refactor(workflow): update JFM research workflow and remove obsolete figures

Revise the JFM research workflow documentation to clarify project structure, decision-making processes, and tool routing. Remove outdated figures and images that are no longer relevant to the current manuscript. Ensure that the workflow aligns with the latest research objectives and safety protocols.

.cursor/rules/jfm-knowledge-terminology.mdc

@@ -0,0 +1,17 @@
+---
+description: JFM manuscript terminology and dangerous scientific boundaries
+globs: "{docs/JFM_WYQ/**/*.md,docs/JFM_WYQ/ROUND5_BOTTOM_UP/**/*.tex,docs/JFM_WYQ/MANUSCRIPT/**/*.tex}"
+alwaysApply: false
+---
+
+# JFM Knowledge and Terminology Scaffold
+
+- Use the current round controls, global decision ledger, owning dossier, named authorities/manifests, and immutable artifacts. Memory and literature do not override them. This rule is a guardrail, not a substitute for exact section authorities.
+- In reader-facing prose name the `parabolic-inflow/no-slip-wall channel configuration` and `uniform-inflow/free-slip-wall channel configuration`. Keep `Legacy` and `V5` only in provenance paths/ledgers; never merge, continue, validate, pool, or imply equivalence across their evidence chains.
+- Spell `Kármán` exactly.
+- Use article-wide $Re_D=U_{ref}D/\nu$: bulk/mean inlet speed for the parabolic configuration, uniform inlet speed for the uniform configuration; state $U_{max}\simeq1.5U_{ref}$ where needed. Convert authoritative $2D$ code labels by one half, verify every inherited label, and prohibit bare `Re100`.
+- Bind every result to configuration, case/seed/role, comparator, metric/estimand, ROI/mask/window, and artifact. Periodic mean/phase, event-relative, mean-only, steady-profile, spatial L2, and native/normalized DTW estimands are non-equivalent; do not pool near-colliding values or call them replication, agreement, validation, or triangulation without exact authority.
+- Positive symbolic regression is Kármán/cloaking-only. Illusion SR is historical/negative.
+- CCD is descriptive and noncausal; use exactly `action-correlated correction-field modes`. Do not imply response, authority, causality, uniqueness, or superiority to POD.
+- OID remains claim-free. Steady evidence is separate-chain context, not mechanism, stability, optimum, or cross-configuration validation. Experiment is qualitative, open-loop, author-owned, and provenance-gated; it is not quantitative validation.
+- Kan99b K2 retains the authority's exact overall `partial pass`. Its opposite mean-lift sign follows the CFD coordinate/force-direction convention; never call it unresolved physics or suspected wrong rotation.

.cursor/rules/jfm-research-workflow.mdc

@@ -0,0 +1,27 @@
+---
+description: General JFM startup, authority, memory, literature, and safety routing
+alwaysApply: true
+---
+
+# JFM Research Start and Tool Routing
+
+## Start from the assigned unit
+- Name the project, active round, exact unit, deliverable, output, and exclusions. R1 indexes evidence; R2 designs; R3 builds dossiers; R4 freezes arguments/displays; R5 writes admitted section prose and controls from frozen R4 packets.
+- Read fresh supplied context once. Follow the coordinator packet's exact reads, ownership, limits, questions, and acceptance checks; do not rediscover the project or recreate R1 coverage.
+
+## Recall, then verify
+- Use 2–3 narrow Nowledge `memory_search` queries containing `DynamisLab JFM` plus round, section/unit, and decision/claim/term. Use `thread_search` and scoped `thread_fetch_messages` only when exact prior wording or conversation history is needed.
+- Memory navigates decisions; `mem_fs` may locate source/original text only when actually stored, with source state retained. It is not a general full-text corpus.
+- R1–R4 indexes, dossiers, plans, controls, and figure plans route architecture, ownership, ceilings, and fallbacks; they do not establish a new internal fact or verify/promote an artifact.
+- Bind internal facts/numbers to current package authority, manifest, code/configuration, or immutable artifact. If unbound, omit, use the frozen fallback, mark an authorized placeholder, or report the gate. Current authority overrides memory and summaries.
+- Search before `memory_add`; update rather than duplicate. Save only durable decisions, procedures, or lessons.
+
+## Literature and tool routing
+- For Undermind: run `get_orientation`, list/select the DynamisLab workspace, inspect existing folders/files/searches, search the library first, and broaden only if needed. Confirm identity, DOI, and availability with `get_paper_info`.
+- Keep each `read_pdfs` request bounded and self-contained: identify the papers/pages or question, request only necessary passages, and preserve `full-PDF`, `full-report`, `mapped`, `snippet`, or `inaccessible` state. Mapped/snippet evidence cannot support detailed wording or structural claims.
+- Literature supports external propositions and rhetorical/structural diagnosis only; it never verifies internal results, numbers, artifacts, priority, readiness, or claim ceilings.
+- Before any authorized read-only script, discover and use the repository-designated environment and documented entry point; do not guess an environment or install/change dependencies.
+
+## Safety and closeout
+- Do not run CFD, training, artifact generation, or expensive/new analysis without explicit authorization. Never fill an evidence gate by inference.
+- Close out whether Working Memory, durable memory, thread recall, `mem_fs`, scripts, and literature were used, distinguishing `read`, `mapped`, and independently verified/audited evidence.

.cursor/rules/jfm-round3-section-drafting.mdc

@@ -0,0 +1,13 @@
+---
+description: Round-3 section dossier workflow; superseded for Round 4/5
+globs: docs/JFM_WYQ/ROUND3_BOTTOM_UP/**/*.md
+alwaysApply: false
+---
+
+# Round-3 Section Dossiers
+- This rule applies only to Round 3. Round 4/5 follow `jfm-research-workflow.mdc` and their current coordinator controls; do not carry Round-3 dossier density into section design or prose drafting.
+- Read the R2 freeze/style guide, relevant R1 dossiers and current authorities. Deliver an author-readable paragraph map plus a compact claim/evidence ledger; record `read`, `mapped`, `not-covered` and `external-inaccessible` without treating reading as verification.
+- Include the section question and bounded answer, paragraph jobs, transitions, display/equation roles, imports/exports, caveats, conflicts, non-claims and blockers. Preserve chain, role, comparator, metric, ROI/window and artifact boundaries.
+- Authority-led artifact deferral is acceptable unless the missing artifact is needed to resolve a conflict, bind an exact role/comparator/window, or substantiate a proposed number or panel.
+- First-pass agents cannot self-declare ready. The coordinator classifies systematic versus module-specific defects, repairs controls first when systematic, requests targeted revision, and performs a second audit.
+- Near-colliding results must be classified as `same estimand`, `related diagnostic` or `different estimand`; prohibit arithmetic pooling and replication/validation/agreement/triangulation wording without exact authority.

.cursor/rules/jfm-round5-current-stage.mdc

@@ -0,0 +1,22 @@
+---
+description: Current Round-5 admission status, outputs, and hard boundaries
+globs: "{docs/JFM_WYQ/ROUND4_TOP_DOWN/**/*.md,docs/JFM_WYQ/FIGURE_PLANS/**/*.md,docs/JFM_WYQ/ROUND5_BOTTOM_UP/**/*.tex,docs/JFM_WYQ/ROUND5_BOTTOM_UP/**/*.md,docs/JFM_WYQ/MANUSCRIPT/**/*.tex,docs/JFM_WYQ/MANUSCRIPT/**/*.md}"
+alwaysApply: false
+---
+
+# Current JFM Stage: Round 5
+
+## Goal and admission
+- R5 produces near-final section-by-section TeX prose plus Markdown audit/control sidecars for later final main-TeX assembly. Do not write directly into the submission master, redesign the paper, reopen frozen R4 decisions, create evidence, or silently fill gates.
+- Start with R5 controls. Record confirmed workflow decisions in repository ledgers/packets as assigned; confirmation does not close source, provenance, caption, artwork, or production-readiness gates.
+- Sections 1, 2, 4–7, and 10 are admitted under their listed local gates.
+- Section 3 is admitted with the author-accepted R4 safe default: a two-figure interface comprising the steady profile and a Kármán-dominant wake composition with one source-bound Vortex and one 0.75-radius Illusion preview. Exact source/provenance/caption and production readiness remain gated.
+- Section 8 is admitted with the author-accepted R4 recommendation: the multi-cylinder Hodge/period outer-solution main equation beside the one-circle incompatibility, plus an integrated CFD, prescribed-circulation outer-reference, and sign/closure-failure figure composition. Exact artwork/source/provenance and production readiness remain gated.
+- Section 9 permits clearly marked placeholder TeX with author fields only. Do not invent apparatus facts, execution details, trial IDs, provenance, or morphology observations; placeholders are not admitted experimental results.
+
+## Scope boundaries
+- Work one assigned section at a time and normally one paragraph job at a time. Change only named outputs; do not edit neighboring sections or interfaces.
+- Draft from the owning current/admitted R4 packet and named current authorities. Preserve jobs, ownership, imports/exports, claim ceilings, displays, equations, fallbacks, and author-only slots.
+- Missing internal authority blocks the dependent factual sentence or number, not unrelated licensed drafting: omit it, mark an explicit placeholder, or use the frozen fallback.
+- Missing artifact readiness allows textual argument only where R4 licenses it; never claim that the artifact exists, is approved, or is production-ready.
+- No agent may promote a section, gate, display, equation, source, artifact, or author-only choice beyond the confirmed workflow decisions and controlling authorities.

.cursor/rules/jfm-round5-production-workflow.mdc

@@ -0,0 +1,27 @@
+---
+description: Controlled lifecycle for Round-5 section TeX and audit production
+globs: "{docs/JFM_WYQ/ROUND4_TOP_DOWN/**/*.md,docs/JFM_WYQ/FIGURE_PLANS/**/*.md,docs/JFM_WYQ/ROUND5_BOTTOM_UP/**/*.tex,docs/JFM_WYQ/ROUND5_BOTTOM_UP/**/*.md,docs/JFM_WYQ/MANUSCRIPT/**/*.tex,docs/JFM_WYQ/MANUSCRIPT/**/*.md}"
+alwaysApply: false
+---
+
+# JFM Round-5 Production Workflow
+
+## Controls and unit contract
+- Initialize R5 controls before prose and record confirmed decisions in the assigned ledger/packet without representing unresolved artifact gates as closed.
+- The coordinator contract names the section, owning packet, authority/evidence reads, paragraph jobs, claim ceiling, output paths, provisional length/share budget, exclusions, local gates, and acceptance checks.
+- Each paragraph requires an evidence card binding its job, claim, comparator, metric/estimand where relevant, authority, caveat, display/equation dependency, and fallback or placeholder.
+- Maintain a live section/article length-and-share ledger; derive provisional budgets from R4 jobs, argument centrality, display burden, and the authorized article target. Historical percentages are context, not quotas.
+
+## Section production
+1. Draft the assigned section as near-final `.tex` prose from paragraph evidence cards; create its audit/control `.md` sidecar and update the live ledger.
+2. After the complete first draft, use bounded full-PDF Undermind evidence for section-level diagnosis of logic, proportion, comparator placement, caveats, and JFM rhetoric.
+3. Perform one integrated rewrite; synthesize cross-paper patterns without imitating one paper. Literature cannot authorize internal content or raise claim ceilings.
+4. Audit contract compliance, source state, paragraph logic, TeX ownership, unresolved gates/fallbacks, word count, and section share in the sidecar.
+5. At each production wave boundary, audit interfaces, ownership, bridges, terminology, displays/equations, duplication, and budget before starting the next wave.
+6. After all eligible sections, run three global passes: structure/logic; evidence/data; length/display. Then run terminology, style, and TeX preflight before later main-TeX assembly.
+
+## Stop conditions and deliverable
+- Bind internal facts/numbers to current package authority, manifest, code/configuration, or immutable artifact. Missing binding blocks only dependent sentences/numbers: omit, use the frozen fallback, or insert an explicit authorized placeholder.
+- Missing artifact readiness permits only R4-licensed textual argument and prohibits claims that a panel, equation artwork, caption, source package, or approval exists.
+- Do not run CFD, training, artifact generation, or new/expensive analysis. Stop only the affected unit when admission, claim license, author choice, or required fallback is absent; report the gate and continue independent licensed work.
+- Deliver the section `.tex`, audit `.md`, updated controls/ledger assigned by the contract, unresolved gates and fallbacks, and knowledge-use closeout. Do not assemble or format the final submission master.

.cursor/rules/jfm-writing-style.mdc

@@ -0,0 +1,24 @@
+---
+description: Operational JFM Round-5 paragraph and TeX drafting protocol
+globs: "{docs/JFM_WYQ/ROUND5_BOTTOM_UP/**/*.tex,docs/JFM_WYQ/ROUND5_BOTTOM_UP/**/*.md,docs/JFM_WYQ/MANUSCRIPT/**/*.tex,docs/JFM_WYQ/MANUSCRIPT/**/*.md}"
+alwaysApply: false
+---
+
+# JFM Round-5 Paragraph Drafting Protocol
+
+## Authority and paragraph contract
+- Draft from the owning current R4 section packet and paragraph evidence card. Obey jobs, ownership, imports/exports, claim ceiling, terminology, display/equation plan, author slots, gates, and fallbacks; current authorities control scientific facts.
+- Literature supports external propositions and rhetorical diagnosis only; it never verifies a DynamisLab result, number, artifact, priority, or readiness. Retain source state and do not use mapped/snippet evidence for detailed wording or structural claims.
+- Order paragraphs as `claim/question → comparator/definition → evidence → bounded interpretation → local caveat → consequence/bridge`. Put the dangerous limitation beside its claim and keep distinct configurations, acquisitions, estimands, masks, windows, and evidence chains separate.
+- Use green wording directly, amber only with the named test/object and adjacent qualifier, and delete red wording unless a dedicated authority licenses it. Avoid causal, mechanism, necessary, sufficient, stable, optimal, robust, general, transferable, validating, or independently confirming claims unless explicitly earned.
+
+## Displays, equations, and TeX
+- Keep prose-ready and artifact-ready separate. Use only licensed displays and captions; otherwise use the recorded textual fallback or clearly marked placeholder. Never imply a planned artifact exists or is approved.
+- Display equations only when owned, source-bound, licensed, and used immediately or repeatedly. Define symbols, normalization, units, sign, and domain at first use; import owner notation rather than re-deriving or duplicating it.
+- Clearly mark every placeholder and name the author-supplied field needed. Do not fabricate `\ref`, `\label`, `\cite`, or bibliography keys; labels, citations, and equations must be owned and source-bound.
+- Produce section TeX only. Do not perform final-master formatting, template integration, submission metadata, or global preamble work.
+
+## Style and self-audit
+- Use concrete physical subjects and verbs. Remove throat-clearing, generic AI promotion, package chronology, inventories, panel narration, repeated templates, decorative lists, unsupported intensifiers, method catalogues, roadmap endings, and heading-repeating conclusions.
+- Keep a number only when it changes the inference. Let captions decode conditions and metadata while body prose carries comparison, inference, limitation, and consequence. Define an acronym once and do not rename fixed terms for variety.
+- Before acceptance verify one paragraph job; authorized claim and exact evidence; comparator before inference; local caveat; valid citation source state; display/equation ownership; no chain pooling or duplication; intended bridge; and no filled author-only slot, invented key, or promoted artifact.

.cursorignore

@@ -0,0 +1 @@
+**/archive/**

.gitignore

@@ -117,7 +117,7 @@ outputs/
 .LSOverride
 
 ref/
-docs/
+# docs/
 # Historical OID authority documents are compact preservation evidence.
 !src/OID_analysis/archive/studies/*/presentation/docs/
 !src/OID_analysis/archive/studies/*/presentation/docs/**
@@ -126,3 +126,29 @@ ParaView/
 src/drl_pinball/legacy_test/output/
 src/drl_pinball/reproduce/output/
 *.tar.gz
+# DRL Train release boundary
+# Historical/failed runs remain local; active best-policy bundles are allowlisted.
+src/drl_pinball/train/archive/
+!src/drl_pinball/train/output/
+src/drl_pinball/train/output/*
+!src/drl_pinball/train/output/*/
+src/drl_pinball/train/output/*/*
+!src/drl_pinball/train/output/*/meta.json
+!src/drl_pinball/train/output/*/calibration.json
+!src/drl_pinball/train/output/*/target.npy
+!src/drl_pinball/train/output/*/best_vecnormalize.pkl
+!src/drl_pinball/train/output/*/models/
+src/drl_pinball/train/output/*/models/*
+!src/drl_pinball/train/output/*/models/best_model.zip
+!src/drl_pinball/train/calibrations/*/target.npy
+# CCD generated evidence/archive payloads remain external immutable data.
+src/CCD_analysis/evidence/*
+src/CCD_analysis/archive/
+src/archive/
+
+# Local/runtime residue and generated publication working trees
+.runtime/
+{output_dir}/
+src/drl_pinball/data/reproduction_flow_plots/
+src/drl_pinball/train/results/latest/tables/training_iterations.csv
+src/drl_pinball/data/reproduction

LegacyCelerisLab/driver.py

@@ -9,6 +9,7 @@ from typing import List, Tuple, Union, Optional
 from . import utils
 from . import preprocess as preproc
 from . import compiler
+from src.CCD_analysis.acquisition.solver_state import copy_ping_pong_ddf, d2q9_q_over_u0_xy
 
 FLUID = 0b00000001
 SOLID = 0b00000010
@@ -112,6 +113,10 @@ class FlowField:
         self.action = np.zeros(0, dtype=self.DATA_TYPE)
         self.obs = np.zeros(0, dtype=self.DATA_TYPE)
         self._control_interval = None
+        self._last_completed_observation = None
+        self._last_effective_action = None
+        self._completed_lattice_steps = 0
+        self._completed_control_intervals = 0
 
         initflow(
             self.flag_gpu,
@@ -126,6 +131,23 @@ class FlowField:
         cuda.memcpy_dtoh(self.flag, self.flag_gpu)
         cuda.memcpy_dtoh(self.ddf, self.ddf_gpu)
 
+    def completed_flags_xy(self) -> np.ndarray:
+        """Return a read-only copy of configured flags in canonical ``(NX, NY)`` order.
+
+        The solver stores flags flat with ``k = x + y * NX``. No bit is
+        interpreted or rewritten here, so FLUID/SOLID and auxiliary bits are
+        preserved exactly.
+        """
+        flat = np.asarray(self.flag)
+        expected = int(self.FIELD_SHAPE[0]) * int(self.FIELD_SHAPE[1]) * int(self.FIELD_SHAPE[2])
+        if flat.dtype != np.dtype("uint8") or flat.ndim != 1 or flat.size != expected:
+            raise RuntimeError("configured solver flag storage is not canonical uint8 flat data")
+        if int(self.FIELD_SHAPE[2]) != 1:
+            raise RuntimeError("canonical CCD flag export supports completed D2 geometry only")
+        result = np.ascontiguousarray(flat.reshape((self.FIELD_SHAPE[1], self.FIELD_SHAPE[0])).T)
+        result.setflags(write=False)
+        return result
+
     def add_cylinder(self, center: Tuple[float, float, float], radius: float, id_obj: Optional[int] = None):
         x_c, y_c, z_c = center
 
@@ -224,6 +246,7 @@ class FlowField:
         self.objects[id_object] = {
             "type": "sensor",
             "center": center,
+            "radius": radius,
         }
 
         self.action = np.zeros(len(self.objects), dtype=self.DATA_TYPE)
@@ -409,6 +432,133 @@ class FlowField:
             cuda.memset_d32_async(self.obs_gpu, 0, self.obs.size, stream)
         stream.synchronize()
         state["completed_steps"] += num_steps
+        self._completed_lattice_steps += num_steps
+
+    def current_step_observation(self):
+        """Return raw telemetry for the latest completed lattice step.
+
+        During an active split interval this is the latest synchronized step. At
+        a completed control boundary it is the persisted final raw step, not the
+        interval-averaged public ``obs``.
+        """
+        state = self._control_interval
+        if state is not None and state["completed_steps"] >= 1:
+            return state["obs_steps"][state["completed_steps"] - 1].copy()
+        if self._last_completed_observation is None:
+            raise RuntimeError("no completed lattice step is available")
+        return self._last_completed_observation.copy()
+
+    def current_effective_action(self):
+        """Return the latest EMA action, including at a completed boundary."""
+        state = self._control_interval
+        if state is not None and state["completed_steps"] >= 1:
+            return np.asarray(state["action"]).copy()
+        if self._last_effective_action is None:
+            raise RuntimeError("no completed lattice step is available")
+        return self._last_effective_action.copy()
+
+    def _require_completed_split_step(self):
+        state = self._control_interval
+        if state is None or state["completed_steps"] < 1:
+            raise RuntimeError("no completed step is available in the active control interval")
+
+    def current_step_velocity_field(self):
+        """Return completed Legacy nondimensional velocity ``q/U0`` as ``(NX, NY)``."""
+        if self._control_interval is not None:
+            self._require_completed_split_step()
+        elif self._last_completed_observation is None:
+            raise RuntimeError("no completed Legacy step is available")
+        # run_control_segment synchronizes before returning. After its pointer swap,
+        # ddf_gpu is the completed state and temp_gpu is the previous/work buffer.
+        cuda.memcpy_dtoh(self.ddf, self.ddf_gpu)
+        flags = self.completed_flags_xy()
+        return d2q9_q_over_u0_xy(
+            self.ddf, int(self.FIELD_SHAPE[0]), int(self.FIELD_SHAPE[1]), flags,
+            float(self.field_config.velocity),
+        )
+
+    def current_step_velocity_probe(self, lattice_index: Tuple[int, int]):
+        """Read one synchronized Legacy nondimensional ``(ux/U0, uy/U0)`` pair."""
+        self._require_completed_split_step()
+        if (not isinstance(lattice_index, tuple) or len(lattice_index) != 2
+                or any(type(value) is not int for value in lattice_index)):
+            raise ValueError("lattice_index must be an (x, y) integer tuple")
+        x, y = lattice_index
+        if not (0 <= x < self.FIELD_SHAPE[0] and 0 <= y < self.FIELD_SHAPE[1]):
+            raise ValueError("velocity probe is outside the lattice")
+        ux, uy = self.current_step_velocity_field()
+        return np.asarray([ux[x, y], uy[x, y]], dtype=self.DATA_TYPE)
+
+    def active_step_clock_state(self):
+        """Return solver lineage during a split interval after a completed step.
+
+        The control clock is the number of fully completed control intervals; it
+        therefore identifies the active interval's zero-based absolute index.
+        """
+        state = self._control_interval
+        if state is None or state["completed_steps"] < 1:
+            raise RuntimeError("active-step clocks require a split interval with a completed step")
+        return {
+            "solver_absolute_lattice_clock": int(self._completed_lattice_steps),
+            "solver_absolute_control_clock": int(self._completed_control_intervals),
+        }
+
+    def solver_clock_state(self):
+        """Return public absolute solver lineage clocks at the current boundary."""
+        if self._control_interval is not None:
+            raise RuntimeError("solver clocks are boundary-safe only")
+        return {
+            "solver_absolute_lattice_clock": int(self._completed_lattice_steps),
+            "solver_absolute_control_clock": int(self._completed_control_intervals),
+        }
+
+    def full_state_checkpoint(self):
+        """Capture exact restart state only at a completed control boundary."""
+        if self._control_interval is not None:
+            raise RuntimeError("full checkpoint requires a completed control boundary")
+        ddf = self.current_step_ddf_checkpoint()
+        return {
+            **ddf,
+            "action": self.action.copy(),
+            "last_effective_action": None if self._last_effective_action is None else self._last_effective_action.copy(),
+            "raw_observation": None if self._last_completed_observation is None else self._last_completed_observation.copy(),
+            "boundary_observation": self.obs.copy(),
+            "solver_absolute_lattice_clock": int(self._completed_lattice_steps),
+            "solver_absolute_control_clock": int(self._completed_control_intervals),
+        }
+
+    def restore_full_state(self, checkpoint):
+        """Restore both ping-pong DDFs and solver-side boundary lifecycle state."""
+        if self._control_interval is not None:
+            raise RuntimeError("cannot restore during an active control interval")
+        current = np.asarray(checkpoint["current_ddf"]); temp = np.asarray(checkpoint["temp_ddf"])
+        action = np.asarray(checkpoint["action"]); boundary = np.asarray(checkpoint["boundary_observation"])
+        raw = checkpoint["raw_observation"]; effective = checkpoint["last_effective_action"]
+        if current.dtype != self.DATA_TYPE or temp.dtype != self.DATA_TYPE or current.shape != self.ddf.shape or temp.shape != self.ddf.shape:
+            raise ValueError("checkpoint DDF storage mismatch")
+        if action.dtype != self.DATA_TYPE or action.shape != self.action.shape or boundary.dtype != self.DATA_TYPE or boundary.shape != self.obs.shape:
+            raise ValueError("checkpoint action/observation mismatch")
+        if raw is not None and (np.asarray(raw).dtype != self.DATA_TYPE or np.asarray(raw).shape != self.obs.shape): raise ValueError("checkpoint raw observation mismatch")
+        if effective is not None and (np.asarray(effective).dtype != self.DATA_TYPE or np.asarray(effective).shape != self.action.shape): raise ValueError("checkpoint effective action mismatch")
+        cuda.memcpy_htod(self.ddf_gpu, current); cuda.memcpy_htod(self.temp_gpu, temp)
+        self.ddf = current.copy(); self.action = action.copy(); self.obs = boundary.copy()
+        self._last_completed_observation = None if raw is None else np.asarray(raw).copy()
+        self._last_effective_action = None if effective is None else np.asarray(effective).copy()
+        self._completed_lattice_steps = int(checkpoint["solver_absolute_lattice_clock"])
+        self._completed_control_intervals = int(checkpoint["solver_absolute_control_clock"])
+        cuda.memcpy_htod(self.action_gpu, self.action)
+
+    def current_step_ddf_checkpoint(self):
+        """Return copies/hashes of current(completed) and temp(previous/work) buffers.
+
+        Synchronous device-to-host copies make this safe both at a completed split
+        step and at a completed control boundary; no solver state is modified.
+        """
+        return copy_ping_pong_ddf(
+            lambda host: cuda.memcpy_dtoh(host, self.ddf_gpu),
+            lambda host: cuda.memcpy_dtoh(host, self.temp_gpu),
+            int(self.FIELD_SIZE * self.LATTICE),
+        )
 
     def end_control_interval(self):
         """Publish obs once, only at the original policy-control boundary."""
@@ -423,6 +573,13 @@ class FlowField:
         self.obs = (self.obs / state["total_steps"]).astype(self.DATA_TYPE)
         cuda.memcpy_dtoh(self.error_flag, self.error_flag_gpu)
         self.last_error_flag = int(self.error_flag[0])
+        # Persist the final lattice-step state before clearing the split lifecycle.
+        # The next begin_control_interval therefore starts its EMA from this exact
+        # action, preserving the historical uninterrupted-run semantics.
+        self.action = np.asarray(state["action"], dtype=self.DATA_TYPE).copy()
+        self._last_effective_action = self.action.copy()
+        self._last_completed_observation = state["obs_steps"][-1].copy()
+        self._completed_control_intervals += 1
         self._control_interval = None
         return self.obs
 

README.md

@@ -2,277 +2,149 @@
 
 **Machine Learning meets Numerical Simulation**
 
-DynamisLab is a research framework for combining machine learning techniques with numerical simulations. Built on top of [CelerisLab](https://github.com/frank14f/CelerisLab), it provides standardized environments and training pipelines for various ML + CFD/Physics projects.
+DynamisLab 是一个面向机器学习与数值模拟交叉研究的框架，当前主线是二维 fluidic pinball（三圆柱）主动流动控制：以 GPU 加速 Lattice Boltzmann Method（LBM）求解流场，用 PPO 等策略驱动圆柱旋转，并用 Symbolic Regression（SR）、Corrected Cross-Covariance Decomposition（CCD）、Observable Identification（OID）和稳态理论分析审查控制规律与证据边界。
 
-## Current Projects
+项目关注的科学问题是：有限的力/速度观测和三个旋转执行器，能否使下游流动接近无扰动基准（Kármán cloaking/stealth）或指定目标流动（Illusion），以及可复现证据究竟支持到“性能、控制结构、统计共变、稳态关联或机制”的哪一层。仓库不是一个已经证明普适机理的成品库；它同时保存可执行代码、当前受限结论、失败结果和历史路线。
 
-### 🎯 FlowStealth
+## 当前状态（2026-08-09）
 
-Deep Reinforcement Learning for Active Flow Control, focusing on:
-- **Flow Stealth**: Drag reduction and flow signature minimization
-- **Flow Illusion**: Manipulating flow patterns for deception and control
-- **Methods**: DRL (PPO) + CFD (Lattice Boltzmann Method)
-- **Location**: `src/flow_stealth/`
+- **活跃（active）**：现代 CelerisLab V5 的 `drl_pinball` 训练/评估工作流与保留的 compact policy bundles；Legacy Kármán SR 的受限结论；corrected Kármán CCD 的描述性 phase-coherent co-variation 结果；稳态 pinball 的 NS-first 后续验证路线。
+- **受限（bounded）**：稳态端点、有限时域瞬态和 article publication 已形成可审计证据，但机制、因果性及渐近/全局稳定性未建立；现代 V5 发布可复现训练接口和部分保留模型，不表示所有 case/seed 已按当前代码完整重训并验证；Legacy 标准化 SR 图包仍是 partial Kármán package。
+- **历史/负结果（historical/negative）**：Illusion SR 已失败并关闭为科学解释路线；OID 的 2026-07 双场景研究降级为条件性历史材料；早期 reproduce、CCD/OID 分支和被替代实现均保留在各级 `archive/`。
 
-## Features
+必须区分两个不同命题：**Illusion PPO 是仓库仍支持的训练与确定性评估能力**，包括现代 `train_illusion.py`、目标及 harmonics 合约；**Illusion SR 则不是有效科学结果**。长窗口 Legacy 标准化采集显示保留的 SR 动作衰减到接近 physical zero，未显示相对 zero 的有意义收益，因此不得把旧的 200/400-step target similarity 写成 Illusion SR efficacy、target tracking 或跨尺寸泛化。
 
-- 🌊 **CFD Environments**: Gymnasium-compatible environments for flow control
-- 🤖 **RL Integration**: Ready-to-use with Stable-Baselines3 and other RL libraries
-- 🚀 **GPU Acceleration**: Leverages CelerisLab's CUDA-accelerated LBM solver
-- 📊 **Experiment Tracking**: Built-in TensorBoard integration
-- 🔧 **Modular Design**: Organized by research projects
-- 📦 **Standard Structure**: Follows Python packaging best practices
+Kármán CCD 只支持 DRL 相对 constant-control phase template 的描述性共变及稳定 rank-3 subspace；它不是配对反事实、因果效应、响应时间、机制、explained variance，也不证明 CCD 优于 POD。稳态结果同样尚未闭合 Navier–Stokes 机制链；potential-circulation cancellation 已失败，有限时域关联不得升级为因果性或全局稳定性。OID 当前是 claim-free CPU method core，不承载活跃科学结论。
 
-## Project Structure
+## 五分钟阅读顺序
 
-```
-DynamisLab/
-├── src/                            # Source code (organized by project)
-│   └── flow_stealth/               # FlowStealth: DRL + CFD Active Control
-│       ├── __init__.py
-│       ├── config.py               # Configuration management
-│       └── environments/           # Gymnasium environments
-│           ├── __init__.py
-│           └── cfd_env.py          # CFD flow control environment
-├── scripts/                        # Training and evaluation scripts
-│   └── train_ppo.py                # PPO training script for FlowStealth
-├── configs/                        # Configuration files
-│   ├── config_cuda.json            # CUDA settings
-│   ├── config_flowfield.json       # Flow field parameters
-│   └── config_gym.json             # Environment settings
-├── CelerisLab/                     # CelerisLab submodule (GPU-accelerated CFD)
-├── models/                         # Trained model checkpoints (gitignored)
-├── output/                         # Training data and results (gitignored)
-├── tensorboard/                    # TensorBoard logs (gitignored)
-├── docs/                           # Documentation
-├── requirements.txt                # Python dependencies
-├── pyproject.toml                  # Package configuration
-└── README.md                       # This file
-```
+先看 [`src/README.md`](src/README.md) 获取源码包总索引，再按项目进入权威文档。
 
-## Installation
+1. 本页：项目范围、状态和证据规则。
+2. [`src/drl_pinball/train/README.md`](src/drl_pinball/train/README.md)：现代 V5 校准、训练与保留产物。
+3. [`src/drl_pinball/eval/README.md`](src/drl_pinball/eval/README.md)：canonical V5 确定性评估及 fail-closed 合约。
+4. [`src/SR_analysis/README.md`](src/SR_analysis/README.md)：Legacy Kármán SR 主张、Illusion SR 负结果和不可变证据链。
+5. [`src/CCD_analysis/README.md`](src/CCD_analysis/README.md)：corrected Kármán CCD 当前唯一结果权威。
+6. [`src/steady_pinball_theory/README.md`](src/steady_pinball_theory/README.md) 与 [`src/OID_analysis/README.md`](src/OID_analysis/README.md)：稳态受限路线与 OID scientific reset。
 
-### Prerequisites
+继续复现实验前，再读对应 package 的 pipeline、contract、results、handoff 和 archive index；不要只依据旧图、文件名或历史 README 作结论。
 
-- Python 3.8+
-- NVIDIA GPU with CUDA support
-- CUDA Toolkit 11.0+
+## 包与目录地图
 
-### Step 1: Clone the repository
+- [`src/drl_pinball/`](src/drl_pinball/) — 主运行包。`train/` 是现代 2000×600 CelerisLab V5 校准/训练；`eval/` 是 V5 评估；`legacy_test/` 在 LegacyCelerisLab 上执行统一的历史角色采集；`legacy_env/`、`legacy_train/` 和局部 `archive/` 服务于兼容与溯源。状态：**active + bounded + historical**。
+- [`src/SR_analysis/`](src/SR_analysis/) — 在冻结 PPO 轨迹上发现、拟合并于 Legacy CFD 闭环验证符号控制器。当前只保留 Kármán/cloaking 为 **active but bounded**；Illusion SR 是 **historical/negative**，steady 仅作幅值校准背景。
+- [`src/CCD_analysis/`](src/CCD_analysis/) — corrected `karman_re100` ROI mean 与 cycle-template CCD 的精简权威入口。状态：**completed descriptive/bounded**；旧 temporal/phase 与 pre-reset 树为 **superseded historical**。
+- [`src/OID_analysis/`](src/OID_analysis/) — weighted snapshot POD、Schlegel LR/LE maps、配对和原子 artifact 校验。状态：**active method core, no active science claim**；2026-07 两场景研究为条件性历史材料。
+- [`src/steady_pinball_theory/`](src/steady_pinball_theory/) — rear-symmetric steady rotation 的分析、finite-Re 数值证据与 NS-first 验证。状态：**active investigation, bounded evidence**；尚无完成的稳态机制证明。
+- [`src/pv_plot/`](src/pv_plot/) — ParaView/FFmpeg 轻量 PNG/MP4 渲染接口，不是独立服务或 pip 项目。状态：**active utility**。
+- [`src/archive/`](src/archive/) — 被替代的现代 solver reproduction Track B 等仓库级历史材料；各 package 自己的 `archive/` 也属于其 provenance chain。状态：**historical, non-authoritative unless an active document explicitly cites it**。
+- [`CelerisLab/`](CelerisLab/) — Git submodule，现代 GPU LBM solver。Legacy solver 及其证据依赖按各 package 文档和本地环境解析；不要把两个 solver 的结果静默合并。
+
+## 两条 solver 证据链
+
+**LegacyCelerisLab** 是 SR article chain、标准化 Legacy acquisition 以及 corrected CCD 输入所依赖的历史 plant。其冻结模型、normalization、动作顺序、EMA、边界条件、采样时钟和 immutable manifests 共同定义证据；可从 [`src/drl_pinball/legacy_test/README.md`](src/drl_pinball/legacy_test/README.md) 进入。
+
+**现代 CelerisLab V5** 是当前 `drl_pinball/train` 与 `drl_pinball/eval` 的 2000×600 工作流，使用新的 Simulation API、calibration/target bundle 和 registry。仓库保留了可推理的 compact bundles 与评估基线，但这不等于当前实现已经完成全矩阵重训，也不允许用 V5 replay 自动替代 Legacy article evidence。
+
+两条链的网格、边界实现、动作换算、force/sensor normalization、policy bundle 和指标版本必须分别报告。跨 solver 比较只能作为明确标注的 migration/reproduction 研究，不能混池统计或互相补证。
+
+## 证据权威层级
+
+由高到低采用以下顺序；若 package README 指定了更严格的内部顺序，以它为准：
+
+1. **不可变原始/派生 artifact**：成功、失败或 partial run 的 manifest、SHA-256、parent identity、原始 telemetry/field 和完整 metric contract。
+2. **可执行代码与测试**：能够重算或 fail closed 的 loader、analysis、runner 和 contract tests。
+3. **当前 package 权威文档**：active README、RESULTS/claim ledger、METHOD/PIPELINE、checkpoint；后写且明确 supersede 的文档优先。
+4. **仓库根 README**：负责导航和跨包边界，不替代 package 数值权威。
+5. **历史文档、计划、旧图和 archive narrative**：只说明当时意图或演进，不证明 run 完成，也不覆盖后续负结果。
+6. **Working Memory/对话摘要**：仅作检索线索，不能替代仓库 artifact。
+
+任何数字若不能解析到 scene、solver、role、realization、window、metric definition、artifact path/hash 和 evidence status，就不应写成权威结论。失败、拒绝、未完成和 partial 记录也是证据，不得覆盖或“清理”掉。
+
+## 环境与安装
+
+基础包声明 Python `>=3.8`，但当前 GPU 工作流以 Linux、NVIDIA GPU、CUDA 和 Python 3.10 的 `pycuda_3_10` conda 环境为准。分析任务通常在 CPU-only `pinball_math` 环境运行；具体依赖和命令以相应 package README 为准。ParaView 自带 Python 不能混入 conda Python 的 `PYTHONPATH`。
 
 ```bash
-git clone --recurse-submodules <your-repo-url> DynamisLab
+git clone --recurse-submodules https://github.com/frank14f/DynamisLab.git
 cd DynamisLab
-```
+git submodule update --init --recursive
 
-> **Note**: If CelerisLab is a submodule, use `--recurse-submodules` to clone it automatically.
-
-### Step 2: Install CelerisLab
-
-#### Option A: Install from submodule (recommended for development)
-
-```bash
-cd CelerisLab
-pip install -e .
-cd ..
-```
-
-#### Option B: Install from pip (if published)
-
-```bash
-pip install CelerisLab
-```
-
-### Step 3: Install DynamisLab dependencies
-
-```bash
-pip install -r requirements.txt
-```
-
-### Step 4: Install DynamisLab in development mode
-
-```bash
+conda create -n pycuda_3_10 python=3.10 -y
+conda activate pycuda_3_10
+pip install pycuda
+pip install -e CelerisLab
 pip install -e .
 ```
 
-## Quick Start
+`requirements.txt`/`pyproject.toml` 提供 NumPy、SciPy、PyTorch、Gymnasium、Stable-Baselines3、绘图和开发工具的基础声明。GPU CFD 应串行初始化；尤其 Legacy 编译会改写共享 kernel/config 文件，不要因使用不同 GPU 就并发编译不同 case。
 
-### Training a PPO Agent
+## 真实 quickstart
 
-Train a Proximal Policy Optimization agent for flow control:
+现代 V5 Kármán target calibration 与单次训练（完整参数见 [`train/README`](src/drl_pinball/train/README.md)）：
 
 ```bash
-python scripts/train_ppo.py \
-    --run-name my_first_run \
-    --device-id 0 \
-    --total-timesteps 100 \
-    --n-steps 3600 \
-    --activation sin
+cd src/drl_pinball/train
+conda run --no-capture-output -n pycuda_3_10 python -u calibrate.py \
+  --case kar_re100 --device-id 0 --si 800 \
+  --config ../../../configs/config_lbm_karman_2000x600.json \
+  --out-dir calibrations/kar_re100
+
+conda run --no-capture-output -n pycuda_3_10 python -u train_karman.py \
+  --case-name kar_re100 --device-id 0 --seed 42 \
+  --config ../../../configs/config_lbm_karman_2000x600.json \
+  --calibration calibrations/kar_re100/calibration.json \
+  --total-episodes 500 --symmetry-prob 0
 ```
 
-**Arguments:**
-- `--run-name`: Name for this training run (used for saving models and logs)
-- `--device-id`: CUDA device ID for CFD simulation
-- `--cuda-device`: CUDA device ID for PyTorch training (can be different from --device-id)
-- `--total-timesteps`: Number of training iterations
-- `--n-steps`: Environment steps per training iteration
-- `--activation`: Activation function (`sin`, `tanh`, or `relu`)
-
-### Monitoring Training
+Illusion PPO 使用同一 calibration contract、`train_illusion.py` 和 `ill_*` case；这表示能力存在，不表示 Illusion SR 成功。V5 retained policy 评估：
 
 ```bash
-tensorboard --logdir tensorboard/
+bash src/drl_pinball/eval/run_all.sh --case kar_re100 --seed 45
+python3 src/drl_pinball/eval/infer_train.py --validate-all --output-root /tmp/v5-eval
 ```
 
-Then open http://localhost:6006 in your browser.
-
-### Using the Environment Programmatically
-
-```python
-from flow_stealth.environments import CFDFlowControlEnv
-from flow_stealth.config import load_celeris_configs
-
-# Load configurations
-config_cuda, config_field = load_celeris_configs()
-
-# Create environment
-env = CFDFlowControlEnv(
-    device_id=0,
-    config_cuda=config_cuda,
-    config_field=config_field,
-)
-
-# Run episode
-obs, info = env.reset()
-for step in range(500):
-    action = env.action_space.sample()  # Random action
-    obs, reward, terminated, truncated, info = env.step(action)
-    
-    if terminated or truncated:
-        break
-
-env.close()
-```
-
-## Configuration
-
-### CFD Configuration
-
-Edit `configs/config_flowfield.json` to change flow parameters:
-
-```json
-{
-  "viscosity": 0.01,          # Fluid viscosity
-  "velocity": 0.1,            # Inlet velocity
-  "field_dim_in_U": [400, 200, 1],  # Grid dimensions
-  ...
-}
-```
-
-### CUDA Configuration
-
-Edit `configs/config_cuda.json` for GPU settings:
-
-```json
-{
-  "threads_per_block": 256,
-  "unit_dimensions": [16, 16, 1],
-  ...
-}
-```
-
-## Advanced Usage
-
-### Resume Training
+CPU-only 稳态分析/测试示例（更多命令及 bounded 条件见 [`steady_pinball_theory/README`](src/steady_pinball_theory/README.md)）：
 
 ```bash
-python scripts/train_ppo.py \
-    --resume models/my_run_best.zip \
-    --run-name my_run_continued
+PYTHONPATH="$PWD/src" conda run -n pinball_math \
+  python -m pytest -q src/steady_pinball_theory/tests/test_core.py
 ```
 
-### Custom Hyperparameters
+绘图接口见 [`src/pv_plot/README.md`](src/pv_plot/README.md)；Legacy SR acquisition 和 CCD 不适合作为首次 GPU quickstart，应先读其 immutable-input、serial execution 与 provenance 合约。
 
-```bash
-python scripts/train_ppo.py \
-    --learning-rate 0.0003 \
-    --gamma 0.99 \
-    --batch-size 512 \
-    --n-steps 7200
-```
+## 2026 年 2–8 月演进
 
-### Multi-GPU Setup
+- **2 月**：仓库建立，形成 “Machine Learning + numerical simulation” 的广义框架和 fluidic pinball 主问题。
+- **6 月**：集中整理早期 SR/SINDy、CCD correction-field 与 OID/POD/LR-LE 分析；同时暴露 alignment、FIFO、ROI 和证据组织问题，开始从“有图/有分数”转向 contract-first。
+- **7 月上旬**：建立参数化现代 V5 Kármán/Illusion 训练、跨 Re/尺寸路线和 solver reproduction；修正 inlet/action/order 等兼容问题。
+- **7 月中下旬**：冻结 Legacy SR Stage 1→2→3 article chain，加入 dual-clock CCD、OID 和轻量 `pv_plot`；闭环、hash、no-clobber 与失败保留成为基本要求。
+- **8 月初**：OID scientific reset 撤回 importance/causality 读法；现代 V5 training release 冻结 compact bundles；steady theory 转向 NS-first；CCD 重做为 corrected ROI mean/cycle-template 权威。
+- **8 月 8–9 日**：统一 Legacy 长窗口角色采集并完成项目收尾。Kármán SR 保持受限活跃；Illusion SR 因接近 physical-zero 的长窗口结果关闭；steady article publication 保持 BOUNDED；根文档改为状态与证据入口。
 
-```bash
-# CFD simulation on GPU 0, PyTorch training on GPU 1
-python scripts/train_ppo.py \
-    --device-id 0 \
-    --cuda-device 1
-```
+## Archive 与 provenance 规则
 
-## Environment Details
+- run/artifact 一经发布即 immutable、no-clobber；新尝试使用新 ID，不覆盖成功、失败或 partial 目录。
+- 移动/重命名前先核对 parent path、hash、manifest、symlink、外部 Optane mapping 和下游消费者。可读性不能破坏 identity。
+- Git 中没有 bulk payload 不代表证据不存在；外部 archive 必须由 inventory/hash 和 package relocation 文档绑定。仅有本地文件也不等于已经 Git-preserved。
+- `archive/` 默认不执行、不导入、不作为当前结论；只有 active authority 明确引用时，才按其限定用途使用。
+- derived plot/report 不自动成为 acceptance evidence；必须回溯其完整输入和生成合约。
+- Legacy 与 V5 的 artifact、normalization、target、metric 和 solver version 分库存放、分开引用。
 
-### CFDFlowControlEnv
+## 科学写作边界
 
-The main environment for active flow control around a cylinder.
+可以写：指定 solver、case、窗口和 comparator 下的性能；Kármán SR 在已测试 deletion window 内的 term ranking；corrected CCD 的 phase-coherent descriptive co-variation；稳态端点/有限时域的审计结果；OID 数学工具的实现与验证。
 
-**Observation Space:**
-- Dimensionality: `n_sensors × 2 × 2` (velocity components, current + derivative)
-- Default: 12 dimensions (3 sensors × 2 velocities × 2)
-- Normalized to zero mean and unit variance
+不能写：普适 Reynolds/尺寸规律、全局最优或唯一 SR 方程、PPO 本身具有等变性、CCD/DTW lag 给出因果机制或物理延迟、稳态旋转已经闭合动量/能量机制、单 realization 给出不确定性、现代全矩阵已完整重训、OID predictive R² 等价于 importance/control authority，或 Illusion SR 已实现 tracking/generalization。
 
-**Action Space:**
-- Dimensionality: `n_control_cylinders`
-- Default: 3 (three controllable cylinders)
-- Range: [-1, 1] (scaled internally to physical velocities)
+写作时应并列给出 observation、interpretation 和 mechanism 的层级，并明确 solver、场景、基线、时长、样本数和状态。论文草稿、计划和 acknowledgements 不得反向升级本地证据。
 
-**Reward:**
-- Drag reduction: `-cd × 0.1`
-- Lift minimization: `-|cl| × 0.05`
-- Flow similarity: `-similarity_distance × 0.5`
-- Total reward is sum of components
+## Citation、License 与致谢
 
-**Episode:**
-- Max steps: 500 (configurable)
-- Simulation runs at 800 LBM steps per environment step
-
-## Development
-
-### Project Guidelines
-
-- Follow PEP 8 style guide
-- Use type hints for function signatures
-- Document classes and functions with docstrings
-- Organize projects under `src/` (e.g., `src/flow_stealth/`)
-- Keep training scripts in `scripts/`
-- Use `config.py` for all path and configuration management
-
-### Adding a New Project
-
-1. Create new project directory in `src/` (e.g., `src/my_new_project/`)
-2. Add `__init__.py`, `config.py`, and project-specific modules
-3. Create environments in `src/my_new_project/environments/`
-4. Create corresponding training scripts in `scripts/`
-
-### Adding a New Environment to FlowStealth
-
-1. Create new environment class in `src/flow_stealth/environments/`
-2. Inherit from `gym.Env`
-3. Register in `src/flow_stealth/environments/__init__.py`
-4. Create corresponding training script in `scripts/`
-
-### Running Tests
-
-```bash
-pytest tests/
-```
-
-## Citation
-
-If you use DynamisLab in your research, please cite:
+若使用本仓库，请引用实际采用的代码版本/commit、对应 artifact/manifest 和底层 solver；建议的软件引用为：
 
 ```bibtex
-@software{dynamis2026,
+@software{dynamislab2026,
   author = {Frank14f},
   title = {DynamisLab: Machine Learning for Computational Fluid Dynamics},
   year = {2026},
@@ -280,31 +152,6 @@ If you use DynamisLab in your research, please cite:
 }
 ```
 
-Also cite CelerisLab:
+同时按实际使用情况引用 [CelerisLab](https://github.com/frank14f/CelerisLab)、Stable-Baselines3、Gymnasium、NumPy/SciPy、PyTorch、ParaView 以及各分析方法的原始论文。代码以 [MIT License](LICENSE) 发布；外部论文、模型、数据和 solver artifact 可能有各自的引用与再分发要求。
 
-```bibtex
-@software{celerislab2026,
-  author = {Frank14f},
-  title = {CelerisLab: GPU-Accelerated Lattice Boltzmann Method Solver},
-  year = {2026},
-  url = {https://github.com/frank14f/CelerisLab}
-}
-```
-
-## License
-
-MIT License - see LICENSE file for details
-
-## Acknowledgments
-
-- Built on [CelerisLab](https://github.com/frank14f/CelerisLab) CFD solver
-- Uses [Stable-Baselines3](https://github.com/DLR-RM/stable-baselines3) for RL
-- Gymnasium API for standardized environments
-
-## Contributing
-
-Contributions are welcome! Please open an issue or pull request.
-
-## Contact
-
-For questions or issues, please open a GitHub issue or contact Frank14f.
+感谢 CelerisLab 提供 GPU-accelerated LBM 基础，Stable-Baselines3 与 Gymnasium 提供强化学习接口，PyTorch、NumPy/SciPy、Matplotlib 和 ParaView 支撑训练、分析与可视化。欢迎通过 issue 或 pull request 讨论可复现性、证据合约和明确限定范围的改进。

docs/JFM_WYQ/CELERISLAB_VALIDATION_DOSSIER.md

@@ -0,0 +1,510 @@
+# CelerisLab Validation Dossier for JFM Writing
+
+**Purpose.** This document is an agent-facing index of the CelerisLab solver, its validation contracts, and the qualification artifacts currently retained under `CelerisLab/tests/output/steady_jfm_qualification/`. It is intentionally more detailed than manuscript prose. It records what was validated, how it was measured, what the reference papers establish, and what must not be claimed.
+
+**Scope.** CelerisLab is the current CFD tool library developed for the DynamisLab programme. `LegacyCelerisLab` is the preceding solver generation and is a separate evidence chain. This dossier concerns current CelerisLab only unless explicitly stated otherwise.
+
+**Status date.** 2026-08-09.
+
+**Important evidence warning.** The qualification directory contains complete physics runs for only Kan99b K2 and Sah04 S2. The validation runners define larger K1--K5 and S1--S4 matrices, but those other cases are not present in this qualification directory. Smoke runs are wiring/development diagnostics and are not physics qualification.
+
+---
+
+## 1. Fast agent index
+
+| Question | Read this first |
+|---|---|
+| What is CelerisLab? | `CelerisLab/README.md` |
+| What is the current solver revision? | parent gitlink `c918ac0...`; submodule commit `6e3756c...` |
+| What exact outputs were retained? | `CelerisLab/tests/output/steady_jfm_qualification/MANIFEST.json` |
+| What is the rotating-cylinder contract? | `CelerisLab/tests/specs/Kan99b_validation.md` and `tests/validation/run_kan99b_rotating_cylinder.py` |
+| What is the confined-cylinder contract? | `CelerisLab/tests/specs/Sah04_validation.md` and `tests/validation/run_sah04_st_matrix.py` |
+| What passed in the retained artifacts? | Sections 7 and 8 below |
+| What failed or remains incomplete? | Section 10 below |
+| How are forces and sensors read? | `CelerisLab/README.md`, Sections “Obs telemetry model” and “Sensor special handling” |
+| What can be written in a JFM methods section? | Section 11 below |
+| What should not be written? | Section 12 below |
+
+Search terms for later agents: `CelerisLab validation`, `steady_jfm_qualification`, `Kan99b K2`, `Sah04 S2`, `MRT`, `D2Q9`, `Bouzidi`, `curved boundary`, `neq_extrap`, `channel_stabilized`, `St_error_pct`, `mean_Cd`, `sensor accuracy`, `FP32`, `EsoPull`, `FP16S`.
+
+---
+
+## 2. Software identity and solver role
+
+DynamisLab is the larger research programme for ML-assisted CFD and active flow control of a two-dimensional fluidic pinball. CelerisLab was written as the modern CFD infrastructure for that programme. It provides:
+
+- GPU-accelerated D2Q9 and D3Q19 LBM;
+- SRT, TRT and MRT collision operators;
+- optional Smagorinsky LES;
+- double-buffer and EsoPull streaming paths;
+- moving curved boundaries using Bouzidi interpolation;
+- runtime rotating-body actuation;
+- force, torque and area-averaged velocity-sensor readback;
+- Python control-loop and asynchronous CUDA-stream APIs;
+- checkpoint/snapshot and runtime body-topology synchronization.
+
+The qualification experiments are not a proof that every solver feature is equally validated. They are targeted anchors for the combinations needed by the current research and paper setting.
+
+### Solver revision provenance
+
+- DynamisLab parent revision that points to the solver: `c918ac0de44d4f5106249c31ad7f8a10f4c53450`.
+- CelerisLab submodule revision: `6e3756c587ec08ab6a55c4a1a5f4671822d3f0a0`.
+- Submodule commit message: `fix(esopull): correct init layout and pre-streaming semantics (v0.5.1)`.
+- Parent commit message: `Update CelerisLab submodule to v0.5.1 (6e3756c)`.
+
+For a reproducible paper, cite the solver revision actually used by the run and preserve the artifact manifest. Do not cite only the repository name while omitting the solver version and configuration.
+
+---
+
+## 3. Governing numerical model and core implementation
+
+The solver advances a low-Mach incompressible approximation using a D2Q9 lattice in the retained two-dimensional qualification cases. The collision path used by the retained Kan99b and Sah04 full runs is MRT. The main stepping path is GPU CUDA code orchestrated through Python/PyCUDA.
+
+The qualification configurations use:
+
+- lattice model: D2Q9;
+- collision: MRT;
+- storage: FP32;
+- LES: disabled;
+- streaming: double buffer;
+- curved-body treatment: current Bouzidi moving-wall implementation;
+- outlet: `neq_extrap`;
+- rotating-cylinder validation inlet: uniform profile with `regularized` primary inlet and one `zou_he_local` sensitivity run;
+- confined-cylinder validation inlet: parabolic profile with `channel_stabilized` inlet;
+- open-domain lateral boundaries: free slip for Kan99b;
+- confined-channel lateral boundaries: no slip, represented by the channel geometry/configuration for Sah04.
+
+### Configuration semantics that matter for writing
+
+`inlet.profile` and `inlet.scheme` are different concepts:
+
+- `profile` defines the physical target profile, uniform or parabolic;
+- `scheme` defines the numerical inlet closure, e.g. `regularized`, `zou_he_local` or `channel_stabilized`.
+
+The body rotation input is a runtime angular velocity. Changing `omega` does not require recompilation because it does not change object topology or telemetry layout. Geometry changes do require topology rebuilding/reinitialization.
+
+The solver-generated kernel configuration headers under `src/CelerisLab/lbm/kernels/config/*.h` are generated artifacts. They are not hand-edited scientific inputs; the JSON config and compiler generation path are authoritative.
+
+---
+
+## 4. Telemetry, force and sensor contracts
+
+The GPU body kernels accumulate telemetry in the `obs_gpu` buffer. The buffer has force, torque and sensor segments. The standard `run(steps)` path clears the observation buffer and resets the internal accumulated-step count by default, advances the requested number of LBM steps, and downloads telemetry to a pinned host buffer.
+
+For force and torque:
+
+- `read_force(id, normalize=True)` returns the accumulated force divided by the number of accumulated time steps;
+- `read_force(id, normalize=False)` returns the raw accumulated sum;
+- the validation runners explicitly use `normalize=False` after a one-step accumulation in their fine-grained loop, so each recorded force is a one-step quantity.
+
+For sensors:
+
+- sensor readings are always area-normalized by the number of cells in the sensor footprint;
+- `normalize=True` additionally divides by the accumulated time count;
+- the sensor area normalization is independent of the time normalization flag.
+
+This distinction is essential when comparing solver readback with a manually computed field average or when reconstructing force coefficients from CSV output.
+
+The dedicated sensor test places four circular sensors at `(120,50)`, `(120,64)`, `(120,78)` and `(150,64)` on a `256 x 128` D2Q9 MRT free-slip uniform-flow setup. It compares each GPU sensor value with a CPU arithmetic mean over exactly the sensor cell footprint and uses an absolute tolerance of `1e-4` in each velocity component. The test is an implementation/readback validation, not a fluid-physics validation.
+
+---
+
+## 5. Validation architecture
+
+The CelerisLab test hierarchy has three levels:
+
+1. `tests/unit/`: CPU-only isolated logic such as pending edits, flag masks and equilibrium helpers.
+2. `tests/integration/`: GPU tests for body-topology synchronization, DDF patching, unified action/observation handling and stream APIs.
+3. `tests/validation/`: long GPU physics-regression runners against published rotating-cylinder and confined-cylinder references, plus sensor accuracy and performance diagnostics.
+
+The two physics references serve different purposes:
+
+- [Kan99] is an open-flow rotating-cylinder benchmark. It checks the moving curved boundary, rotation input, force integration, shedding frequency and rotation-dependent mean/fluctuating force coefficients.
+- [Sah04] is a confined stationary-cylinder benchmark. It checks parabolic channel inflow, no-slip wall confinement, blockage mapping, developed velocity normalization, wake shedding frequency and high-blockage geometry handling.
+
+These are validation anchors for solver behaviour. They are not validation of the full fluidic-pinball DRL controller, the symbolic-regression pipeline, or any hydrodynamic-cloaking claim.
+
+---
+
+## 6. Kan99b rotating-cylinder contract
+
+### 6.1 Reference problem
+
+[Kang, Choi and Lee, “Laminar flow past a rotating circular cylinder”] studies fully developed two-dimensional flow around a circular cylinder rotating steadily in a uniform viscous stream. The paper uses
+
+\[
+Re = U_\infty D/\nu,
+\qquad
+\alpha = \frac{\omega D}{2U_\infty},
+\]
+
+and reports Strouhal number, mean lift and drag, and lift/drag fluctuation amplitudes. It also studies suppression of vortex shedding as the rotation parameter increases.
+
+The strongest exact numeric anchor selected in the repository is the paper's `Re=100`, `alpha=1.0` convergence case:
+
+| Quantity | Kan99 reference |
+|---|---:|
+| `St` | 0.1655 |
+| mean `C_L` | -2.4881 |
+| mean `C_D` | 1.1040 |
+| `C'_L` | 0.3631 |
+| `C'_D` | 0.0993 |
+
+The paper also supports a low-rotation lift trend near `Re=100`, approximately `mean C_L ≈ -2.48 alpha`, and qualitative suppression thresholds near `alpha=1.4` at `Re=60`, `alpha=1.8` at `Re=100`, and `alpha=1.9` at `Re=160`. The repository correctly treats those threshold values as regime guides rather than exact single-point gates.
+
+### 6.2 Lattice mapping
+
+The runner fixes:
+
+\[
+U_\infty=0.03,\qquad D=30,\qquad R=15,
+\]
+
+so
+
+\[
+\nu=\frac{U_\infty D}{Re}=\frac{0.9}{Re},
+\qquad
+\omega_{body}=\frac{2\alpha U_\infty}{D}=0.002\alpha.
+\]
+
+For the retained K2 run:
+
+- `Re=100`;
+- `alpha=1.0`;
+- `nu=0.009`;
+- body omega `=0.002`;
+- medium domain `M = 1351 x 601`;
+- cylinder diameter `D=30` lattice cells;
+- total steps `280000` (`80000` burn-in + `200000` measured);
+- force recorded every 100 steps;
+- 2800 recorded samples per run.
+
+The runner uses the exact force definitions
+
+\[
+C_D=\frac{2F_x}{U_\infty^2D},
+\qquad
+C_L=\frac{2F_y}{U_\infty^2D}.
+\]
+
+The fluctuation amplitude is half the peak-to-peak range over the post-burn time series, using cycle-aware crossing logic when sufficient crossings are present.
+
+### 6.3 Primary and sensitivity variants
+
+The retained full package has two K2 runs:
+
+1. **Primary baseline:** MRT + uniform profile + `regularized` inlet.
+2. **Inlet sensitivity:** MRT + uniform profile + `zou_he_local` inlet.
+
+Both use double-buffer streaming, FP32 storage, no LES, free-slip y boundaries and `neq_extrap` outlet. The second run is not a separate physical reference case; it tests sensitivity to the inlet numerical closure.
+
+### 6.4 Retained full results
+
+| Variant | St | St error | mean CL | mean CD | C'L | C'D |
+|---|---:|---:|---:|---:|---:|---:|
+| MRT + regularized | 0.1696483 | 2.5065% | +2.5881365 | 1.1430751 | 0.3529538 | 0.0984419 |
+| MRT + zou_he_local | 0.1685349 | 1.8338% | +2.5764851 | 1.1294693 | 0.3488342 | 0.0952994 |
+| Kan99 reference | 0.1655 | — | -2.4881 | 1.1040 | 0.3631 | 0.0993 |
+
+The frequency, drag and fluctuation magnitudes are close to the reference. The mean lift has the opposite sign relative to the stored Kan99 anchor in both variants, with a relative magnitude error of approximately 204%. The manifest therefore classifies the Kan99b full result as a **partial pass**, not an aggregate pass: `St`, `C_D` and fluctuation amplitudes pass their declared bands, but the mean-`C_L` sign/convention mismatch fails the all-metric gate.
+
+This is scientifically important. Do not write “all Kan99b metrics agree” or call the sign mismatch a minor rounding issue. Before using the rotating-cylinder lift in a mechanism argument, the object orientation, wall-point velocity convention, positive rotation direction, force sign convention and reference-paper coordinate convention must be audited.
+
+### 6.5 Declared K2 bands
+
+The spec declares the following preferred relative bands:
+
+- `St`: within 3%;
+- mean `C_L`: within 4%;
+- mean `C_D`: within 5%;
+- `C'_L`: within 8%;
+- `C'_D`: within 10%.
+
+For the two retained runs, the band results are:
+
+| Variant | St | mean CL | mean CD | C'L | C'D | aggregate |
+|---|---|---|---|---|---|---|
+| regularized | pass | fail | pass | pass | pass | partial pass |
+| zou_he_local | pass | fail | pass | pass | pass | partial pass |
+
+### 6.6 Matrix specified but not retained here
+
+The runner/spec defines:
+
+- K1: `Re=100`, `alpha=0.5`, lift-trend check;
+- K2: `Re=100`, `alpha=1.0`, hard anchor;
+- K3: `Re=60`, `alpha=1.6`, suppression classification;
+- K4: `Re=100`, `alpha=2.0`, suppression classification;
+- K5: `Re=160`, `alpha=2.0`, suppression classification;
+- optional K0: `Re=100`, `alpha=0`.
+
+Those cases are not present in `steady_jfm_qualification`; only K2 baseline and K2 inlet sensitivity are present. The current dossier therefore treats K1 and K3--K5 as **specified/planned validation cases**, not completed qualification evidence.
+
+---
+
+## 7. Sah04 confined-cylinder contract
+
+### 7.1 Reference problem
+
+[Sahin and Owens, “A numerical investigation of wall effects up to high blockage ratios on two-dimensional flow past a confined circular cylinder”] studies a stationary circular cylinder midway between parallel no-slip walls. The paper defines
+
+\[
+\beta=D/H,
+\qquad
+Re=U_{max}D/\nu,
+\qquad
+St=\frac{D}{TU_{max}}.
+\]
+
+The paper covers blockage ratios up to approximately 0.9 and Reynolds numbers up to approximately 280. It documents Hopf shedding, pitchfork symmetry breaking, asymmetric states and high-blockage wall/wake interactions.
+
+The selected direct periodic anchors are:
+
+| Case | beta | Re | target St |
+|---|---:|---:|---:|
+| S1 | 0.3 | 100 | 0.2115 |
+| S2 | 0.5 | 200 | 0.3513 |
+| S3 | 0.8 | 160 | approximately 0.5537 |
+| S4 | 0.9 | 200 | 0.5314 |
+
+The repository deliberately avoids using values interpolated from figures or critical-onset points as hard numerical gates.
+
+### 7.2 Retained S2 setup
+
+The full retained run uses:
+
+- case S2;
+- nominal `beta=0.5`;
+- nominal `Re=200`;
+- physical target profile: parabolic;
+- numerical inlet scheme: `channel_stabilized`;
+- collision: MRT;
+- outlet: `neq_extrap`;
+- no-slip confined walls;
+- `D=30` lattice cells;
+- fluid height `H=60` cells;
+- realized `beta=30/60=0.5`;
+- wall gap: 15 cells;
+- grid: `2402 x 62` including boundary rows;
+- `120000` total steps;
+- `45000` burn-in steps;
+- forces sampled every 5 steps;
+- 24000 stored lift/drag samples.
+
+The analysis measures the developed downstream `U_max,real`, and reports
+
+\[
+Re_{real}=\frac{U_{max,real}D}{\nu}.
+\]
+
+The spectrum is computed from the post-burn lift signal using a mean-subtracted Hanning-windowed real FFT, a target-guided frequency band, and local log-parabolic sub-bin interpolation.
+
+### 7.3 Retained full S2 result
+
+| Quantity | Result |
+|---|---:|
+| nominal Re | 200 |
+| realized Re | 204.8809 |
+| nominal beta | 0.5 |
+| realized beta | 0.5 |
+| target St | 0.3513 |
+| measured St | 0.3566902 |
+| relative St error | 1.5344% |
+| gate | pass, threshold 5% |
+| mean Cd | 2.5801072 |
+| realized Umax | 0.10244045 |
+| curved links | 292 |
+| fallback links | 0 |
+| low-q links | 72 |
+| final rho range | 0.9820464 to 1.0512210 |
+
+The full run passes the declared S2 Strouhal gate. It is a solver/reference anchor at a specific confined-channel setup. It is not evidence that all high-blockage cases, all collision models or all bifurcation boundaries have been validated.
+
+The saved NPZ contains:
+
+- 24000 lift samples;
+- 24000 drag samples;
+- sample lattice steps;
+- burn index and recording interval;
+- post-burn frequencies and power spectrum;
+- final `rho`, `ux`, `uy` fields;
+- measured `St`, peak frequency, realized `Re`, and realized `beta`.
+
+### 7.4 Smoke S2 result and interpretation
+
+The smoke run uses only `5000` steps with `1500` burn-in and retains 1000 force samples. It gives:
+
+- realized `Re=161.4173` rather than the nominal 200;
+- measured `St=0.3039154`;
+- relative error `13.4884%`;
+- gate fail.
+
+This is expected diagnostic behaviour: the short run has not developed the target state and has insufficient statistical duration. It must be labelled wiring/initialization diagnostic only and must never be quoted as a failed physical validation of CelerisLab.
+
+### 7.5 Matrix specified but not retained here
+
+The runner/spec defines S1--S4, with high-blockage refinement recommended for S3/S4. The retained qualification directory contains only S2 full and S2 smoke. S1, S3 and S4 are not completed artifacts in this directory.
+
+The specification requires realized blockage and realized Reynolds number to be reported, especially at high blockage. It also recommends at least approximately doubled resolution for `beta >= 0.8` before treating S3/S4 results as validation-quality. This is a methodological requirement, not evidence that those refined cases have already been run.
+
+---
+
+## 8. Manifest and artifact inventory
+
+The manifest schema is `steady-pinball-cfd-anchor-qualification/v1`.
+
+### Manifest interpretation
+
+- `kan99b_full`: partial pass; frequency, drag and fluctuation amplitudes pass, but mean lift sign/convention mismatch fails aggregate gate.
+- `sah04_full`: pass; S2 Strouhal error 1.534% is below the 5% gate.
+- `smoke`: diagnostic only; insufficient duration for physics gates.
+
+### Retained artifact hashes
+
+| Relative artifact | SHA-256 |
+|---|---|
+| `kan99b_full/force_csv/k2_baseline_domM_re100_a1p000_regularized_mrt.csv` | `e5ba55a789254322408b6657c175b0c1ac82ce92c3db4216716727ff8cc7df1c` |
+| `kan99b_full/force_csv/k2_k2_inlet_control_domM_re100_a1p000_zou_he_local_mrt.csv` | `84b999757c13c0d975be968f22b7e4cbf9ad9229d9ce91eb096c5ecdc92bbf72` |
+| `kan99b_full/summary_runs.csv` | `b009ffc75615d65ec6d0d351da37a2d99ac4e57378d8ee635207e65e7edc6677` |
+| `kan99b_full_summary.json` | `18b19b15bd426c68d5d919d3449f436a2284386726e2cd19e8334ccba2f0c6bf` |
+| `sah04_full/s2_mrt.npz` | `6e3ef38de8cd1bc5c67ffa693868b1c6fd97453e1d7721a9287ca6fdf741f5ec` |
+| `sah04_full_summary.json` | `e212fb4e0653e9ca3a3715fefef1d552e7dfac8acc1e46cd732769c0d7ce13e1` |
+| `kan99b_smoke/force_csv/k2_baseline_domM_re100_a1p000_regularized_mrt.csv` | `049382058562dbfda0aedb836827311302878d6589d790a94942aa973dee6262` |
+| `kan99b_smoke/force_csv/k2_k2_inlet_control_domM_re100_a1p000_zou_he_local_mrt.csv` | `45552003e8c9f8149d06eaac9f8696691e1e8607d92fd0e57135ea0919a664e4` |
+| `kan99b_smoke/summary_runs.csv` | `468188a9cbf54b4491999f306b0e82352be91880038fd58487cdb040034518a3` |
+| `kan99b_smoke_summary.json` | `24c101fcc7d1568c1ea0c4e8a9fe7c00b9427f9046b183bc9018624b65f76247` |
+| `sah04_smoke/s2_mrt.npz` | `40438246e8b9c63dd1b28c9982a273d48885f6d378ba9756afae3db61b8eafd5` |
+| `sah04_smoke_summary.json` | `902b211d28ca2c4d4205aae12cdd1ab6f563231e31ab23a8bed626d55451871` |
+
+The manifest is the hash authority. The full artifact files are the evidence, not this prose dossier.
+
+---
+
+## 9. Additional CelerisLab validation relevant to setting writing
+
+### 9.1 Sensor accuracy
+
+`tests/validation/test_sensor_accuracy.py` validates the sensor kernel against a direct CPU average over the same sensor footprint. It uses four sensor locations, a free-slip uniform-flow MRT test, and an absolute component-wise difference threshold of `1e-4`. The test establishes that area-averaged sensor readback is consistent with the underlying macroscopic field sampling for the tested setup.
+
+This supports using area-averaged probes in a DRL observation loop. It does not establish full-field observability, sensor optimality, or equivalence between sparse sensor DTW and full-field wake error.
+
+### 9.2 Streaming equivalence
+
+CelerisLab v0.5.1 includes an EsoPull single-buffer streaming path. The README records verification for D2Q9 curved-boundary MRT configurations, including fixed/rotating-cylinder Kan99b K2 comparisons and runtime body synchronization. The current documented scope is two-dimensional D2Q9; D3Q19 EsoPull is not yet implemented in that verification statement.
+
+The README reports approximate Kan99b K2 drag comparisons between EsoPull and double-buffer for `D=20` and `D=30`, with differences below approximately 3.2% and 5.6%, respectively. These are implementation equivalence diagnostics and should not be confused with the retained `steady_jfm_qualification` full-run manifest.
+
+### 9.3 FP16S limitation
+
+FP16S stores the distribution function in half precision with FP32 computation and scaling. The documented validation found that FP16S combined with Bouzidi curved boundaries can produce approximately 30--40% drag error even with DDF shifting, while no shifting can exceed 100% error for Kan99b K2. Sah04 S2 Strouhal can remain within approximately 1.5%, but force-critical curved-boundary work should use FP32.
+
+Therefore the JFM steady/cloaking qualification should use FP32 storage when force or torque is interpreted scientifically. A good Strouhal result under FP16S is not enough to authorize force-mechanism claims.
+
+### 9.4 DDF shifting
+
+DDF shifting stores `f_i-w_i` rather than `f_i` to improve half-precision accuracy. The documented supported scope includes D2Q9 MRT with double-buffer or EsoPull and the tested inlet families. It is important for FP16S but does not cure all Bouzidi/quantization force errors. The retained steady qualification uses ordinary FP32 and `ddf_shifting=false`.
+
+### 9.5 Performance
+
+The CelerisLab README reports approximately 4400 MLUPS for a 384 x 192 D2Q9 V100 benchmark under Re100 MRT no-LES, for both double-buffer and EsoPull at the stated benchmark level. The performance runner is a separate engineering benchmark. MLUPS is not a validation of physical accuracy and should be reported separately from physics metrics.
+
+---
+
+## 10. Known limitations and unresolved issues
+
+1. **Kan99 mean-lift sign mismatch.** This is the most important unresolved physics-validation issue in the retained package. The magnitude is close but the sign is opposite to the Kan99 anchor. Audit coordinate and force conventions before interpreting rotating-cylinder lift.
+2. **Qualification matrix is partial.** Only K2 and S2 full runs are retained in this directory. Do not claim K0--K5 or S1--S4 as completed based only on runner definitions or README tables.
+3. **Smoke runs are not physics evidence.** The S2 smoke run demonstrates that the setup runs but has realized Re drift and insufficient periodic duration.
+4. **No uncertainty estimate.** The retained artifacts are deterministic solver runs, not independent ensembles. They provide no statistical confidence interval.
+5. **No mesh-convergence result in this manifest.** Domain/grid choices are documented in the runners/specs, but the qualification directory does not contain a multi-resolution convergence package.
+6. **No full field snapshots for Kan99 full.** The retained Kan99 full output contains force histories and summaries, not vorticity images or field snapshots.
+7. **Sah04 high-blockage cases remain unqualified here.** High blockage is especially sensitive to narrow wall gaps, realized Re and geometry resolution.
+8. **Reference/solver contract differences matter.** Agreement is expected only after matching Re definition, inlet profile, wall condition, outlet treatment, rotation convention, coefficient normalization and analysis window.
+9. **No mechanism proof.** A benchmark match validates selected numerical observables; it does not prove a particular wake mechanism or guarantee transfer to the three-cylinder pinball.
+10. **No Legacy equivalence.** LegacyCelerisLab results cannot be silently pooled with current CelerisLab results. The two solver generations have different APIs, accumulation semantics, generated kernels, boundary implementation details and provenance.
+
+---
+
+## 11. Writing-ready methods language
+
+The following is suitable as a starting point, subject to checking the final artifact paths and solver revision:
+
+> The flow solver was implemented in CelerisLab, a CUDA-accelerated D2Q9 lattice-Boltzmann code with MRT collision, double-buffer streaming and Bouzidi interpolation for moving curved boundaries. Force and velocity-sensor observables were accumulated on the GPU and read back through the Python API; force coefficients were formed from the accumulated force using the same diameter and free-stream normalization as the reference problem. Before applying the solver to the fluidic-pinball control problem, we qualified selected numerical components against two canonical single-cylinder benchmarks. For a steadily rotating cylinder, the `Re=100`, `alpha=1` case reproduced the reference Strouhal number, mean drag and fluctuation amplitudes within the declared tolerance bands, although the mean-lift sign convention remained discrepant and was therefore not used as an accepted validation observable without further convention auditing. For a stationary cylinder in a confined channel at `beta=0.5`, `Re=200`, the measured Strouhal number was `0.35669` compared with the reference value `0.3513`, a relative error of `1.53%`. These tests qualify the selected solver/configuration combination for the reported numerical setting; they do not constitute a proof of universal solver accuracy or of the later flow-control mechanism.
+
+### What to put in a table
+
+At minimum include:
+
+- solver revision;
+- lattice model and dimension;
+- collision model;
+- streaming path;
+- storage precision;
+- LES state;
+- inlet physical profile and numerical scheme;
+- outlet mode;
+- wall condition;
+- diameter and grid;
+- `Re`, `nu`, `U_inf` or `U_max` definition;
+- body rotation parameter and physical omega;
+- burn-in and statistics window;
+- sample interval;
+- target and measured `St`;
+- measured force metrics where accepted;
+- artifact path and manifest/hash identifier.
+
+---
+
+## 12. Claims that are authorized, bounded or prohibited
+
+### Authorized with scope
+
+- CelerisLab implements a GPU-accelerated D2Q9 MRT LBM suitable for the tested two-dimensional rotating and confined cylinder configurations.
+- The retained Sah04 S2 full run passes the declared 5% Strouhal gate.
+- The retained Kan99b K2 full runs reproduce frequency, drag and fluctuation magnitudes within their selected bands.
+- The selected sensor readback path can be checked against direct area averages over the tested footprints.
+- FP32 is the appropriate storage choice for force-critical curved-boundary calculations in the documented validation scope.
+
+### Bounded / conditional
+
+- The Kan99b K2 result is a partial pass because mean-lift sign/convention agreement is unresolved.
+- The `regularized` versus `zou_he_local` K2 comparison is an inlet sensitivity check, not evidence that one inlet scheme is universally superior.
+- The Sah04 result qualifies one moderate-blockage confined-channel point, not all blockage ratios or bifurcation regimes.
+- EsoPull equivalence is bounded to the documented D2Q9 MRT configurations and comparison metrics.
+- Performance figures describe hardware/configuration throughput, not accuracy.
+
+### Prohibited without new evidence
+
+- “All Kan99b K0--K5 and Sah04 S1--S4 cases passed.”
+- “The solver is validated for all Reynolds numbers, all blockage ratios or all collision models.”
+- “The mean-lift sign mismatch is irrelevant.”
+- “A matching Strouhal number proves force accuracy, boundary accuracy or hydrodynamic mechanism.”
+- “The solver validation proves the DRL policy, SR law, CCD interpretation or cloaking mechanism.”
+- “LegacyCelerisLab and CelerisLab are numerically interchangeable.”
+- “Smoke output is a physical benchmark result.”
+
+---
+
+## 13. Reference papers
+
+- [Kan99] S. Kang, H. Choi and S. Lee, “Laminar flow past a rotating circular cylinder,” *Physics of Fluids* 11, 3312--3321 (1999). DOI: `10.1063/1.870190`.
+- [Sah04] M. Sahin and R. G. Owens, “A numerical investigation of wall effects up to high blockage ratios on two-dimensional flow past a confined circular cylinder,” *Physics of Fluids* 16, 1305--1320 (2004). DOI: `10.1063/1.1668285`.
+
+Undermind cite-key resolution in the DynamisLab workspace: `Kan99` and `Sah04`. In Undermind workspace Markdown, use bare markers `[Kan99]` and `[Sah04]`; in manuscript prose, resolve them through the project's bibliography workflow.
+
+---
+
+## 14. Reproduction entry points
+
+From the `CelerisLab/` directory, the validation runners are:
+
+```bash
+conda run -n pycuda_3_10 python tests/validation/run_kan99b_rotating_cylinder.py
+conda run -n pycuda_3_10 python tests/validation/run_sah04_st_matrix.py
+conda run -n pycuda_3_10 python tests/validation/test_sensor_accuracy.py
+```
+
+The retained qualification artifacts should not be overwritten. New runs must use a new no-clobber output directory and must record the exact command, environment, GPU, solver revision, configuration, artifact hashes and gate interpretation.
+
+**Final operational rule:** read the code and manifest before quoting any number. This dossier is an index and interpretation aid; the JSON/NPZ/CSV artifacts and current validation code remain the numerical authority.

docs/JFM_WYQ/FIGURE_PLANS/README.md

@@ -0,0 +1,49 @@
+# JFM figure-specification index and production controls
+
+This directory is the Round-4 figure-specification layer for Sections 1–10. It organizes display intent, provenance, readiness and safe omission. It does **not** make a panel, table, equation, source payload or result publication-ready. No file here authorizes CFD, training, raw-array inspection, computation, rendering or artifact generation.
+
+## Section index
+
+- **Section 1 — Introduction:** no owned empirical display. No plan file required; reason: the frozen introduction owns framing only.
+- **Section 2 — Problem formulation, control framework and evaluation:** `SECTION_02_FIGURE_01_CONFIGURATIONS_AND_SENSORS.md`; `SECTION_02_FIGURE_02_DRL_INFORMATION_ACTION_REWARD.md`; `SECTION_02_TABLE_01_CONFIGURATION_METHOD.md`; `SECTION_02_TABLE_02_CASE_METRIC_SEMANTICS.md`; `SECTION_02_TABLE_03_CFD_QUALIFICATION.md`; `SECTION_02_EQUATION_SPEC.md`.
+- **Section 3 — Capability progression:** `SECTION_03_FIGURE_01_STEADY_DISCOVERY_PROFILE.md`; `SECTION_03_FIGURE_02_WAKE_CLOAKING_AND_RETARGETING.md`. Section 3 remains `discuss`: Fig. 1 composition and representative Vortex offset are author choices.
+- **Section 4 — Kármán performance:** `SECTION_04_FIGURE_01_KARMAN_CONDITIONS_AND_LEARNING.md`; `SECTION_04_FIGURE_02_REFERENCE_FIELD_SIGNAL_ACTION.md`.
+- **Section 5 — Compact Kármán SR law:** `SECTION_05_KARMAN_SR_CONTROL_LAW.md`.
+- **Section 6 — Kármán mean fields and CCD:** `SECTION_06_FIGURE_SPEC.md`; S6-S2 remains blocked.
+- **Section 7 — Cross-case extensions:** `SECTION_07_FIGURE_01_VORTEX_EVENT_RESPONSE.md`; `SECTION_07_FIGURE_02_ILLUSION_ERASE_EXTENSIONS.md`. Role fields exist; quantitative Vortex and Erase products remain blocked; Illusion is bounded PPO capability.
+- **Section 8 — Steady context and theory boundary:** `SECTION_08_STEADY_CONTEXT_AND_THEORY.md`. Existing outer assets are bounded redraw sources only; no new MFS computation is authorized. Frank's equation/composition choice remains open.
+- **Section 9 — Qualitative open-loop experiment:** `SECTION_09_EXPERIMENT.md`; `author-supply-gated`, with strict omission fallback.
+- **Section 10 — Discussion and conclusions:** no display, table or equation by frozen text-only contract. No plan file required.
+
+Every actual plan file in this directory is indexed above. Existing filenames are intentionally unchanged in this pass.
+
+## Controlled cross-figure rules
+
+1. Stable reader roles are target/reference → controlled → physical zero/uncontrolled; add constant or SR only when acquired and explicitly bound. A target has no action trajectory.
+2. Never pool or visually imply equivalence between configurations. Internal chain labels stay in provenance.
+3. Bind case, seed/realization, role and comparator. Use neutral source-case labels where inherited parabolic velocity-reference conversion is unresolved; never guess a reader-facing Reynolds value.
+4. Mean, phase, event-relative, steady-profile, native DTW, normalized DTW, standardized rolling DTW and fixed-cycle deletion DTW are non-interchangeable. State whether larger or smaller is better.
+5. Record ROI/domain, geometry guard or exact mask, weighting/normalization and retained window/phase/event coordinate. A display crop is not a metric ROI.
+6. Record x/y direction, array order/transposition, action/body order and sign, and color/sign semantics.
+7. Record exact authority, source artifact, script/evaluator, manifest, role rows/windows and hashes in the production manifest. A partial plot manifest authenticates only the diagnostics it lists; it does not prove completeness, portability or readiness.
+8. Share crop, scale, units and mask only within a direct, authority-bound comparison.
+9. Captions decode configuration, roles, condition, coordinate and limits; they do not promote morphology to efficacy, signal matching to field matching, qualification to control validation, or association to causality.
+10. Omit unresolved panels, scalars or annotations. Never fill a gap with a non-equivalent source, inferred geometry, unsupported conversion or diagnostic screenshot.
+
+## Normalized readiness schema
+
+Each plan should use these five independent fields. `N/A` must include a reason.
+
+- **Plan state:** `specified`, `discuss`, `author-supply-gated`, or `omitted (N/A: reason)`.
+- **Evidence state:** `read`, `path-confirmed`, `mapped`, `not-covered`, `missing-exact-leaf`, or `author-supply-gated`. `read` is not independent verification; `mapped` is not read or locally resolved.
+- **Metric state:** `definition-ready`, `authority-bound`, `dependency-gated`, `blocked`, or `N/A: method-only/no quantitative result`.
+- **Render state:** `not-started`, `draft-diagnostic-only`, `redraw-required`, `blocked`, or `N/A: omitted/text-only`.
+- **Disposition:** `retain`, `retain-with-gates`, `supplement`, `fallback-only`, or `omit`.
+
+## Required production record
+
+Before production, record: authority; evidence state; chain/configuration; case/seed/roles/comparator; metric and orientation; ROI/domain/mask/weight/window; display crop; array/action/flow orientation; exact source/script/evaluator/manifest and hashes; render settings/output hash; caption ceiling; fallback. Mark unavailable fields `N/A` with a reason. Missing required fields prevent a publication-ready claim.
+
+## Coverage boundary
+
+`read`: current Round-4 controls/owning section packets and the two Section-2 repository reports. `path-confirmed`: exact repository paths named in plans. `mapped`: authority-described or manifest-linked payloads not opened or independently audited here. `not-covered`: raw arrays, external payload verification, computation, CFD/training, rendering, figure generation and publication typography. Durable memory was recalled for navigation only; no thread recall or literature search was needed.

docs/JFM_WYQ/FIGURE_PLANS/SECTION_02_EQUATION_SPEC.md

@@ -0,0 +1,15 @@
+# Section 2 equation specification
+
+## Required equations
+1. `Re_D=Uref D/nu`; bulk/mean `Uref` for parabolic inflow, uniform speed for uniform inflow, and `Umax ~= 1.5 Uref`. Parabolic inherited labels remain unconverted until source metadata bind `U0`.
+2. `(Uw,Vw)=(-omega r_y, omega r_x)`, positive omega counter-clockwise; `s_i=omega_i R/Uref`. Historical source-native `alpha_i=omega_i/U0`; general `s_i=alpha_i R(U0/Uref)`, reducing to `alpha_i R` only when `U0=Uref` is established.
+3. Online objective: compact force plus sparse-sensor discrepancy; exact weights optional and run-bound.
+4. Mean field L2: area/coordinate-weighted, `Uref`-normalized vector error over declared ROI with roles/comparator explicit.
+5. Phase L2: same spatial contract by periodic slot; slots are not repeated-cycle samples.
+6. DTW recurrence/path or rigorous prose; state distance/similarity, normalization, channels, aggregation, window and constraints.
+7. Comparator-relative reduction optional only under one exact role/ROI/mask/weight/window contract.
+
+## Readiness contract
+Authority is frozen Section 2 plus the sensor/field report. Equations 1–2 are configuration definitions; 3–7 require result-local chain/config/case/seed/role/comparator bindings. ROI/mask/weight/window are `N/A: physical/method definition` for 1–3 and mandatory for 4–7. Action sign/order and metric orientation accompany reuse. Implementation constants and rows are `dependency-gated`; no equation implies an artifact. Caption ceiling is definitions only. Fallback retains Reynolds, action, mean L2 and rigorous prose DTW; recurrence/detail may move to supplement.
+
+Plan `specified`; evidence `read`; metric `definition-ready`; render `N/A: typeset equations, not artifact panels`; disposition `retain-with-gates`.

docs/JFM_WYQ/FIGURE_PLANS/SECTION_02_FIGURE_01_CONFIGURATIONS_AND_SENSORS.md

@@ -0,0 +1,17 @@
+# Section 2 Figure 1 — Flow configurations and downstream sensors
+
+## Job and layout
+A method-only schematic defines, side by side, the parabolic-inflow/no-slip-wall and uniform-inflow/free-slip-wall channel configurations. Show inlet profile, lateral and outlet boundaries, the three-cylinder pinball, body order `(F,U,L)`, x-right/y-up coordinates, rotation sign, the case-dependent upstream disturbance or target body, and three circular two-component sensors at `x/D=10`, `y/D=0,±2`, radius `0.25D`. Disturbance/target geometry remains a labelled variable unless an exact selected-case source supplies dimensions.
+
+## Production contract
+- **Authority:** `docs/JFM_WYQ/ROUND4_TOP_DOWN/SECTION_02_PROBLEM_METHOD_EVALUATION.md`; `docs/JFM_WYQ/WAKE_L2_AND_SENSOR_PLACEMENT_RESEARCH_NOTES.md`.
+- **Chain/configuration:** both physical configurations, kept separate; internal chains are N/A in reader labels.
+- **Case/seed/roles/comparator:** `N/A: method definition, not a performance comparison`.
+- **Metric orientation; ROI/mask/weight/window:** `N/A: schematic defines observations, not efficacy`. A registered field-evaluation ROI may be shown only as a separate evaluation region.
+- **Orientation:** x right, y up; positive solver omega counter-clockwise under `(Uw,Vw)=(-omega*r_y,omega*r_x)`; action order front, rear-y+, rear-y-; flow direction printed.
+- **Exact source/script/manifest:** geometry drawing source and row-level dimensions are `missing-exact-leaf`; no production script/manifest exists. Do not guess cylinder centres, domain lengths, disturbance diameter or target placement.
+- **Caption ceiling:** geometry, boundaries, action order/sign and sparse sensing only; no equivalence, optimal sensing, observability, solver-exact scale or performance.
+- **Fallback:** explicitly not-to-scale minimal line drawing plus Table 2-1; omit unbound dimensions/case body.
+
+## Readiness
+Plan state `specified`; evidence authorities `read`, paths `path-confirmed`, geometry leaves `missing-exact-leaf`, raw/config arrays `not-covered`; metric `N/A: method-only`; render `not-started`; disposition `retain-with-gates`.

docs/JFM_WYQ/FIGURE_PLANS/SECTION_02_FIGURE_02_DRL_INFORMATION_ACTION_REWARD.md

@@ -0,0 +1,17 @@
+# Section 2 Figure 2 — DRL information, action and reward flow
+
+## Job and layout
+Draw one left-to-right closed-loop method schematic: physical configuration/state → body forces plus six velocity channels from three downstream sensors → fixed physical scaling and matched online normalization → PPO policy → three rotations `(F,U,L)` → CFD evolution. A separate target-signal branch enters force- and sensor-discrepancy reward terms. A dashed independent-evaluation branch leads to signal DTW and registered field L2; it does not feed the policy unless exact run authority says so.
+
+## Production contract
+- **Authority:** frozen Section 2 and `docs/JFM_WYQ/WAKE_L2_AND_SENSOR_PLACEMENT_RESEARCH_NOTES.md`.
+- **Chain/config/case/seed/roles/comparator:** canonical method; run-specific chain/case/seed/constants `N/A: no selected performance run`. Roles are target signals, controlled state and physical-zero comparator only in evaluation.
+- **Metric orientation:** reward higher-better only when implementation-bound; DTW must state distance/similarity and native/normalized; spatial L2 lower-better. No value is printed.
+- **ROI/mask/weight/window:** field branch names the registered `20D x 10D` geometry-guarded ROI; exact weights/windows are run-specific. Sensor disks are not the ROI.
+- **Orientation:** body/action order and sign follow Section 2; arrows show information/execution, not causality or convergence.
+- **Exact source/script/manifest:** conceptual item; no drawing script/manifest. Policy/normalizer/target/reward rows are `dependency-gated`.
+- **Caption ceiling:** flow only; no convergence, optimality, observability, field equivalence, mechanism or canonical-rerun claim.
+- **Fallback:** compact prose flow or one-column block diagram without run constants.
+
+## Readiness
+Plan `specified`; evidence authorities `read`, code families `mapped`, selected-run bindings `not-covered`; metric `definition-ready` and numerical binding `dependency-gated`; render `not-started`; disposition `retain-with-gates`.

docs/JFM_WYQ/FIGURE_PLANS/SECTION_02_TABLE_01_CONFIGURATION_METHOD.md

@@ -0,0 +1,14 @@
+# Section 2 Table 1 — Configuration and numerical-method definitions
+
+## Job and columns
+Compress definitions without implementation history. Candidate columns: configuration; inlet/lateral/outlet boundaries; pinball/disturbance or target geometry; `Uref` and article `Re_D`; solver/collision/curved-body treatment; grid/diameter/time sampling; evidence ownership; caveat.
+
+## Contract
+- **Authority:** frozen Section 2; `docs/JFM_WYQ/CELERISLAB_VALIDATION_DOSSIER.md` for qualification context.
+- **Rows:** parabolic/no-slip and uniform/free-slip configurations only. Exact grid, diameter, time-step, precision and selected-run rows are `dependency-gated`; never infer from a different run or benchmark.
+- **Metric/ROI/window:** `N/A: definition table, not a result`; record coordinate/action orientation in notes.
+- **Exact source/script/manifest:** row-level config/manifest leaves are `missing-exact-leaf`; bind each numerical row before production. The dossier cannot substitute for control-run metadata.
+- **Caption ceiling:** separate plants and methods only; no transfer, canonical-rerun completion or performance.
+- **Fallback:** retain source-bound boundaries, nondimensionalization, D2Q9/MRT and Bouzidi rows; move run matrix to supplement.
+
+Plan `specified`; evidence Section 2/dossier `read`, run authorities `mapped`; metric `N/A: method-only`; render `not-started`; disposition `retain-with-gates`.

docs/JFM_WYQ/FIGURE_PLANS/SECTION_02_TABLE_02_CASE_METRIC_SEMANTICS.md

@@ -0,0 +1,14 @@
+# Section 2 Table 2 — Case and metric semantics
+
+## Job and rows
+Define non-equivalent families: periodic Kármán; Vortex event-relative; periodic Illusion; Erase mean-only candidate; steady endpoint/profile. Columns: roles; temporal coordinate/window; field estimand; signal estimand/orientation; ROI/mask/weight; allowed inference; prohibited pooling.
+
+## Contract
+- **Authority:** frozen Section 2 and `docs/JFM_WYQ/WAKE_L2_AND_SENSOR_PLACEMENT_RESEARCH_NOTES.md`.
+- **Metric orientation:** spatial error lower-better; DTW orientation printed per output; native and normalized DTW separate. Eight slots are independently phased nearest snapshots, not an ensemble. Vortex offsets are not phase. Erase efficacy is blocked. Steady uses endpoint/profile criteria.
+- **ROI/mask/weight/window:** registered ROI `-6 <= (x-xs)/D <= 14`, `|(y-y0)/D| <= 5`, geometry-guarded because exact masks were not persisted. Changed estimands renew their own contract.
+- **Bindings:** definitions only; exact chain/config/case/seed/roles/comparator and source/evaluator/manifest remain downstream result-owner dependencies.
+- **Caption ceiling:** definitions only; no pooled efficacy, validation or transfer.
+- **Fallback:** concise prose list preserving all non-equivalences.
+
+Plan `specified`; evidence `read`; metric `definition-ready`, quantitative Vortex/Erase `blocked`; render `not-started`; disposition `retain-with-gates`.

docs/JFM_WYQ/FIGURE_PLANS/SECTION_02_TABLE_03_CFD_QUALIFICATION.md

@@ -0,0 +1,15 @@
+# Section 2 Table 3 — Compact CFD qualification
+
+## Job and rows
+Qualify selected CelerisLab observables. Columns: benchmark/configuration; retained case; observable; authority; resolution/window; status; caveat; artifact/manifest binding.
+
+## Contract
+- **Authority:** `docs/JFM_WYQ/CELERISLAB_VALIDATION_DOSSIER.md` (`read/full-report`).
+- **Permitted rows:** Kan99b K2, aggregate status exactly `partial pass`; Sah04 S2, `pass` for its confined-channel Strouhal gate. K2 frequency, drag and fluctuation observables pass their bands; the opposite mean-lift sign follows the CFD coordinate/force-direction convention and is not unresolved physics or suspected wrong rotation.
+- **Excluded rows:** K1/K3–K5, S1/S3/S4 and smoke runs are not completed retained physics evidence.
+- **Metric/ROI:** row-specific benchmark gates; exact values/resolution/window require artifact/manifest row binding. ROI `N/A: benchmark observables are not the manuscript wake ROI`.
+- **Roles:** current CelerisLab qualification only; PPO/SR/CCD/control roles `N/A`.
+- **Caption ceiling:** selected solver-observable qualification only; no controller, interpretation, older-solver or cloaking validation.
+- **Fallback:** bounded paragraph naming K2 partial pass and S2 pass; matrix to supplement.
+
+Plan `specified`; evidence dossier `read`, artifact leaves `mapped`; metric `dependency-gated`; render `not-started`; disposition `retain-with-gates`.

docs/JFM_WYQ/FIGURE_PLANS/SECTION_03_FIGURE_01_STEADY_DISCOVERY_PROFILE.md

@@ -0,0 +1,110 @@
+# Section 3 Figure 1 — Steady discovery through downstream velocity profiles
+
+## Scientific question
+Did the historically first approximately steady rear-cylinder actuation recover the uniform target velocity profile downstream relative to a stationary pinball?
+
+## Intended comparison
+Show the actual cross-stream downstream velocity profiles against target, controlled and stationary roles. The profile is the evidence; streamlines are retained only if they add a distinct topology view. A scalar station summary cannot substitute for the profiles.
+
+## Source plots, data and scripts
+
+- Profile data: `src/steady_pinball_theory/figures/profiles_long.csv`.
+  - Columns used: `case`, `x_over_D`, `y_over_D`, `ux_over_U`, `uy_over_U`, `fluid_mask`, `source_sha256`.
+  - Selected cases: `cfd-accepted-s3p55` and `cfd-stationary`.
+  - Mandatory station: `x_over_D=10.0`; optional secondary station: `6.0` only if useful at final size.
+  - Uniform target: $u/U_\infty=1$, $v/U_\infty=0$ on the same valid/common-fluid y coordinates.
+- Scalar support only: `figures/summary/profile_Einf_vector.csv`, `figures/field_metrics.csv`, `figures/manifest.json`.
+- Optional field context: `figures/fields/streamlines/cfd-accepted-s3p55.png` and `cfd-stationary.png`, with plate companions.
+- External authoritative `qin`, `accepted-s3.55` and `stationary` endpoint NPZ/manifests are path/hash-bound in `figures/README.md` and `manifest.json`.
+- Production script: `src/steady_pinball_theory/plot_results.py`; no CFD is required.
+- Readiness: profile rows and source hashes are present; **ready for profile redraw, dependency-gated for final crop/typography**.
+
+## Layout
+
+- Full two-column width, **178 mm × 88–105 mm** before caption.
+- Grid: **2 rows × 2 columns**.
+  - Top row, 35–40% height: optional accepted and stationary streamline panes, one column each.
+  - Bottom row, 60–65% height: streamwise profile $u/U_\infty$ and transverse profile $v/U_\infty$, one column each.
+- If streamlines fail final-size legibility, use a **1 × 2 profile-only** figure and enlarge both profile panels.
+- Reading path: target/controlled/stationary $u$ profile → transverse $v$ residual → optional topology context → small endpoint summary.
+
+## Panels and role order
+
+### (a–b) Optional total-velocity streamlines
+
+- (a) accepted controlled endpoint; (b) stationary/uncontrolled pinball. Exact actuation belongs to Section 8; retain source-case identity in production provenance only.
+- Variable: total velocity rendered as streamlines, not a speed scalar.
+- Crop: $x/D=[-2,10]$, $y/D=[-3.025,2.975]`; array order `yx`; solver-fluid masks; identical density/style.
+- Uniform target is not a blank field pane. It is explicit in the profile panels.
+- Temporal status: accepted endpoint and stationary reference. Section 8 owns settling qualification.
+
+### (c) Streamwise velocity profile at $x/D=10$
+
+- Horizontal axis: $u/U_\infty$; vertical axis: $y/D$.
+- Curves/order: **uniform target** $u/U_\infty=1$ (black dashed) → **accepted controlled** (blue) → **stationary/uncontrolled** (dark grey/orange).
+- Data: rows with `fluid_mask=1` for the selected role/station. Use the common valid y support; do not bridge masked gaps.
+- The target curve uses the same y support and is not inferred from stationary data.
+- $x/D=10$ is mandatory. If $x/D=6$ is included, encode station by line weight/alpha within each role: $x/D=10$ heavy/opaque, $x/D=6$ thin/40–50% alpha. Do not add stations 2 and 4 to the main panel.
+
+### (d) Transverse velocity profile at $x/D=10$
+
+- Horizontal axis: $v/U_\infty$; vertical axis: $y/D$, shared with panel (c).
+- Curves/order: **uniform target** $v/U_\infty=0$ (black dashed) → accepted controlled → stationary/uncontrolled.
+- Use a symmetric x-axis about zero based on the pooled accepted/stationary valid values at the selected station, with modest padding; never normalize each role separately.
+- Keep $u$ and $v$ separate. Their target levels and numerical ranges differ, and a vector-magnitude-only curve would hide sign and transverse cancellation.
+
+### Small station-summary annotation/inset
+
+- Optional, no more than 12–15% of profile area.
+- Do not print exact endpoint-error values in this discovery preview; Section 8 owns them.
+- If a tiny station strip is retained, keep it qualitative and subordinate; it must not displace either component profile.
+- Definition: maximum full-cross-section vector difference from uniform target on common fluid points, normalized by $U_\infty$.
+
+## Normalization, crop, masks and scales
+
+- Velocity components are already exported as `ux_over_U`, `uy_over_U`; do not renormalize.
+- Vertical coordinate is `y_over_D`, x-right/y-up. Retain the full available cross-section by default. A symmetric crop may be used only after checking it excludes no material wake deviation and is documented.
+- Apply identical y limits and common-fluid handling across target/controlled/stationary and across $u/v$ panels.
+- No color scale applies. Role colors are consistent across both components; station is encoded by weight/alpha, not color.
+
+## Labels and legends
+
+- Header: `uniform-inflow / free-slip-wall configuration, Re_D=50`.
+- Profile titles: `(c) streamwise velocity, x/D=10`; `(d) transverse velocity, x/D=10`.
+- One shared role legend in target → controlled → stationary order.
+- If $x/D=6$ is included, use a separate compact station key; do not duplicate legends.
+
+## Caption skeleton
+
+> **Steady discovery through downstream velocity-profile recovery.** Cross-stream profiles at $x/D=10$ compare the uniform target, accepted controlled endpoint, and stationary pinball in the uniform-inflow/free-slip-wall configuration at $Re_D=50$: (c) $u/U_\infty$ and (d) $v/U_\infty$ versus $y/D$. [If retained: (a,b) total-velocity streamlines provide a separate wake-topology view for the accepted and stationary roles.] The controlled profile approaches the target in both components. These are endpoint profiles, not periodic means, and establish neither optimum, stability nor mechanism.
+
+## Inference limits
+The component profiles show that the accepted controlled endpoint approaches the uniform target and that the stationary pinball retains a much larger downstream departure; optional streamlines add a separate topology view. The endpoint comparison does not determine an optimum, attracting branch, stability or mechanism, and it says nothing about periodic phase matching or near-body field error.
+
+## Normalized readiness
+
+- **Plan state:** `specified`; the two-figure Section-3 composition author choice is closed. Exact source/provenance, crop, scale, caption, artwork, rendering and publication-readiness gates remain open.
+- **Evidence state:** named CSV/script/manifest paths are `path-confirmed`; source arrays are `mapped`, not reopened; raw arrays and rendering are `not-covered`.
+- **Metric state:** profile contract `authority-bound`; final crop is `dependency-gated`.
+- **Render state:** `redraw-required`; existing streamlines are diagnostics/source material.
+- **Disposition:** `retain-with-gates`; author may choose the profile-only fallback.
+- **Caption ceiling/fallback:** endpoint profile approach only; no optimum, stability or mechanism; never replace profiles with a scalar or streamline-only view.
+
+## Production checklist
+
+1. Read `profiles_long.csv`; select accepted/stationary, `fluid_mask=1`, $x/D=10$ and optionally 6.
+2. Verify each selected role's `source_sha256` against `figures/manifest.json`.
+3. Build the uniform target arrays on the identical valid y support: $(u/U_\infty,v/U_\infty)=(1,0)$.
+4. Audit accepted/stationary y coordinates and use their common support; preserve masked gaps.
+5. Draw separate $u$ and $v$ panels with shared $y/D$, role colors and target curves; test full y range first.
+6. Keep the Section-3 preview qualitative; omit exact $E_{\infty,vector}$ values and settling qualification, which Section 8 owns.
+7. Test optional streamline panes at final dimensions; retain only if topology remains legible.
+8. Record final y range, optional station, source hashes and styling in figure provenance.
+
+## Fallback
+
+- Required: separate $u$ and $v$ profiles at $x/D=10$ with target, controlled and stationary roles.
+- First drop: $x/D=6$ secondary curves/inset.
+- Second drop: both streamline panes; use profile-only 1 × 2 layout.
+- Never replace profiles with $E_{\infty,vector}$ versus station or a streamline-only panel.
+- Do not add reverse or MFS material to fill space.

docs/JFM_WYQ/FIGURE_PLANS/SECTION_03_FIGURE_02_WAKE_CLOAKING_AND_RETARGETING.md

@@ -0,0 +1,74 @@
+# Section 3 Figure 2 — Wake cloaking and retargeting
+
+## Scientific question
+How does learned rotational control preserve a periodic street, respond through an isolated vortex event and create a declared non-zero wake, and why does the periodic Kármán case warrant deeper analysis?
+
+## Source plots, data and scripts
+
+**Kármán.** Use `src/drl_pinball/data/reproduction/legacy/karman_re100/{target,controlled,zero}/{metadata.json,phase_fields.npz}`, the diagnostic `src/drl_pinball/data/reproduction_plots/legacy/karman_re100.png`, the bound `src/drl_pinball/eval/wake_l2.py` summary, and `src/drl_pinball/plot_reproduction_summary.py`/`src/drl_pinball/plot_flow_fields.py`. Crop/recombine the diagnostic for a layout proof; redraw phase slot 0 for production so crop, scale, units and role labels are controlled and the diagnostic sensor portraits are removed.
+
+**Taylor vortex.** Use `src/drl_pinball/data/reproduction/legacy/vortex_taylor_y000/{target,controlled,zero}/event_fields.npz`, which stores offsets `[-10,-5,0,5,10]`. Existing draft renders for every role/offset are under `src/drl_pinball/data/reproduction_flow_plots/all/legacy/vortex_taylor/vortex_taylor_y000/`; `src/drl_pinball/plot_flow_fields.py` supplies the rendering logic. `docs/My_Confirmation/Graphs/cloak_tylor.pdf` is narrative precedent for the sequence only. Select one source-bound representative role comparison from existing event fields; no CFD or event extraction is required. Detailed offset evolution belongs to Section 7.
+
+**Illusion.** Use `src/drl_pinball/data/reproduction/v5/ill_075L/target`, `.../ill_075L_seed43/controlled`, `.../ill_075L/zero`, `src/drl_pinball/data/reproduction_plots/v5/ill_075L_seed43.png`, and `src/drl_pinball/data/reproduction_plots/manifest.json`. Crop/recombine for a proof, then redraw phase slot 0. This is seed-43 PPO capability for target radius $0.75L_0$.
+
+## Layout and dimensions
+- Two-column width, **178 mm × 128–148 mm** before caption.
+- Top 52–58%: large Kármán target → controlled → physical-zero triplet, each pane equal width.
+- Bottom left 55–60%: one representative Taylor-vortex target → controlled → physical-zero comparison at one source-bound event offset; no offset evolution.
+- Bottom right 40–45%: Illusion target → controlled → physical-zero triplet.
+- Minimum final type 7 pt. Test the single-offset role comparison at 178 mm.
+
+## Panels, variables and coordinates
+
+### Periodic Kármán cloak
+- Variable: $\omega_zD/U_{ref}$ only if normalization is reconstructed from metadata; otherwise print lattice-vorticity units.
+- Coordinate: literal phase slot 0, not a mean or ensemble.
+- Role order: target/reference → frozen PPO controlled → physical zero.
+- Annotation: `complete-cycle mean and eight-slot phase velocity errors each about 82% lower than physical zero`.
+- Exact values remain in Section 4.
+
+### Centred Taylor event
+- Variable: vorticity at one source-bound representative event offset.
+- Role order: target → controlled → physical zero on the same crop and scale.
+- Record the selected literal offset in production provenance; do not show multi-offset evolution in Section 3.
+- No event L2, phase label or inferred continuous-time conversion.
+
+### 0.75-radius Illusion
+- Variable: phase-slot-0 vorticity.
+- Role order: non-zero target-cylinder wake → controlled pinball → physical zero.
+- Label target radius $0.75L_0$, seed 43, PPO.
+- No action, DTW, field-L2 or SR annotation.
+
+## Crops, masks and scales
+Use one scale and physical crop within each direct comparison. All three roles at the selected event offset share one event scale and crop. Kármán and Illusion may use separate case-specific scales because their configurations and raw units differ; print the limits. The manifest's fixed $[-10^{-3},10^{-3}]$ range is a diagnostic plotting choice, not automatic nondimensional authority. Preserve x-right/y-up orientation and source geometry. Do not call geometry overlays persisted solver-exact masks. Avoid full-domain strips that make vortices unreadably small.
+
+## Reading order
+Read the large periodic target/control/zero comparison first, read the single isolated-vortex role comparison, then finish with the changed non-zero target. Area and position carry the hierarchy; no common score or decorative cross-case scale is used.
+
+## Caption draft
+> **Wake cloaking and retargeting under distinct physical comparisons.** (a–c) Target, frozen-PPO controlled and physical-zero vorticity at phase slot 0 for the periodic Kármán cloak in the parabolic-inflow/no-slip-wall configuration. The controlled complete-cycle mean and separately defined eight-slot phase velocity errors are each about 82% lower than physical zero; exact values and windows accompany the detailed Kármán result. (d–...) Target, controlled and physical-zero vorticity at one printed event-relative offset for a centred Taylor-vortex event. This is a representative role comparison, not event evolution or periodic phase, and no event L2 is reported. (...) Non-zero target, seed-43 PPO controlled and physical-zero vorticity at phase slot 0 for the target-radius-$0.75L_0$ Illusion case in the uniform-inflow/free-slip-wall configuration. Scales are shared within each direct comparison and may differ between configurations. The composition compares distinct physical tasks rather than assigning them a common efficacy measure.
+
+## Inference limits
+The Kármán triplet and rounded field statement justify deeper analysis. The representative event view identifies role-field morphology without a quantitative event-field or efficacy claim. The Illusion triplet shows positive PPO retargeting. These views do not supply a pooled result, arbitrary-flow transfer, positive Illusion symbolic reduction or a common mechanism.
+
+## Normalized readiness
+
+- **Plan state:** `specified`; the two-figure Section-3 composition author choice is closed. The exact representative Vortex offset and all source/provenance, crop, scale, caption, artwork, rendering and publication-readiness gates remain open.
+- **Evidence state:** exact repository paths are `path-confirmed`; role payloads are `mapped`, including nonportable links where applicable; raw arrays are `not-covered`.
+- **Metric state:** Kármán rounded comparison is authority-bound; Vortex quantitative efficacy is `blocked`; Illusion is bounded PPO morphology/capability.
+- **Render state:** `redraw-required`; existing montages are diagnostic only.
+- **Disposition:** `retain-with-gates`; fallback keeps one source-bound Vortex triplet and omits unsupported scalars.
+- **Caption ceiling:** distinct role morphology and bounded Kármán comparison only; no pooled efficacy, arbitrary-flow transfer or Illusion SR.
+
+## Production checklist
+1. Verify role metadata, schemas and hashes against the reproduction manifest.
+2. Export Kármán/Illusion phase slot 0 and one selected Taylor offset for all roles; reserve the full offset sequence for Section 7.
+3. Audit vorticity units; retain lattice units if nondimensional conversion is not source-bound.
+4. Produce the single-offset target/control/zero Taylor comparison at 178 mm.
+5. Freeze and record crop, scale, clipping, masks and orientation within each comparison.
+6. Keep target → controlled → physical-zero labels stable and event offsets literal.
+7. Verify that Kármán remains dominant and Illusion remains readable.
+8. Record source files, hashes, scripts, selections and final dimensions.
+
+## Fallback
+Use one source-bound target/controlled/physical-zero comparison at a single literal event offset. If crowded, reduce its footprint but retain all three roles; do not add multi-offset evolution, which Section 7 owns. If Illusion physical zero fails role audit, retain target/controlled and make no over-zero claim. Do not fill space with Reynolds-series, reward, SR/CCD, Erase or experiment panels.

docs/JFM_WYQ/FIGURE_PLANS/SECTION_04_FIGURE_01_KARMAN_CONDITIONS_AND_LEARNING.md

@@ -0,0 +1,57 @@
+# Section 4 Figure 1 — Kármán conditions and repeated learning
+
+## Scientific question
+Across which retained periodic Kármán conditions were high-performing policies found, and how repeatably did the only five-realization condition progress toward high reward?
+
+## Source plots, data and scripts
+Use `src/drl_pinball/train/results/latest/tables/latest_eval_summary.csv`, `src/drl_pinball/train/results/latest/tables/latest_results_summary.json`, `latest/HOW_TO_READ.md`, `train/results/training_csv/training_iterations.csv` and its manifest, `train/results/figures/scratch_training_summary.csv`, and `train/output/kar_re100_seed41–45/meta.json`. Existing logic is in `train/plot_latest_results.py` and `plot_scratch_results.py`; generated `fig01_re100_multiseed_learning`, `fig07_eval_cross_scene_metrics`, `fig12_vorticity_cross_re` and `fig13_vorticity_disturbance` are design references when present locally. **Redraw a focused manuscript composition from the exported tables.** Do not copy dashboards, retention panels or drawdown panels. No CFD or training is needed.
+
+## Layout and dimensions
+- Two-column width, **178 mm × 120–138 mm**.
+- Upper 35–40%: two aligned condition groups—retained reward and DTW versus $Re_D=30,50,100,200$, and versus disturbance-radius ratio $0.75,1.0,1.5,2.0$.
+- Lower 60–65%: the five $Re_D=50$ reward trajectories over 500 training iterations; this is the visual anchor.
+- Reward and DTW occupy separate aligned mini-axes. No dual y-axis or composite score.
+- Reading order: tested range → highlighted reference condition → five repeated learning histories.
+
+## Condition panels
+
+### Reynolds series
+Use retained rows `kar_re60`, `kar_re100`, `kar_re200`, `kar_re400` after the article-wide conversion. Plot reward and six-sensor DTW, both higher-is-better, on separate axes. Label $Re_D=50$ `five searches; selected retained policy shown`; all other points are $n=1$. Use points only: no fitted line, uncertainty bar, interpolation or monotonic-law wording.
+
+### Disturbance-radius series
+Use `kar_d075`, `kar_re100`, `kar_d15`, `kar_d2`. Mark ratios 0.75 and 2.0 with hollow symbols/daggers because they used learning rate $10^{-4}$ rather than the common $3\times10^{-4}$. Every non-reference point is $n=1$. Describe a tested condition series, not a pure diameter ablation.
+
+## Five learning trajectories
+Filter `training_iterations.csv` for `case_id=kar_re100`, seeds 41–45 and iterations 1–500. Show raw thin lines. A 20-iteration rolling median may be overlaid only if the raw curves remain visible and the window is stated. Mark authoritative best checkpoints:
+- seed 41: 0.926638 at iteration 471;
+- seed 42: 0.930196 at 315;
+- seed 43: 0.916007 at 360;
+- seed 44: 0.922196 at 465;
+- seed 45: 0.941161 at 447.
+
+Label the horizontal coordinate `training iteration`—one PPO learning chunk plus deterministic evaluation—not episode. Use a colourblind-safe palette and direct labels or a compact legend. Do not shade the final 50 iterations, print late-window ratios, draw best-to-late slopes or tell a retention-boundary story.
+
+## Scales and statistics
+Use 0–1 reward and DTW ranges for the condition map by default. Keep 0–1 on the trajectories so early development and the common high-reward region remain visible. Do not collapse the five curves to a mean or uncertainty band. Training best and retained deterministic evaluation are different streams; never connect them as paired before/after measurements.
+
+## Caption draft
+> **Periodic Kármán conditions and repeated high-reward learning.** (a,b) Retained deterministic reward and six-sensor DTW similarity for the uniform-inflow/free-slip-wall Reynolds series $Re_D=30,50,100,200$ and disturbance-radius ratios 0.75, 1.0, 1.5 and 2.0. Higher values indicate closer online-objective or signal matching, respectively. Only $Re_D=50$ has five searches; all other conditions are single realizations. The 0.75 and 2.0 disturbance cases used learning rate $10^{-4}$ rather than $3\times10^{-4}$ and are marked accordingly. (c) Reward trajectories for all five $Re_D=50$ stochastic realizations over 500 training iterations, with authoritative best checkpoints marked. All five progress from low reward to high-performing checkpoints, whose best rewards span 0.916–0.941. This supports repeatable high-performance policy discovery across the five tested realizations. Reward and DTW do not measure registered field error, and the displayed points do not define a continuous Reynolds or geometry law.
+
+## Inference limits
+The figure shows a finite tested condition range and repeated high-performance discovery at the only replicated condition. It does not show broad hyperparameter/Reynolds/geometry robustness, statistical uncertainty at $n=1$, convergence guarantees, stability, transfer, current-contract retraining or independent field matching.
+
+## Normalized readiness
+
+Plan `specified`; table/metadata paths `path-confirmed`, generated figures `mapped` when present, raw reruns `not-covered`; reward/DTW orientation and retained rows `authority-bound`, Reynolds labels require metadata audit; render `redraw-required`; disposition `retain-with-gates`. Caption ceiling is finite tested conditions and repeated discovery only; fallback moves disturbance conditions to supplement.
+
+## Production checklist
+1. Verify condition rows, selected seeds and metric orientation in `latest_eval_summary.csv`.
+2. Verify article $Re_D$ labels and disturbance-radius meaning against Section 2 and metadata.
+3. Extract five complete 500-iteration histories; match metadata best points to CSV flags.
+4. Render raw traces and optional documented smoothing; remove retention/drawdown elements.
+5. Mark realization counts and learning-rate confounds directly.
+6. Test separate reward/DTW strips at 178 mm and at least 7 pt.
+7. Record source rows, metadata, hashes, smoothing and plotting revision.
+
+## Fallback
+If crowded, move the disturbance-radius group to the supplement and retain the Reynolds strip plus five trajectories. Do not split the curves into a third main figure, replace them with a mean band or restore retention graphics. The five trajectories are non-negotiable.

docs/JFM_WYQ/FIGURE_PLANS/SECTION_04_FIGURE_02_REFERENCE_FIELD_SIGNAL_ACTION.md

@@ -0,0 +1,79 @@
+# Section 4 Figure 2 — Field matching and periodic control in the reference wake
+
+## Scientific question
+In the deeply documented parabolic-inflow/no-slip-wall Kármán wake, how much closer is the controlled downstream field to its target, and what periodic sensor and applied-action organization accompanies it?
+
+## Source plots, data and scripts
+
+### Fields and scalar errors
+Use `src/drl_pinball/data/reproduction/legacy/karman_re100/{target,controlled,zero}/{metadata.json,phase_fields.npz,timeseries.csv,dtw_summary.json}`, `src/drl_pinball/eval/wake_l2.py` (`drl-pinball-wake-l2-v2`), the diagnostic `src/drl_pinball/data/reproduction_plots/legacy/karman_re100.png`, and `src/drl_pinball/plot_reproduction_summary.py`/`src/drl_pinball/plot_flow_fields.py`. This is the parabolic-inflow/no-slip-wall reference case (historical source label `karman_re100`; manuscript $Re_D$ pending the source-velocity conversion audit). The source label and any apparent numerical correspondence do not join it to the uniform-inflow/free-slip-wall $Re_D=50$ learning condition. Reuse the diagnostic for a proof; redraw/crop the three phase fields for production. Scalar L2 values are ready; fields still require role/hash/unit/crop audit.
+
+### Applied PPO actions and sensors
+Use `src/SR_analysis/results/runs/article2-timeseries-csv-20260720/karman_re100/{target_wide.csv,ppo_wide.csv,ppo_manifest.json}` and `src/SR_analysis/results/runs/article2-plotting-package-20260721/{README.md,phase_alignment.json}`. The exact window is target indices 96–145, $t_D=38.4$–58.0, 50 samples and about three target cycles; current alignment lag is 0. Reuse the CSV window and redraw. Plot PPO applied-action columns only: the target has no action trajectory.
+
+### Spectrum and visual precedent
+Confirmation views `docs/My_Confirmation/Graphs/{vort,error,sensor,freq,action}_cloak_vortex.pdf` guide reading order and compactness only. Recompute a centre-sensor spectrum offline from the same declared target/PPO/zero window if exact zero data and clock are bound; otherwise omit. Do not import old percentages or the claim that harmonics are essential.
+
+## Layout and dimensions
+- Two-column width, **178 mm × 150–172 mm**.
+- Band 1, 40–44%: target → controlled → physical-zero vorticity triplet.
+- Band 2, 18–22%: two L2 dumbbells or paired points (mean and phase) over about 70% width; small target-normalized DTW pair at right.
+- Band 3, 34–40%: applied PPO rotations at left; centre-sensor portrait at upper right; optional compact spectrum at lower right.
+- Reading order: field roles → mean/phase errors → sparse-signal comparison → applied motion.
+
+## Field panels
+- Variable: phase-slot-0 vorticity, labelled $\omega_zD/U_{ref}$ only after unit audit; otherwise exact lattice units.
+- Role order: target/reference → frozen PPO controlled → physical zero.
+- One common crop, geometry treatment and symmetric scale.
+- Display-crop candidate: $x/D=7$–48, $y/D=-5.975$–5.975 from the source-ready PPO package. This is not the metric ROI. It becomes a final display crop only after zero is reconciled on exactly the same region.
+- No eight-phase montage; the displayed field is one literal phase.
+
+## Mean and phase L2
+- Complete-cycle mean: controlled 0.085771; physical zero 0.475206; relative reduction 0.819508.
+- Eight-slot phase diagnostic: controlled 0.112680; physical zero 0.630878; relative reduction 0.821392.
+- Show absolute pairs first and relative reductions as small labels.
+- Registered region: $-6\le(x-x_s)/D\le14$, $|(y-y_0)/D|\le5$, a $20D\times10D$ rectangle.
+- State the geometry guard and absence of a persisted solver-exact common-fluid mask.
+- Mean and phase are separate estimands, not uncertainty or repeated measurements.
+
+## DTW, sensor portrait and spectrum
+- Target-normalized six-channel DTW similarity: controlled 0.932899; physical zero 0.625127; higher is better.
+- Keep DTW visually subordinate to L2; native values belong in the supplement.
+- Portrait axes: centre-sensor $u_x/U_{ref}$ and $u_y/U_{ref}$, role order target → controlled → physical zero, common limits.
+- Use the declared stable window for target/PPO. No exact source for a zero-role sensor portrait is currently bound; treat it as absent/unavailable. Show target/PPO only unless a role-bound comparable zero source is located, and retain zero only in authority-bound scalar metrics.
+- Optional spectrum uses centre-sensor $u_y$ from the exact portrait window, one normalization and one frequency axis. It may show dominant periodic alignment but not harmonic necessity.
+
+## Applied rotations
+- Plot three applied PPO series for samples 96–145: front, upper and lower in native order.
+- Use source-native historical $\alpha=\omega/U_0$ only when the selected columns and role-local metadata bind that definition. Physical conversion is general and conditional: $s=\alpha R(U_0/U_{ref})$, reducing to $s=\alpha R$ only if authority establishes $U_0=U_{ref}$. Otherwise plot exact $\omega$ units. Do not mix commanded and effective/smoothed quantities.
+- Use one shared vertical scale and stable body colours. Print the time window and approximately three-cycle duration.
+- No target action, zero-action target proxy, SR trace or constant-control line.
+
+## Colour, mask and crop controls
+Vorticity uses one symmetric target/control/zero scale and shared colour bar. L2 and DTW use neutral paired-point styling and remain visibly different quantities. Sensor colours encode roles; action colours encode bodies. Preserve x-right/y-up, source geometry and array orientation. Record clipping percentile or fixed limits.
+
+## Caption draft
+> **Registered field matching and periodic control in the reference Kármán wake.** (a–c) Target, frozen-PPO controlled and physical-zero vorticity at phase slot 0 on a common crop and symmetric scale. (d) Complete-cycle mean velocity error is 0.0858 for controlled and 0.4752 for physical zero; the separate eight-slot phase diagnostic is 0.1127 and 0.6309, corresponding to reductions of 0.8195 and 0.8214 over the registered $20D\times10D$ downstream region. The phase diagnostic uses one nearest saved snapshot per independently phased slot, and the region is geometry-guarded because a solver-exact common mask was not persisted. (e) Target-normalized six-channel DTW similarity is 0.933 for controlled and 0.625 for physical zero; higher is better and DTW is a signal measure. (f) Applied PPO rotations over samples 96–145 ($t_D=38.4$–58.0$, approximately three target cycles). (g) Centre-sensor phase portrait [and, if retained, (h) spectrum from the same window]. The target has no action trajectory.
+
+## Inference limits
+The field triplet and L2 show that controlled is closer to target than physical zero under both declared spatial estimands. DTW, portrait and spectrum describe complementary sparse-signal organization, and the applied rotations show the accompanying PPO motion. They do not identify a mechanism, universal periodic law or equivalence with later SR/CCD analyses.
+
+## Portability and normalized readiness
+
+Symlink/external role payloads and manifest-linked outputs are `mapped` but not locally resolved or portable merely because paths are named. Where an Illusion-family directory is referenced by inherited manifests, exact leaf bindings remain `missing-exact-leaf` unless printed in the final production record. A partial plot manifest authenticates only listed diagnostics, not package completeness or panel readiness.
+
+Plan `specified`; controlling summaries/scripts `read` or `path-confirmed`, unresolved payloads `mapped`, raw arrays `not-covered`; L2/DTW `authority-bound`, action units and zero portrait `dependency-gated`; render `redraw-required`; disposition `retain-with-gates`. Caption ceiling is registered field proximity plus accompanying signal/action organization, not mechanism or SR/CCD equivalence; fallback omits spectrum, then DTW, and never fabricates zero alignment.
+
+## Production checklist
+1. Audit role metadata, arrays, clocks, action order and hashes.
+2. Resolve the historical source-velocity convention and manuscript $Re_D$ conversion; until then, omit a Reynolds-number assignment from reader-facing text and retain the parabolic-inflow/no-slip-wall configuration label.
+3. Verify frozen wake-L2 rows without changing the evaluator.
+4. Export phase slot 0 for all roles; join zero only under common crop, phase and scale.
+5. Extract target/PPO indices 96–145 and a role-bound zero window if available.
+6. Select one applied-action quantity and label it consistently with Section 2.
+7. Compute a spectrum only from the displayed signal/window and record detrending/windowing/normalization.
+8. Keep L2 primary; drop spectrum before shrinking text below 7 pt.
+9. Record roles, indices, crop, scales, scripts and output hashes.
+
+## Fallback
+First drop the spectrum. Second, move DTW to caption/prose. The required core is the target/control/zero field triplet, separate mean/phase L2, PPO action traces and one sensor portrait. If zero cannot be phase-aligned visually, retain target/PPO fields and show zero in L2; never fabricate alignment. Do not substitute SR, constant-reference or CCD fields, target actions, uncertainty bars or an eight-phase gallery.

docs/JFM_WYQ/FIGURE_PLANS/SECTION_05_KARMAN_SR_CONTROL_LAW.md

@@ -0,0 +1,89 @@
+# Section 5 figure specification — compact Kármán-cloaking control law
+
+**Proposed main figure:** `A compact control law for Kármán cloaking`. **Owner:** Section 5. **Status:** production `dependency-gated`; existing images are diagnostic drafts only. **Configuration:** parabolic-inflow/no-slip-wall channel configuration. Internal provenance IDs remain in paths, not reader-facing labels.
+
+## Overall layout and argument
+
+Use a physics/results-centred four-panel composition: top row (a) closed-loop SR actions and performance, (b) direct controlled-flow result; bottom row (c) compact finite-deployment envelope if retained, (d) parent-relative deletion. Panel (c) is drop-first. No geometry/action schematic, reflection-map diagram or formula decomposition; Section 2 owns geometry/sign and the text owns $G$ and the law.
+
+## Panel (a) — Closed-loop SR behavior in the reference source case
+
+- **Argumentative job:** show finite, organized closed-loop actuation dominated by persistent opposite rear rotation with smaller modulation; connect action structure to achieved signal/field performance.
+- **Exact layout:** left two-thirds: three aligned time traces of the historical effective applied $\alpha_F,\alpha_U,\alpha_L$ over 160 retained boundaries, one shared nondimensional time/control-boundary axis. Right inset: native-DTW mean and spatial-L2 summary for SR and physical zero relative to target. Do not plot PPO unless the common-target dependency closes.
+- **Variables:** historical effective applied $\alpha=\omega/U_0$, labelled with its source lattice/nondimensional-length convention; optionally centre-sensor $u_y/U_{ref}$ only if legible. This trace is not `effective_smoothed_omega/U_ref`. If a physical surface command is discussed, use $s=\alpha R(U_0/U_{ref})$ and reduce it to $s=\alpha R$ only after source authority establishes $U_0=U_{ref}$. Never mix commanded and effective action.
+- **Roles/comparators:** SR primary; target reference; physical zero comparator. PPO only after reconciliation.
+- **Case/Re/configuration:** internal `legacy/karman_re100`, neutral reference source-case label, parabolic inflow/no-slip walls, 1280×512 acquisition.
+- **Metric/window:** 480 warm-up intervals, 160 retained boundaries; native DTW window 30. SR native mean `0.942357`; $E_{mean}=0.125404$ versus zero `0.475206`; $E_{phase8}=0.156040$ versus zero `0.630878`. Registered ROI $-6\le(x-x_s)/D\le14$, $|y-y_0|/D\le5$, geometry-guarded without a persisted solver-exact mask.
+- **Data/artifact sources:** `src/drl_pinball/data/reproduction/legacy/karman_re100/{sr,zero,target}/timeseries.csv`; role `metadata.json` and `dtw_summary.json`; `src/drl_pinball/data/reproduction/sr_wake_l2_summary.{json,csv}`.
+- **Processing/render source:** new publication extraction/plotting required. Parsing conventions exist in `src/drl_pinball/plot_reproduction_summary.py`, which currently renders sensor portraits rather than action traces.
+- **Readiness:** `available-data` for SR/zero/target; `dependency-gated` for PPO; publication renderer absent.
+- **Visual conventions:** front dark grey; upper rear blue; lower rear orange; zero/reference light grey. Identical action limits; annotate persistent rear offsets without decorative decomposition.
+- **Caption claim:** the executable law sustains dominant opposite rear rotation with smaller time-dependent corrections and remains close to the declared target under the named standardized window.
+- **Explicit non-claims:** no PPO symmetry, plant-symmetry proof, mechanism, stability, robustness, necessity, field equivalence from DTW or cross-configuration transfer.
+- **Supplement fallback:** full action/sensor/reward traces and commanded-versus-effective comparison.
+
+## Panel (b) — Direct SR-controlled wake result
+
+- **Argumentative job:** directly show that SR deployment moves the downstream wake toward the Kármán target relative to physical zero.
+- **Exact layout:** horizontal triptych Target | SR | physical zero, all phase slot 0, one shared domain retaining pinball and registered downstream ROI, one shared vertical colour bar. Put compact $E_{mean}$ and $E_{phase8}$ annotations under SR and zero; no sensor portraits.
+- **Variables:** spanwise vorticity $\omega_z=\partial_xu_y-\partial_yu_x$ from `ux`,`uy`; optional unobtrusive ROI outline.
+- **Roles/comparators:** target reference; SR controlled; physical-zero comparator. PPO omitted by safe default.
+- **Case/Re/configuration:** same neutral source case as panel (a).
+- **Metric/window:** slot 0 is one nearest-boundary snapshot on each independently phased role's common mathematical phase coordinate; complete-cycle mean and eight-slot diagnostics as above; no global phase minimization/interpolation.
+- **Data/artifact sources:** `.../karman_re100/{target,sr,zero}/phase_fields.npz`, metadata/hashes in `src/drl_pinball/data/reproduction_plots_sr/manifest.json`; metrics in `sr_wake_l2_summary.json`.
+- **Processing/render source:** vorticity computation in `src/drl_pinball/plot_reproduction_summary.py`; draft `src/drl_pinball/data/reproduction_plots_sr/legacy/karman_re100.png`.
+- **Readiness:** `available-data`, `draft-only` rendering; needs publication crop, typography, labels, ROI annotation and hash/access check.
+- **Visual conventions:** fixed symmetric vorticity scale, initially $[-10^{-3},10^{-3}]$ lattice$^{-1}$; diverging map centred at zero; identical extent/aspect and cylinder outlines. No role-specific limits.
+- **Caption claim:** SR is closer to target than physical zero in complete-cycle mean and eight-slot spatial L2; the slot-0 vorticity view illustrates that named comparison.
+- **Explicit non-claims:** snapshot is not itself the L2 estimand; no exact solver mask, full-state identity, causality, repeated-cycle uncertainty or equivalence to Section 4 PPO fields.
+- **Supplement fallback:** all eight phase slots and current sensor phase portraits.
+
+## Panel (c) — Finite formula-parent deployment envelope (optional)
+
+- **Argumentative job:** show one frozen law remains executable over named finite deployments while similarity weakens toward the highest tested condition.
+- **Exact layout:** compact points of native Legacy DTW similarity against neutral source-case labels, separate markers for 200- and 400-control-step deployments; no fitted trend.
+- **Variables:** native similarity (higher is better) versus neutral source-case label.
+- **Roles/comparators:** formula-parent SR versus each declared target; not PPO/zero.
+- **Case/Re/configuration:** internal `Re_code` values 50/100/200/400 retained in provenance only; use neutral source-case labels pending velocity-reference authority, in the same parabolic/no-slip configuration.
+- **Metric/window:** native `legacy_dtw_v1_abs_n_unclipped`; 200-step `0.9543,0.9427,0.8560,0.7827`; 400-step `0.952178,0.944339,0.863850,0.833365`.
+- **Data/artifact sources:** `src/SR_analysis/results/runs/article-L3-karman-20260718/`, `article2-long-karman-20260720/`; `src/SR_analysis/CLAIMS.md`, `results/README.md`.
+- **Processing/render source:** no publication renderer identified; Article2 plotting/CSV products are derived inputs.
+- **Readiness:** `available-data`, production `dependency-gated`; exact row extraction/labels needed.
+- **Visual conventions:** discrete markers and neutral source-case labels; lines only as visual guides; state one realization per condition/window. No reader-facing Reynolds values.
+- **Caption claim:** the law completes the named 200- and 400-step deployments, with weaker similarity in the later neutral-labelled source cases.
+- **Explicit non-claims:** no asymptotic stability, robustness distribution, continuous Reynolds law, generalization or same-protocol PPO/zero comparison.
+- **Supplement fallback:** omit from main figure and move deployment/unseen-condition diagnostics to supplement (preferred compression).
+
+## Panel (d) — Parent-relative term deletion
+
+- **Argumentative job:** rank tested components: rear constant largest, rear-lift feedback secondary, front correction weak/mixed.
+- **Exact layout:** grouped dot plot of $\Delta S=S_{deleted}-S_{parent}$ versus the same neutral source-case labels, series for delete rear constant (`k_rear1`), rear-lift feedback (`k_rear0`) and front correction (`k_front0`); zero line; negative means degradation. No uncertainty bars.
+- **Variables:** change in native DTW similarity relative to same-case parent.
+- **Roles/comparators:** each deletion only against its canonical parent at the same case and 40-step window.
+- **Case/Re/configuration:** same four Kármán cases/configuration as panel (c).
+- **Metric/window:** exact native DTW contract: six channels; one shared historical circular lag; score each channel at that shared lag and then take the channel mean; higher is better. Use the same fixed 40-control-step window for each parent/deletion pair and plot $\Delta S=S_{deleted}-S_{parent}$. Rear constant: `-0.106537,-0.093869,-0.178525,-0.191082`; rear feedback: `-0.028755,-0.041660,-0.061363,-0.048948`; front: `-0.006520,+0.000579,-0.005179,-0.025194`.
+- **Data/artifact sources:** `src/SR_analysis/results/runs/article2-plotting-package-20260721/tables/term_deletion.{csv,md}` and immutable run families named there.
+- **Processing/render source:** new publication plot required. Existing standardized field drafts `src/drl_pinball/data/reproduction_plots_sr/legacy/variants/karman_re100_k_{front0,rear0,rear1}.png` use longer acquisitions and do not generate the 40-step deltas.
+- **Readiness:** numerical plot `available-data`; existing field images `draft-only` and non-equivalent in window; renderer absent.
+- **Visual conventions:** colourblind-safe palette, common marker positions/y-axis, label physical term rather than internal ID; caption says parent-relative fixed-window deletion.
+- **Caption claim:** deleting the persistent rear constant produces the largest listed degradation across all four cases; rear-lift feedback is smaller and non-zero; tested front correction is weak/mixed.
+- **Explicit non-claims:** not causal ablation, necessity, sufficiency, unique decomposition, mechanism, uncertainty, long-horizon ranking or universal sensitivity.
+- **Supplement fallback:** full deletion table, retained formulas and standardized variant field diagnostics.
+
+## Historical-alpha extraction and DTW non-interchangeability
+
+For every retained role, extract historical effective applied alpha by dividing that role's `effective_smoothed_omega_*` columns by its role-local `metadata.json` `u0`. Record the exact retained rows, control-boundary axis, source units, role metadata path and source hashes in the production manifest. Do not infer a reader physical Reynolds label and do not mix commanded omega/alpha with effective smoothed quantities. A neutral internal-to-source-label mapping and exact row/run binding are mandatory.
+
+The standardized rolling native-DTW diagnostic uses window 30 and a role summary; its lag channel is recorded in that summary. The fixed-cycle deletion diagnostic uses the separately specified 40-control-step implementation contract with one shared historical circular lag. They are non-interchangeable and must not share labels, rows or interpretations. Partial plot manifests authenticate only the diagnostics they enumerate, not payload completeness, portability or readiness.
+
+## PPO inclusion dependency and honest fallback
+
+The standardized tree and draft `karman_re100.png` contain Target/PPO/SR/Zero, but PPO and SR acquisitions bind independently generated target arrays with different hashes. Before a quantitative four-role panel, reconcile target identity and compute every role with one declared metric/window. Until then, use formula-parent deployment and standardized Target/SR/physical-zero evidence; Section 4 retains PPO performance ownership.
+
+## Normalized readiness
+
+Plan `specified`; repository authorities/paths `read` or `path-confirmed`, symlink/external payloads `mapped` and possibly not locally resolved, raw arrays `not-covered`; standardized SR metrics and deletion rows `authority-bound`, exact extraction/run/hash manifest `dependency-gated`, PPO common-target comparison `blocked`; render `redraw-required`; disposition `retain-with-gates`. Caption ceiling excludes causality, mechanism, physical Reynolds conversion and commanded/effective mixing. Fallback omits PPO and deployment envelope, retaining law, Target/SR/zero and fixed-window deletion.
+
+## Production checklist
+
+Verify external payload access and hashes; extract effective actions and retained axis; freeze neutral source-case labels pending velocity-reference authority; decide panel (c) omission; reconcile targets before PPO inclusion; preserve slot/phase/mask caveats; create a dedicated publication renderer/output sibling without overwriting evidence; record source hashes, environment and final crop/colour limits.

docs/JFM_WYQ/FIGURE_PLANS/SECTION_06_FIGURE_SPEC.md

@@ -0,0 +1,172 @@
+# Section 6 figure specifications — Kármán mean fields and CCD
+
+**Status:** textual production specification. Inventory and existing drafts do not imply manuscript-panel readiness.
+
+## Shared scientific identity
+
+- **Physical case:** periodic Kármán cloak, parabolic-inflow/no-slip-wall channel configuration; inherited code label 100 retained in provenance only; use a neutral reference source-case label pending velocity-reference authority.
+- **Mean-field roles:** target, physical zero, constant reference at the exact retained DRL physical action mean, and DRL.
+- **Primary mean estimand:** coordinate-weighted vector RMS target error of retained role means on the exact four-role common solver-fluid mask, inclusive `34<=x/D<=54`, `|y/D|<=5`, 80,200 fluid points. Persisted velocity is already nondimensional.
+- **Residual identity:** separately centred DRL cycle/bin field minus the separate constant-control 19-cycle ensemble phase template; 19 DRL cycles × 10 bins, `N=190`, no cross-role pairing.
+- **Internal provenance only:** current corrected result and publication paths may retain internal chain identifiers; captions use the physical configuration above.
+
+## S6-1 — Mean streamwise correction and vector-error accounting
+
+### Argumentative job
+
+Show the physically legible streamwise pattern first—passive deficit relative to target and controlled addition relative to physical zero—then bind it to the exact two-component scalar error-reduction estimand. Establish mean dominance without calling it a physical/causal contribution.
+
+### Exact layout
+
+- **Top row, four equal field panels (a–d):** `mean_target[u_x]`, `mean_zero[u_x]`, `mean_constant_mean[u_x]`, `mean_drl[u_x]`.
+- **Bottom-left, two wider signed-difference panels (e–f):** `mean_zero[u_x] - mean_target[u_x]`; `mean_drl[u_x] - mean_zero[u_x]`.
+- **Bottom-right, one compact ledger (g):** target-error markers/bars at zero `0.4694352273`, constant `0.1110140739`, DRL `0.0857787860`; ordered brackets `zero→constant=0.3584211534`, `constant→DRL=0.0252352879`, total `0.3836564413`; labels `93.42%` and `6.58%` explicitly as shares of the measured scalar error reduction.
+- Crop every field to the declared ROI; show body/ROI context only in a small locator inset if needed, because bodies lie upstream of this ROI.
+
+### Variables and subtraction order
+
+- Panels (a–d): retained mean streamwise velocity, not speed or vector magnitude.
+- Panel (e): **physical zero minus target**. Negative values denote lower streamwise velocity than target under the repository coordinate/sign convention.
+- Panel (f): **DRL minus physical zero**. Positive values denote streamwise velocity added relative to passive zero.
+- The two fields are compared visually as signed complements; do not label one as an exact negative of the other.
+- Ledger uses both `u_x` and `u_y`; it is not computed from panels (e–f) alone.
+
+### Roles, case, metric, ROI, mask and window
+
+- Roles: target / physical zero / constant reference / DRL from one corrected four-role campaign.
+- Case/configuration: periodic Kármán cloak, parabolic inflow and no-slip channel walls, inherited code label 100.
+- Metric: `E_r=[sum_K w_k ||mean(q_r)-q_target||_2^2/sum_K w_k]^(1/2)`; coordinate weights duplicated across both velocity components.
+- ROI/mask: inclusive `34<=x/D<=54`, `|y/D|<=5`; exact intersection of all four solver-derived fluid masks; 80,200 points.
+- Window: retained means bound by the corrected role campaigns; do not substitute Section-4 complete-cycle standardized means. Exact retained role/window provenance must be copied into the final production manifest: role identities, parent source leaves and hashes, retained row/slice bounds, cycle/bin windows, mask/weight contract, array orientation/transposition, render command/revision, crop/scale/clipping, and output hashes. Until that record exists, production remains dependency-gated.
+
+### Data and artifact sources
+
+- Exact arrays: `/home/frank14f/optane/DynamisLab/ccd/karman-dynamic/canonical/results/roi-mean/arrays.npz`.
+- Result summary/manifest: same directory `summary.json`, `manifest.json`; source manifest hash `f7f6141042678e52314a4d8f76d589108da8e844c8ce504f8f523a6f88a17e73`.
+- Repository authority: `src/CCD_analysis/{RESULTS_INDEX.md,FINAL_RESULTS.md,MATHEMATICAL_DERIVATION.md,FIGURES_AND_DATA.md}` and `src/CCD_analysis/karman_dynamic/{CORRECTED_MATH_CONTRACT.md,dynamic_increment.py,publication.py}`.
+- Existing drafts: external and repository reader copy `publication-review-fix-per-cycle-20260808T1736+0800/{01_roi_mean_performance,02_roi_mean_target_error_fields,03_roi_mean_corrections}.{png,pdf}`.
+- Existing render source: `src/CCD_analysis/karman_dynamic/publication.py`.
+
+### Readiness
+
+- **Arrays/values:** ready and authority-bound.
+- **Existing useful panels:** draft-ready only. `02` already contains `zero-target` for `u_x/u_y`; `01` supplies the scalar ledger; `03` supplies `constant-zero` and `DRL-constant` but not the requested `DRL-zero`.
+- **Requested composite:** dependency-gated new render. `DRL-zero` must be directly calculated as `mean_drl - mean_zero` from the stored means and recorded in a future output manifest with input paths/hashes, subtraction order, units, orientation and output hash; no substitution by `DRL-constant` is allowed. No CFD or new scientific estimator is required.
+- **Independent sign check:** weighted ROI mean-$u_x$ cosine of requested fields `-0.9961` and relative residual `0.1662` were computed read-only for design verification. Do not print these values unless the coordinator extends current claim authority.
+
+### Color and sign conventions
+
+- Four absolute mean-$u_x$ panels: one shared sequential or diverging scale chosen from all four arrays; label units/nondimensionalization exactly from authority. If a sequential scale obscures deviations, use a physically meaningful common scale but never independently autoscale roles.
+- Difference panels: one shared symmetric diverging scale centred at zero, preferably `RdBu_r`; positive must mean the named first role has larger `u_x` under the printed subtraction.
+- Mask non-fluid cells; transpose stored `(x,y)` arrays for plotting. Never infer sign from color without a labelled color bar.
+- Optional zero contours on both difference panels should use identical styling.
+
+### Caption claim
+
+“Relative to the target, the passive pinball produces a downstream mean-streamwise deficit over much of the declared wake ROI, while DRL adds a largely opposite-signed streamwise correction relative to physical zero. This pattern is compatible with adding streamwise velocity where the passive configuration produces a deficit. The scalar ledger evaluates both velocity components on the exact common mask: the constant reference accounts for 93.42% of the measured zero-to-DRL scalar mean-field error reduction and the constant-to-DRL increment accounts for the remaining 6.58%.”
+
+### Explicit non-claims
+
+No momentum restoration; no exact cancellation; no closed momentum/energy budget; no causal actuator effect; no mechanism; no physical contribution percentage; no energy/explained variance; no SR-term share; no Section-4 replication/agreement; no cross-case universality.
+
+### Supplement fallback
+
+If the seven-panel composition is illegible, retain panels (e–g) in main text and move the four absolute means to Supplement S6-S0. If raw composite production fails, use the existing scalar ledger plus `zero-target` draft and state that `DRL-zero` is omitted; do not substitute `DRL-constant` under the wrong label.
+
+## S6-2 — Non-paired residual and action-coordinate phase organization
+
+### Argumentative job
+
+Explain what CCD adds after the mean ledger: a low-rank representation of the separately centred residual associated with action coordinates and organized over phase. Keep the result descriptive and avoid a mode gallery.
+
+### Exact layout
+
+- **Panel (a), left 25%:** residual construction schematic with the equation `U_(c,b)=(q^D_(c,b)-bar q_D)-(hat q^C_b-bar q_C)`, `19×10`, and a visible “cycles not paired” annotation.
+- **Panel (b), centre 25%:** compact 3×3 action-coordinate matrix or bars: rows `front`, `rear-symmetric`, `rear-antisymmetric`; columns modes 1–3. State global mode-sign arbitrariness.
+- **Panels (c–d), right 50%:** rank-3 mean residual reconstructions at two registered representative phase bins nearest `0` and `pi` (use four bins nearest `0,pi/2,pi,3pi/2` only if final-size legibility passes). Show signed `u_x`; use the exact registered bins from `representative_phase_bins`.
+- Do not include all individual mode fields in main text. A small rank/stability note may state `rank 3`, not singular-strength shares.
+
+### Variables, roles and estimand
+
+- Field residual uses both velocity components internally; displayed phase reconstruction is `u_x` and must be labelled as a component view.
+- Constant phase template is the ensemble mean over its own 19 cycles. DRL and constant cycles are not paired.
+- Observable is same-boundary DRL effective action in native front/upper/lower order, exact retained physical action mean removed, then empirical centering.
+- Operator: `A=P_c(W^(1/2)U_c)^T/(N sqrt(3))`, `N=190`, `Q=1`; no whitening, standardization, lag, wrap or POD preprojection.
+
+### Domain, source and readiness
+
+- Preferred display domain: pinball-inclusive `29<=x/D<=54`, `|y/D|<=5`, exact common-fluid mask, 99,264 points, `M=198528`.
+- Preferred result: `/home/frank14f/optane/DynamisLab/ccd/karman-dynamic/canonical/results/pinball-domain-20260809T1925+0800/`, manifest hash `7e107a895364ed3e34abf2be513a3ce9e12736766ec6383c5b831891802d5828`.
+- Repository plot arrays: `src/CCD_analysis/data/karman-dynamic/canonical/publication-pinball-domain-20260809T1925+0800/plot_data.npz` and `plot_data.json`.
+- Existing drafts: `01_domain_estimand`, `02_ccd_mode_identity`, `03_phase_reconstruction`, `04_raw_pod_vs_ccd` PNG/PDF groups.
+- Render sources: `src/CCD_analysis/karman_dynamic/{domain_comparison.py,mode_diagnostics.py,pod_baseline.py}`; standalone single-mode redraw helper `src/CCD_analysis/plot_ccd.py` is insufficient for the full composite.
+- **Readiness:** source arrays and existing views ready; manuscript recomposition dependency-gated. Phase panels may be extracted/redrawn without new CFD. The residual schematic is specification-ready.
+
+### Color and sign conventions
+
+- Phase residual `u_x`: one shared robust symmetric diverging limit across selected phases, centred at zero.
+- Mode/action columns: if individual mode signs are shown, use the same sign convention as the bound publication and state that global sign is arbitrary. Physical phase sign is carried by coefficient × mode, not mode color alone.
+- Stored plot arrays use `(x,y)` and require transpose for display. Overlay exact mask; body circles are geometry guides only.
+
+### Caption claim
+
+“After separate centring, the DRL fields are compared with an independently formed constant-control phase template; cycles are not paired. CCD associates the resulting rank-3 residual representation with centred same-boundary action coordinates and displays its phase organization. These are action-correlated correction-field modes, not causal responses or unique flow structures.”
+
+### Explicit non-claims
+
+No paired counterfactual; no response time/lag; no causal direction; no actuator authority; no mechanism; no wall-vorticity interpretation; no explained variance from CCD singular strengths; no individual-mode stability claim; no SR derivation; no independent triangulation.
+
+### Supplement fallback
+
+If action matrix plus phase fields remains crowded, main text keeps residual schematic + two phase reconstructions and moves action-coordinate matrix with all individual modes to S6-S1. Text then states the association without showing a mode gallery.
+
+## Shared normalized readiness
+
+Plan `specified`; repository authorities and named repository copies `read/path-confirmed`, external absolute payloads `mapped` rather than newly audited, raw-array recomputation `not-covered`; S6-1/S6-2 estimands `authority-bound` but exact retained role/window and final render-manifest provenance `dependency-gated`; render `redraw-required`; disposition `retain-with-gates`. A partial plot manifest authenticates only available diagnostics, not completeness/readiness. Caption ceiling remains descriptive/noncausal and different-estimand; fallback uses existing scalar ledger plus `zero-target`, never a substituted difference.
+
+## S6-S1 — Raw POD boundary and CCD diagnostics (supplement)
+
+### Argumentative job and layout
+
+Define familiar POD as a variance/energy-optimal field organization under the declared weighted metric without action information; compare its rank-3 subspace with CCD on the same centred residual. Use a compact principal-cosine strip/table plus optional aligned POD/CCD mode pairs and stability inset. Detailed mode parity and body-local rotation proxy, if retained, belong here.
+
+### Exact source and metric
+
+- Same pinball-inclusive residual/domain/result as S6-2.
+- Raw full-fit weighted POD Gram matrix `G=U_c^T W U_c`; no action observable selects POD.
+- Principal cosines are singular values of `Phi^T W Psi`: `[0.9999616609, 0.9998971005, 0.9976439804]`.
+- Minimum declared bin/cycle rank-3 stability cosine `0.9974651867`.
+- Source: preferred `04_raw_pod_vs_ccd` draft and compact result arrays; math in `MATHEMATICAL_DERIVATION.md`.
+
+### Readiness, caption and non-claims
+
+**Readiness:** numerical and draft-view ready; publication recomposition required. Caption claim: POD and CCD span essentially the same tested rank-3 field subspace; CCD supplies action-coordinate association, not a distinct field basis. Do not claim CCD/POD superiority, exact identity, held-out prediction, independent uncertainty, unique modes, causal relevance or energy meaning for CCD singular strengths.
+
+## S6-S2 — Optional cross-cloak signed mean-$u_x$ audit (supplement only)
+
+### Argumentative job
+
+Test whether the visual `zero-target` / `controlled-zero` sign pattern recurs beyond the deep case without transferring its quantitative estimand or interpretation.
+
+### Candidate source inventory
+
+- Linked root: `src/drl_pinball/data/reproduction` → external reproduction authority.
+- Periodic `phase_fields.npz` normally stores `mean_ux`, `mean_uy` from complete cycles.
+- Directly inspected available means:
+  - parabolic/no-slip inherited Kármán labels `karman_re50`, `karman_re100`, `karman_re200`, `karman_re400`;
+  - uniform/free-slip Reynolds cases `kar_re60`, five `kar_re100` controlled realizations, `kar_re200`, `kar_re400`;
+  - uniform/free-slip disturbance-size cases `kar_d075`, `kar_d15`, `kar_d2`.
+- Existing images: `src/drl_pinball/data/reproduction_plots/{legacy,v5}/*.png` show phase-0 vorticity, not mean-$u_x$ differences.
+- Scripts: `plot_reproduction_summary.py` renders phase-0 vorticity; `plot_flow_fields.py` reads complete means but renders nonnegative vector target-error magnitude, not signed mean-$u_x$ pairs.
+
+### Required layout and per-panel contract if commissioned
+
+For each selected case, two adjacent fields: `mean_zero[u_x]-mean_target[u_x]` and `mean_controlled[u_x]-mean_zero[u_x]`. Print physical configuration, case, controlled realization/seed, complete-cycle role windows, mask policy and scale above each pair. Use shared symmetric scale within a case pair; use one global scale across cases only after unit/nondimensionalization and range audit. Never combine configurations in a common scalar score.
+
+### Readiness and dependencies
+
+**Source arrays:** available for the cases listed. **Existing signed plots:** absent. **Render code:** absent for this exact panel. **Panel readiness:** **S6-S2 remains blocked** pending case selection, role/window and mask audit, nondimensionalization/scale decision, and a dedicated read-only renderer. A preliminary unweighted full-grid check found broadly opposite signs in most cases but weakening at harder/high-Re/large-disturbance conditions; it is diagnostic only and must not appear as a result.
+
+### Caption claim and non-claims
+
+Allowed caption after audit: “Selected periodic cloak cases show qualitatively opposing passive-deficit and controlled-minus-passive mean-streamwise patterns under their own role and window contracts.” Prohibit common quantitative complementarity, exact cancellation, universal cloak mechanism, transfer of `93.42/6.58`, transfer of the exact common-mask ROI, CCD interpretation, robustness/generalization, or momentum restoration.

docs/JFM_WYQ/FIGURE_PLANS/SECTION_07_FIGURE_01_VORTEX_EVENT_RESPONSE.md

@@ -0,0 +1,84 @@
+# Section 7 Figure 1 — Vortex event response
+
+## Argumentative job
+
+Show role identity and morphology for one nonperiodic incoming-event acquisition under an event-relative contract at central Taylor-vortex incidence and offset `+10`. No efficacy result, reactive-control conclusion or arbitrary-flow implication is available before a versioned evaluator closes the quantitative gate.
+
+## Selected case, roles and variables
+
+- Physical case: central Taylor-vortex incoming disturbance in the parabolic-inflow/no-slip-wall channel configuration.
+- Provenance ID: `legacy/vortex_taylor_y000`; internal label remains caption provenance only.
+- Selected event coordinate: `relative_offset=+10`, array slot 4. Call it an event offset, never phase.
+- Exact roles: independently acquired `target`, frozen-PPO `controlled`, and `physical-zero`.
+- Primary variable: vorticity `omega_z` for morphology, with optional streamlines as a thin overlay or paired strip.
+- Quantitative target-error variables remain blocked pending a versioned evaluator. Do not print provisional normalization, ROI-error maps or scalar comparisons.
+- Event/action context: a narrow action trace or author-drawn event schematic may locate offset `+10`. The author draws any geometry/event schematic; do not infer physical time from the integer offset without an authority conversion.
+
+## Source artifacts
+
+- Fields: `src/drl_pinball/data/reproduction/legacy/vortex_taylor_y000/{target,controlled,zero}/event_fields.npz`.
+- Metadata/timing/actions: sibling `metadata.json`, `event_summary.json`, `timeseries.csv` and `timeseries.npz`.
+- Draft renders: `src/drl_pinball/data/reproduction_flow_plots/all/legacy/vortex_taylor/vortex_taylor_y000/{target,controlled,zero}/`.
+- Relevant ready files include `vorticity/event4_offset+10.000.png`, `streamlines/event4_offset+10.000.png`, and controlled/zero `error/event4_offset+10.000.png`.
+- Rendering authority: `src/drl_pinball/plot_flow_fields.py`; source/render lineage is recorded in `data/reproduction_flow_plots/all/manifest.json`.
+- Existing summary montage `data/reproduction_plots/legacy/vortex_taylor_y000.png` uses offset `-10` and is navigation/draft material, not the selected main result.
+
+## Layout
+
+Use a compact two-tier composition, not five offsets by three roles.
+
+1. **Top context strip (15–20% height):** author-drawn incoming-event marker plus the three physical action commands/effective actions around the selected event. Mark the selected boundary `o=+10`. If trace legibility is poor, retain only the schematic and move action traces to supplement.
+2. **Main row (55–60%):** target, controlled and physical-zero vorticity fields, in that order, cropped on identical coordinates. Label roles above panels; label `event offset +10` once for the row.
+3. **Bottom evidence row (20–25%):** omit target-error maps and scalar comparisons until a versioned evaluator exists. Fallback: physical-variable target, controlled and physical-zero role fields only, with a compact acquisition-status note outside the visual comparison.
+
+Do not include Lamb and five vertical Taylor incidences in the main figure. They may form a supplementary coverage strip only after the central result is fixed.
+
+## Crop and morphology scales
+
+- Full-field crop should retain incoming structure, pinball and downstream sensing context while removing uninformative far boundaries. Use identical pixel/domain bounds for all three roles.
+- Vorticity: one symmetric common scale across target/controlled/zero; current draft setting is `[-0.0015,+0.0015]` in `plot_flow_fields.py`. Confirm final nondimensional label before publication.
+- Streamlines use identical density/crop; do not let streamline density imply metric weight.
+
+## Metric semantics and event-L2 status
+
+No stored event-L2 result was found. `src/drl_pinball/eval/wake_l2.py` excludes Legacy Vortex and reads only periodic `phase_fields.npz` or steady `late_field.npz`; it does not consume `event_fields.npz`. Existing error PNGs are pixelwise draft maps, not stored ROI-L2 summaries.
+
+No provisional evaluator values, normalization constants or ROI-derived efficacy diagnostics are retained in this plan.
+
+## Candidate versioned evaluator requirements
+
+Before any quantitative display can be reconsidered, a separately authorized versioned evaluator would need to:
+
+1. validates exact `event_fields.npz` keys, finite equal shapes and identical `relative_offsets` across target/controlled/zero;
+2. selects offset `+10` by value, never by assumed slot alone;
+3. verifies role metadata, case/scenario identity, normalization authority, field quantity and source hashes;
+4. applies the declared geometry guard/common-fluid policy and exact inclusive ROI;
+5. computes per-role two-component RMS error and `eta=1-E_controlled/E_zero`;
+6. stores versioned JSON/CSV with formula, units, ROI, role paths/hashes, offset and limitations;
+7. regenerates publication-resolution vorticity/error panels from arrays, not by resizing draft PNGs.
+
+This is an evaluator-design gate, not authorization to run it and not a pre-result.
+
+## Caption interpretation
+
+Caption sequence: identify the central Taylor event and offset `+10`; define target/controlled/physical-zero roles; state that offsets are event-relative; state quantitative semantics only after a versioned output passes audit; until then report role identity, event coordinate and morphology only.
+
+Allowed interpretation before the gate closes: the named acquisition provides target, controlled and physical-zero event-relative role fields with distinguishable morphology. No efficacy or reactive/local-information conclusion is licensed.
+
+## Readiness
+
+- Role fields and metadata: **available and audited for existence/schema**.
+- Offset `+10` draft vorticity/streamline/error images: **available; draft/source quality**.
+- Event action timeseries: **available; selection/render still required**.
+- Stored event-L2: **absent**.
+- Candidate evaluator inputs: **identified but not a result or execution authorization**.
+- Publication scalar/panel: **dependency-gated on versioned evaluator output and redraw**.
+- Fallback: show physical-variable target/controlled/physical-zero role fields only; no error maps or scalars.
+
+## Normalized readiness and fallback
+
+Plan `specified`; role fields/metadata paths `path-confirmed`, payload contents `mapped`, raw arrays `not-covered`; morphology metric `N/A: physical-variable role display`, event efficacy `blocked`; render `draft-diagnostic-only` then `redraw-required`; disposition `fallback-only` until a versioned evaluator exists. Exact role fields are available, but quantitative Vortex products are blocked. Caption ceiling is role identity, event offset and morphology; fallback is target/controlled/physical-zero fields only.
+
+## Nonclaims
+
+No periodic phase; no arbitrary-vortex or arbitrary-flow universality; no unseen amplitude/family/offset generalization; no robustness, stability or optimality; no physical-flow memorylessness; no full observability; no proof that the useful information or cloaking physics occupies a small spatial region; no Kármán SR/CCD transfer; no causal response time; no common score with Illusion or Erase; no claim that existing draft renders are publication-ready.

docs/JFM_WYQ/FIGURE_PLANS/SECTION_07_FIGURE_02_ILLUSION_ERASE_EXTENSIONS.md

@@ -0,0 +1,116 @@
+# Section 7 Figure 2 — Illusion and Erase target extensions
+
+## Argumentative job
+
+Place two changes of target contract side by side without treating them as one score: (A) PPO control toward a declared non-zero Illusion target and (B) PPO control toward an independently acquired clean Erase target in the presence of an upstream disturbance. Illusion retains a named non-zero-target PPO capability; Erase retains role identity and morphology only until a versioned evaluator output exists. 
+## Panel A — Illusion roles and variables
+
+- Physical configuration: uniform-inflow/free-slip-wall channel configuration.
+- Selected result: retained `0.75` target-size PPO demonstration, seed 43.
+- Provenance IDs: `v5/ill_075L` target/zero and `v5/ill_075L_seed43/controlled`; retain internal IDs only in caption/source notes.
+- Roles: non-zero `target`, PPO `controlled`, `physical-zero`.
+- Variables: registered slot-0 vorticity/streamlines or target-error magnitude, plus one compact force/sparse-sensor or DTW/reward readout from the exact retained evaluation if legible.
+- The physical target-size definition must follow Section 2/current environment authority; do not relabel the historical `0.75` variable as a diameter ratio without that authority.
+- No archived surrogate panel, sentence, number, legend item or comparator.
+
+## Panel B — Erase roles and variables
+
+- Physical configuration: parabolic-inflow/no-slip-wall channel configuration with an upstream disturbance cylinder for controlled and physical-zero roles and a clean independently generated target role.
+- Provenance ID: `legacy/erase`; internal label is provenance only.
+- Roles: clean `target` from `late_field.npz`, frozen-PPO `controlled`, and `physical-zero` from `phase_fields.npz`.
+- Variables: current production physical-variable role fields only: target late field, plus controlled and physical-zero retained mean or explicitly labelled eight-snapshot average fields. No target-error magnitude, error comparison or target-error scale is active; all such products remain behind a future versioned evaluator gate.
+- Historical reward summaries remain provenance-only and are omitted from reader-facing efficacy interpretation.
+
+## Exact sources
+
+### Illusion
+
+- Role manifest/navigation: `src/drl_pinball/data/reproduction_plots/manifest.json`, group `ill_075L_seed43`.
+- Fields: `src/drl_pinball/data/reproduction/v5/ill_075L/{target,zero}/phase_fields.npz` and `.../v5/ill_075L_seed43/controlled/phase_fields.npz`.
+- Draft flow renders: `src/drl_pinball/data/reproduction_flow_plots/all/v5/ill_075L/` and corresponding manifest entries.
+- Evaluation/training summaries: `src/drl_pinball/train/results/latest/tables/latest_eval_summary.csv`, `latest_results_summary.json`, and existing wake-L2 outputs under the reproduction root where exact run/role identity matches.
+- Rendering/evaluation: `plot_flow_fields.py`, `plot_reproduction_summary.py`, `eval/wake_l2.py`.
+
+### Erase
+
+- Fields/metadata/summaries: `src/drl_pinball/data/reproduction/legacy/erase/{target,controlled,zero}/`.
+- Target: `target/late_field.npz`; controlled/zero: `phase_fields.npz`, `phase_cycle.npz`, `timeseries.npz`, `erase_summary.json`.
+- Draft renders: `src/drl_pinball/data/reproduction_flow_plots/all/legacy/erase/erase/`.
+- Rendering authority: `plot_flow_fields.py`; acquisition authority: `legacy_test/acquire.py`; reward implementation: `legacy_test/metrics.py` and `legacy_env/legacy_env_erase.py`.
+
+## Layout
+
+Use two clearly separated columns with different local subtitles and metric labels.
+
+1. **Left, Illusion (about 50% width):** target, controlled and physical-zero vorticity or streamlines in a three-panel row; beneath, controlled/zero target-error maps or one exact retained signal comparison. Label `declared non-zero target` prominently.
+2. **Right, Erase (about 50% width):** clean target, controlled and physical-zero mean/average fields in a three-panel row; beneath, physical-variable controlled/zero role fields only pending a versioned evaluator; no target-error maps or scalar pair. Label `clean target; upstream disturbance present in controlled/zero`.
+3. A narrow central divider states `target contract changes`; it must not contain an arrow implying numerical progression or transfer.
+4. No geometry schematic is required from repository output. If desired, the author draws a small target-object versus upstream-disturbance icon.
+
+## Crop, ROI and common-scale rules
+
+- Within each case, register all roles to identical coordinates and crop. Include the controlled/zero bodies and enough downstream wake to interpret the target comparison.
+- Within each role triplet, use one common symmetric physical-variable scale. A sequential target-error scale may be used for Illusion only after the exact retained evaluation is bound; no Erase target-error scale is active.
+- Do **not** force a common physical-variable scale across Illusion and Erase unless units, normalization and legible dynamic range are explicitly verified. Distinct labelled scales are safer because configurations, targets and temporal estimands differ.
+- Any quantitative ROI remains part of a future versioned evaluator contract, not a current display annotation. Do not imply that a rectangular ROI is the observation set or proves local sufficiency.
+
+## Metric semantics
+
+### Illusion
+
+Prefer an already versioned exact acquisition result: complete-cycle mean L2 and eight-snapshot phase L2 remain separate; normalized DTW is a complementary signal metric. Bind target, controlled, physical-zero, seed, window, ROI/mask, normalization and summary path before printing a value. If this binding is not completed, retain the role fields and exact PPO evaluation readout without an independent field scalar.
+
+### Erase
+
+The clean target is nonperiodic and stored as one late field; controlled and physical-zero roles are periodic and store eight selected fields but no `mean_ux/mean_uy`. Therefore three candidate readouts must not be conflated:
+
+1. **Eight-snapshot average-field error:** average each role's eight retained fields, then compare that average with the clean target over the declared ROI.
+2. **Eight-snapshot aggregate error:** RMS across all eight role snapshots and ROI cells relative to the same clean target.
+3. **Stored clean-sensor RMSE:** signal-level summary from `erase_summary.json`, not field L2.
+
+These three non-equivalent readouts remain candidate evaluator-design alternatives only. No provisional scalar, normalization value, recommendation of a winning estimand or efficacy pre-result is retained. A future coordinator-authorized evaluator must select and version one exact contract before quantitative use.
+
+## Historical reward caveat
+
+The old Erase reward computes a same-rollout rolling similarity using an evolving controlled-force/sensor history. That construction does not provide an independently scaled clean target-field efficacy metric. Current acquisition provides separate clean target, controlled and physical-zero field roles. Do not use native reward or reward DTW as field-efficacy authority.
+
+## Candidate evaluator and redraw requirements
+
+### Illusion
+
+1. Select one exact retained seed/acquisition and resolve role metadata/hashes.
+2. Load its existing wake-L2/evaluation summary; verify target size, role window, ROI/mask, phase convention and metric orientation.
+3. Re-render selected fields/errors at publication resolution from arrays.
+4. If no exact scalar binding survives, omit the scalar rather than borrow a historical-chain number.
+
+### Erase
+
+1. Validate target/controlled/zero metadata, field schemas, shapes, normalization authority and source hashes.
+2. If separately authorized, choose one candidate estimand and version its target-relative field and ROI contract; do not assume a reduction result.
+3. Store a versioned JSON/CSV including formula, ROI, snapshot indices/phases, role paths/hashes, geometry guard and the fact that the target is one late clean field.
+4. After that future evaluator is selected, versioned and audited, error maps or scalars may be specified in a future figure revision; they are not part of the current production panels.
+5. Optionally compute a true complete-cycle mean only if retained candidate fields and exact cycle windows are available; this is offline processing, not new CFD.
+
+## Caption interpretation
+
+Open with the changed target contracts, not case names. For Illusion, say that PPO is evaluated against a declared non-zero target under its exact retained acquisition. For Erase, identify independently acquired clean-target/control/zero roles and morphology only. Quantitative efficacy remains blocked until a versioned output passes audit; reward and field estimands remain different objects.
+
+Close with: these are named objective extensions under distinct configurations and estimands; adjacency does not establish transfer, common efficacy or a shared mechanism.
+
+## Readiness
+
+- Illusion target/controlled/zero arrays and draft fields/errors: **available**; final seed-to-summary scalar binding and redraw remain **dependency-gated**.
+- Erase target/controlled/zero arrays, metadata, phase/error drafts and signal summaries: **available and audited for existence/schema**.
+- Stored Erase field L2: **absent**.
+- Erase candidate evaluator inputs: **identified; not a versioned result or execution authorization**.
+- True Erase complete-cycle mean: **not stored in the current published field artifact**; optional candidate-field extraction remains to be audited.
+- Publication role-field composite: **dependency-gated on redraw**; any quantitative display is **blocked** pending a versioned evaluator output.
+- Fallback: retain Illusion and Erase physical-variable role fields with no error maps or scalar annotations.
+
+## Normalized readiness and fallback
+
+Plan `specified`; exact repository roots `path-confirmed`, payloads `mapped`, and any directory lacking printed role leaves is `missing-exact-leaf`; raw arrays `not-covered`. Illusion is bounded seed-43 PPO capability with scalar binding `dependency-gated`; Erase quantitative efficacy is `blocked` although physical role fields are available. Render is `draft-diagnostic-only`/`redraw-required`; disposition `retain-with-gates` for Illusion and `fallback-only` for Erase. Caption ceiling is changed target contracts and morphology, never common efficacy, transfer or positive Illusion SR.
+
+## Nonclaims
+
+no common Illusion/Erase metric; no Legacy/current chain equivalence; no cross-target or cross-configuration transfer; no robustness, universality, optimality or stability; no reward/field equivalence; no causal force-to-field mapping; no full-state matching from sparse signals; no complete-cycle Erase mean unless produced under that exact contract; no quantitative claim from draft PNG appearance alone; no publication-ready status for existing outputs.

docs/JFM_WYQ/FIGURE_PLANS/SECTION_08_STEADY_CONTEXT_AND_THEORY.md

@@ -0,0 +1,133 @@
+# Figure plan — Section 8 steady context and theory boundary
+
+**Status:** `discuss`; textual production specification with Frank’s main equation/composition choice open. Existing images are bounded drafts/redraw sources, not publication-ready panels.  
+**Figure question:** What changes between the stationary, accepted and reverse finite-Re fields; what circulation changes in the prescribed-circulation outer reference; and why does the attractive outer cancellation fail as an explanation of the accepted finite-Re endpoint?  
+**Bounded answer:** The accepted wall actuation accompanies a distinct low-deficit viscous endpoint and reorganized near-wake field. Prescribed circulation strongly reorganizes outer streamlines, but the circulation sign favored by the outer downstream objective is opposite to the descriptive sign measured from the accepted CFD field. The figure visualizes context and a failed closure, not a mechanism.
+
+## Main-text composition: one integrated figure, no action schematic
+
+Use a two-row composition with aligned reading order:
+
+- **Top row — finite-Re evidence:** three compact endpoint columns: stationary reference, accepted $s=3.55$, reverse $s=5.2$. Each column overlays total-velocity streamlines on the corresponding nondimensional CFD vorticity field. A narrow right-hand profile inset plots the full-cross-section vector error $|\boldsymbol q-\boldsymbol q_{in}|/U_\infty$ at $x/D=10$ (or, if clearer in production, $u_x/U_\infty$ with $q_{in}$ shown explicitly) for the same three roles. Visually emphasize accepted; reduce stationary/reverse size or saturation. State that stationary is unsteady and accepted/reverse amplitudes differ.
+- **Bottom row — prescribed-circulation outer reference:** equal-size streamline panels at $g=0$, $g_{opt}=+6.147944$, and fitted $g=-12.200554$, followed by a compact sign/closure box. Use streamline arrowheads and small circulation glyphs on the rear cylinders so the sign is readable directly from the field; these are circulation labels, not actuator-direction arrows. The box states `outer objective: +6.147944`, `accepted CFD descriptive fit: -12.200554`, `opposite sign`, and `wall spin does not prescribe Γ under Euler impermeability`.
+
+This arrangement changes inference in three steps: accepted endpoint distinctness; circulation-dependent outer organization; failed sign/closure. Do not add an integrated evidence map or action-direction schematic.
+
+## Panel contracts
+
+### A–C. Finite-Re endpoint fields
+
+**Variables:** total velocity $(u_x/U_\infty,u_y/U_\infty)$ for streamlines; conservative masked $\omega D/U_\infty$ beneath them.  
+**Roles:** stationary CFD unsteady reference; accepted controlled endpoint $[0,+\Omega,-\Omega]$, $s=3.55$; reverse endpoint, $s=5.2$.  
+**Source arrays:** hash-bound endpoint NPZ files identified in `src/steady_pinball_theory/figures/manifest.json`:
+
+- stationary: `.../steady-contract-v5-torque-d20-20260808/stationary/endpoint.npz`, SHA-256 `c2edc9e...8dde8f3`;
+- accepted: `.../accepted-s3.55/endpoint.npz`, SHA-256 `b6c70b76...9ac72db`;
+- reverse: `.../reverse-s5.2/endpoint.npz`, SHA-256 `760c1e84...2ec3`;
+- comparator $q_{in}$: `.../qin/endpoint.npz`, SHA-256 `88cbeac1...e419b`.
+
+**Existing redraw material:** `figures/fields/streamlines/cfd-*.png`, `figures/fields/vorticity/cfd-*.png`, and PDF/PNG companions under `figures/plates/`; generated by `src/steady_pinball_theory/plot_results.py`.  
+**Crop/coordinates:** source crop $x/D=[-2,10]$, $y/D=[-3.025,2.975]$, front-cylinder origin, x right/y up, array order `yx`, $241\times121$ pixels. For publication, preserve $x/D=[-2,10]$ if the $x/D=10$ profile marker is integrated; otherwise a field crop to $x/D\approx[-1.5,6]$, $|y/D|\le2.5$ is allowed only if the separate profile inset retains the full cross-section and the crop is disclosed. Keep cylinder geometry exact.  
+**Scale:** one shared symmetric CFD vorticity scale for all three fields. Current diagnostic source uses pooled 97.5-percentile $\omega D/U_\infty\in[-2.374694,2.374694]$ with clipping; production may use a reviewed robust limit, but one limit and clipping statement must remain common. Streamline speed is nondimensionalized by $U_\infty$ and should not use color as a second unlabelled quantity.  
+**Role in argument:** establish observed viscous organization and endpoint contrast before theory.  
+**Readiness:** `source-array ready; publication redraw required`. The accepted manifest is settling-qualified to $tU_\infty/D=250$. Existing plates are presentation drafts.
+
+**Caption limits:** stationary is an unsteady reference, not a failed steady endpoint; reverse is a distinct, non-amplitude-matched endpoint. Spatial differences do not establish a causal chain, branch stability, force ownership, momentum restoration or optimum.
+
+### D. Full-cross-section endpoint/profile inset
+
+**Primary variable:** $E_{\infty,\mathrm{vector}}=\max_y|\boldsymbol q-\boldsymbol q_{in}|/U_\infty$ at $x/D=10$ must be printed for accepted: `0.0231672176752314`. If plotting a curve in $y$, use vector error versus $y/D$ from `figures/profiles_long.csv`; if plotting downstream station summaries, use `figures/summary/profile_Einf_vector.csv` and label stations $x/D=2,4,6,10$.  
+**Exact profile source:** `figures/profiles_long.csv` (hash in figure manifest `dc0597bd...f0980a`) and `figures/summary/profile_Einf_vector.csv` (`643720d0...a5d2`).  
+**Current comparator values at $x/D=10$:** stationary `1.0456532678`, accepted `0.0231672177`, reverse `0.6898418412`; do not place potential values on the same quantitative axis unless the families and governing equations are separated explicitly.  
+**Scale:** for a pointwise-$y$ profile, use the complete common-fluid cross-section and dimensionless velocity/error. A broken/log axis is discouraged because it can overdramatize role differences; use an inset/table annotation if one linear axis hides accepted structure.  
+**Role:** convert field appearance into the declared steady endpoint evidence.  
+**Readiness:** `CSV ready; publication redraw required`.
+
+**Nonclaims:** no common estimator with Sections 5–6, no stationary phase average, no optimum/stability, and no potential-versus-CFD metric agreement.
+
+### E–G. Outer-theory streamline sequence
+
+**Variables:** total outer velocity $(u_x/U_\infty,u_y/U_\infty)$ and streamlines for rear-opposite prescribed-circulation basis $(\Gamma_1,\Gamma_2,\Gamma_3)=(0,+\Gamma,-\Gamma)$ under geometric counter-clockwise convention; dimensionless coefficient $g$.  
+**States:** $g=0$; favorable outer value $g_{opt}=+6.1479442395622$; descriptive finite-Re fitted value $g=-12.200554143243473$.  
+**Source arrays/solver:** `steady-strip-modal-analysis-20260804-v4/analysis.json` plus its authenticated `strip_arrays.npz` (SHA-256 `4af103fa...0f03b`); current plotting code reconstructs the 40-image-layer affine strip fields via `strip_basis_affine(...)` and `affine_velocity(...)` in `plot_results.py` lines 109–121. Existing direct assets are:
+
+- `figures/fields/streamlines/mfs-zero.png` and `plates/streamlines/mfs-zero.{png,pdf}`;
+- `figures/fields/streamlines/mfs-gopt.png` and `plates/streamlines/mfs-gopt.{png,pdf}`;
+- `figures/fields/streamlines/mfs-gfit.png` and `plates/streamlines/mfs-gfit.{png,pdf}`.
+
+These are the exact current plots to redraw, not a request for a new MFS computation.  
+**Crop/coordinates:** use the same physical geometry and field extent as A–C where feasible. For circulation legibility, a tighter common crop $x/D\approx[-1.5,5]$, $|y/D|\le2.5$ is preferred, with a small downstream continuation cue; all three use identical crop and streamline density.  
+**Scale/style:** black streamlines on a light neutral background; matched streamline density and arrow size. Add rear-cylinder circulation glyphs keyed to geometric $\Gamma$ only. Do not use vorticity fill: the potential exterior vorticity is analytically zero, and a blank/zero vorticity panel adds no inference.  
+**Role:** show visually that circulation is a prescribed outer degree of freedom and that favorable/fitted signs reorganize the flow oppositely.  
+**Readiness:** `bounded outer-reference display ready for redraw`, not `finite-Re mechanism validated`. Current derivation reports exact-circle and failure controls, bounded unbounded MFS–Laurent agreement, and bounded 20/40/80 finite-image consistency; it explicitly does not claim an asymptotic strip-convergence tail.
+
+**Caption limits:** outer impermeability/free-slip reference only; no rotating no-slip wall trace, separation, wall vorticity, drag, torque, stability, or $\Gamma(\Omega)$ law. Similar streamline organization would not establish mechanism; the actual sign comparison rejects the proposed identification.
+
+### H. Sign-conflict and solution-boundary box
+
+**Content:**
+
+- compact equation label: $\boldsymbol u=\boldsymbol u_N+\sum_j\Gamma_j\boldsymbol h_j$;
+- `Γj prescribed periods; not selected by Ω`;
+- $g_{opt}=+6.147944$ versus $g_{fit}=-12.200554$;
+- `opposite sign → prescribed-circulation cancellation does not explain accepted CFD`;
+- one-circle boundary in small type: $u_\theta(R,\theta)=-2U_\infty\sin\theta+\Gamma/(2\pi R)$ cannot equal constant $R\Omega$ for all $\theta$ when $U_\infty\ne0$.
+
+**Source:** `DERIVATION.md` §§3–6 and current strip/finite-Re analyses.  
+**Role:** prevent visual resemblance from outrunning the mathematical and finite-Re boundary.  
+**Readiness:** `text/equation ready; graphic typesetting required`.
+
+## Physical interpretation supported by the figure
+
+The figure permits this sequence, and no stronger one:
+
+1. The accepted wall law directly moves both rear gap-facing cardinals downstream and both outer cardinals upstream.
+2. Rotating no-slip walls directly prescribe tangential wall velocity only. The endpoint fields exhibit different shear/vorticity, shear-layer, inter-cylinder-transport, recirculation and near-wake organization; links among these observations remain hypotheses.
+3. It is reasonable to hypothesize that these near-field changes alter downstream mean transport and yield the low-deficit profile.
+4. The figure does not isolate the links in step 3. Accepted/reverse differ in amplitude; stationary is unsteady; the partial NS ledger lacks storage/time-consistent and body-adjacent terms and has no residual.
+5. The potential panels explain only how prescribed circulations organize an outer field. The opposite-sign result blocks using that family as the accepted finite-Re wall-spin mechanism.
+
+No caption should use `gap jet` as a causal noun without specifying the measured transport direction/field. No boat-tail wording.
+
+## Caption draft
+
+**Steady finite-Re context and the boundary of a prescribed-circulation explanation.** (a–c) Total-velocity streamlines over conservative masked vorticity for the stationary unsteady reference, the settling-qualified accepted endpoint $[0,+\Omega,-\Omega]$ at $s=3.55$, and the reverse endpoint at $s=5.2$ in the uniform-inflow/free-slip-wall configuration, $Re_D=50$. (d) Full-cross-section comparison with the no-body inflow at $x/D=10$; the accepted endpoint has $E_{\infty,\mathrm{vector}}=0.0231672176752314$. The reverse endpoint differs in both sign and amplitude and is not a controlled same-amplitude intervention. (e–g) Prescribed-circulation outer-reference streamlines for zero circulation, the favorable rear-opposite value $g_{opt}=+6.147944$, and the descriptive finite-Re fit $g=-12.200554$. Circulations are independent periods of the Euler impermeability problem, not wall-spin boundary data. (h) Their opposite signs reject the proposed identification of potential cancellation with the accepted finite-Re state. The panels show endpoint organization and an outer-theory boundary; they establish neither a viscous mechanism, momentum/energy closure, force ownership, stability nor cross-configuration validation.
+
+## Normalized readiness
+
+Plan state `discuss`: Frank's main equation/composition choice remains open. Evidence state: current manifests/reports and existing outer assets were `read/path-confirmed`; external endpoint/strip payloads are `mapped`; new arrays/computation are `not-covered`. Metric state: endpoint/profile and sign comparison are `authority-bound`; viscous mechanism is `N/A: hypothesis not established`. Render state `redraw-required`; the existing outer assets are bounded redraw sources only and **no new MFS computation is authorized**. Disposition `retain-with-gates`; fallback preserves accepted endpoint/profile and the `g_opt` versus fitted-g sign conflict. Caption ceiling preserves the fact of prescribed wall velocity while keeping wall-vorticity, shear-layer and transport links as viscous hypotheses, not established mechanism.
+
+## Production and derivation work
+
+1. Re-render A–C from authenticated NPZ arrays, using the current conservative masked-vorticity implementation and a shared reviewed scale; preserve source hashes in production metadata.
+2. Draw D from `profiles_long.csv` or the station-summary CSV; bind every plotted role and value to the figure manifest.
+3. Re-render E–G from the current 40-layer affine strip reconstruction. This is redraw work only—do not alter the MFS model or generate a new claimed field.
+4. Add consistent geometry, flow-direction key, panel labels, circulation glyphs, $x/D=10$ marker, scale bars and status labels. Use `prescribed-circulation outer reference` above E–G.
+5. Typeset H and the main-text Hodge/period equation consistently. Keep detailed MFS validation, layer/order/source-radius diagnostics, NS budgets, endpoint sweep and vorticity/error galleries in supplement.
+6. Export vector PDF for linework and a high-resolution raster fallback. Record code commit, command, source paths/hashes, crop, masks, scales and any clipping in a production manifest.
+
+## Readiness and fallback
+
+- **Required and ready as source material:** accepted endpoint field/profile; outer $g=0/g_{opt}/g_{fit}$ streamline sources; sign values and solution boundary.
+- **Dependency-gated production:** final visual hierarchy, shared field scales, profile format, panel typography and publication export.
+- **Not required/main-text excluded:** action schematic, MFS validation plots, full NS budget, gap-flux/power bars, sweep, error gallery and potential-vorticity panels.
+- **Fallback if eight-part composition is crowded:** retain accepted CFD vorticity/streamline panel + compact stationary/reverse profile inset; retain all three theory streamline panels + sign box. Drop stationary field first, then reverse field, but never drop the reverse/profile boundary or the $g_{opt}$ versus fitted-$g$ sign conflict.
+
+## Exact production choices for author approval
+
+1. Approve top-row CFD vorticity+streamline triptych with profile inset; safe fallback is accepted field plus three-role profile inset.
+2. Approve bottom-row three-theory-streamline sequence and circulation glyphs; no action schematic.
+3. Approve theory label `prescribed-circulation outer reference (bounded finite-image consistency)`.
+4. Choose profile encoding: recommended pointwise vector error versus $y/D$ at $x/D=10$; alternative downstream-station $E_\infty$ summary only if the pointwise inset is illegible.
+5. Confirm whether stationary/reverse fields survive final page-width trial; accepted field and sign-conflict sequence are non-drop items.
+
+## Compact asset ledger
+
+| State | Asset | Use |
+|---|---|---|
+| `read/verified` | `figures/manifest.json`, `figures/README.md`, endpoint manifest, display README | Coordinates, masks, cases, hashes, rendering provenance and status |
+| `read/verified` | CFD streamline/vorticity raw images and plates; `profiles_long.csv`; profile/near-wake/gap CSVs | Redraw sources and bounded endpoint interpretation |
+| `read/verified` | MFS zero/$g_{opt}$/fitted-$g$ streamline raw images and plates; strip arrays/analysis | Outer-reference redraw and sign comparison |
+| `not publication-ready` | All existing PNG/PDF plates | Draft/source material only; do not paste into final composition unchanged |
+| `supplement` | MFS validation diagnostics, full NS ledgers/budgets, error/vorticity galleries, power/gap bars and sweeps | Reproducibility and limits, not main inference |
+| `not-covered` | New CFD, MFS, budget, stability or causal analysis | No result inferred |

docs/JFM_WYQ/FIGURE_PLANS/SECTION_09_EXPERIMENT.md

@@ -0,0 +1,83 @@
+# Section 9 figure plan — qualitative experiment
+
+## Status and candidate composition
+
+**Status:** textual production plan only; `author-supply/provenance-gated`. No rendering or artifact promotion is authorized. **Candidate composition:** one compact apparatus schematic plus a selected subset of single-frame experimental/CFD pairs: one steady row and zero or more Illusion rows. Do not require every Illusion case. If provenance or legibility fails, omit a row; if all image gates fail, use schematic-only or no figure.
+
+**Figure job:** document the physical setup and let the reader inspect named-condition morphology in separate experimental and CFD views. Juxtaposition means correspondence by author-bound condition only. It is not registration, overlay, experimental validation, CFD agreement or quantitative comparison. A single frame cannot establish shedding frequency or any temporal behavior.
+
+## Layout
+
+Use a two-column grid after a full-width or left-side apparatus panel:
+
+- **(a) Apparatus schematic:** compact facility → carriage/pinball actuation → open-loop command → image-acquisition path.
+- **(b–c) Selected steady row:** experimental single frame | corresponding named-condition CFD single frame.
+- **(d–e), optional:** one selected Illusion row in the same order.
+- **Additional Illusion rows:** include only if each adds a distinct morphology needed by Frank's argument and remains legible. Drop before compressing provenance or claim ceilings.
+
+Keep experimental and CFD labels explicit. Do not share a colour bar, scale bar, coordinates or annotations in a way that implies common calibration. No difference image, overlay, registration marks, metric, frequency annotation, velocity vectors, vorticity equivalence or efficacy score.
+
+## Panel contracts
+
+### Panel (a) — apparatus schematic
+
+- **Role:** establish physical configuration and open-loop execution/acquisition path.
+- **Exact source:** `[AUTHOR SUPPLY: exact schematic path/hash; original editable source; version/date; permission]`. No matching file was located under the inspected `/home/frank14f` paths.
+- **Author fields:** facility/test section and boundaries; towing direction; carriage/rail/rack; pinball geometry/body order; cylinder drive; command source and units; imaging/tracer path; any sensor shown; elements not synchronized or not used for feedback.
+- **Processing:** `[redraw or direct use; dimensions preserved; labels added; changes from original]`.
+- **Caption ceiling:** apparatus context and open-loop signal path only. Do not claim calibrated accuracy, closed-loop operation, synchronized force acquisition or experimental outcome.
+- **Readiness:** `author-supply-gated`.
+- **Fallback:** concise apparatus prose; omit schematic if it becomes a component diary or lacks source permission.
+
+### Panels (b–c) — selected steady experimental/CFD views
+
+- **Role:** show one author-selected named steady morphology comparison.
+- **Experimental source:** `[AUTHOR SUPPLY: exact raw image path/hash]`; `[trial ID/date/setup]`; `[reference or controlled condition]`; `[frame index/time]`; `[camera/view/FOV/spatial scale or uncalibrated]`; `[dye injection]`; `[crop/enhancement]`; `[repeat and selection rule]`; `[permission]`.
+- **CFD source:** `[AUTHOR SUPPLY or audit: exact image path/hash]`; `[CelerisLab/DynamisLab case identity and configuration]`; `[action/body order/sign]`; `[simulation step/frame]`; `[field-to-streakline processing and render settings]`. Candidate generation/history paths exist at `/home/frank14f/CelerisLab/tests/postproc/run_exp_ctrl_matrix_streakline.py`, `/home/frank14f/CelerisLab/tests/postproc/experiment.ipynb`, and `/home/frank14f/CelerisLab/tests/output/{exp_ctrl_matrix_streak_nx1500,stealth_steady_sweep_nx1500}/manifest.json`, but the manifest-referenced PNGs were not present and are not proven to be the author's counterpart.
+- **Layout:** same row, experiment left and CFD right, independent labels/scales; use similar crop only if honestly disclosed, never geometric registration.
+- **Caption ceiling:** `Single-frame dye visualization and separately generated CFD view for the named steady condition; qualitative morphology only.` Frank inserts the exact visible observation after provenance review. No field matching, straightness estimator, velocity/force inference, validation or agreement.
+- **Readiness:** `author-supply-gated`; computational source additionally needs exact case/frame audit.
+- **Fallback:** apparatus-only figure or omit this row. Never reconstruct from a manifest or notebook screenshot.
+
+### Panels (d–e) — selected Illusion experimental/CFD views (optional)
+
+- **Role:** show one selected named Illusion morphology; not all available cases.
+- **Experimental source:** `[AUTHOR SUPPLY: raw path/hash, trial ID, exact Illusion condition, comparator role, frame index/time, camera/view/FOV/scale, dye protocol, processing, repeats/selection, permission]`.
+- **CFD source:** `[AUTHOR SUPPLY or audit: exact path/hash, case/target identity, configuration, action/body mapping, step/frame and streakline/vorticity processing]`. Potential CFD analytic-action definitions and historical outputs exist in `/home/frank14f/CelerisLab/tests/specs/exp_ctrl_matrix.md`, `/home/frank14f/CelerisLab/tests/postproc/run_exp_ctrl_matrix_{vorticity,streakline}.py`, and the corresponding `tests/output/.../manifest.json` records; exact manuscript mapping remains unresolved.
+- **Layout:** experiment left, CFD right; independent acquisition labels. If multiple Illusion rows are retained, each needs its own condition and observation; do not imply an ordered frequency series.
+- **Caption ceiling:** `Single-frame dye visualization and separately generated CFD view for the named Illusion condition; qualitative morphology only.` **Mandatory:** `A single frame does not establish shedding frequency or any temporal change.` No frequency multiplier, phase, waveform recovery, agreement, validation or quantitative efficacy.
+- **Readiness:** `author-supply-gated`.
+- **Fallback:** omit the Illusion row(s); the section remains a short author interface.
+
+## Author production manifest requirements
+
+Before any panel survives, Frank must resolve and approve: experiment configuration and boundaries; trial ID to exact CFD case/frame mapping; flow direction, image orientation and any mirror/rotation; whether the action shown is commanded, converted apparatus input or measured body motion (never conflate them); action/body order/sign/units; exposure, illumination and frame rate, or explicit absence/nonavailability; raw-media path/hash through every crop/enhancement to the rendered output; comparator semantics; apparatus-schematic provenance/editable source; final author crop and caption approval; and publication/reuse rights. Also record trial selection/repeats and whether spatial scale is calibrated or absent. Missing any claim-critical field triggers the strict omission fallback.
+
+## CFD streakline processing disclosure
+
+If a CelerisLab streakline image is selected, disclose the exact release points/mode, injection count, sampled velocity-frame interval, integration (`bilinear space + linear time interpolation + RK4` in the current library), substep settings, diffusion coefficient, particle filtering, age decay, density binning, Gaussian blur, colour/background, simulation step and crop. These fields describe rendering; they do not make the CFD image a calibrated dye model.
+
+If a vorticity image is selected instead, label it `CFD vorticity`, give the plotted range and frame/step, and do not visually equate it with dye intensity. Prefer a streakline-like CFD counterpart when Frank's purpose is morphology, but only from an existing audited artifact or separately authorized future production.
+
+## Caption skeleton — Frank to complete
+
+> **Figure X.** (a) [apparatus and open-loop command/acquisition path; source/version]. (b,c) [steady experimental trial/frame/condition] and a separately generated CFD view for [exact named condition/case/frame/rendering]. (d,e) [optional selected Illusion trial/frame/condition] and its separately generated CFD view. Experimental and CFD images are unregistered, acquisition-specific single-frame views shown only for qualitative morphology. [Exact processing/scale statements.] A single frame does not establish shedding frequency, velocity, force, field matching, quantitative agreement or CFD validation.
+
+## Readiness gate and fallback order
+
+A panel passes only with exact source, case/trial role, condition/action, frame/step, view/scale status, processing, selection/repeats and permission. The caption must preserve open-loop, separate-acquisition and morphology-only limits.
+
+Fallback order:
+1. drop extra Illusion rows;
+2. drop the Illusion row entirely;
+3. retain schematic + steady row only;
+4. retain schematic only;
+5. omit the figure and keep the author-owned textual boundary.
+
+## Normalized readiness
+
+Plan `author-supply-gated`; author-described assets `mapped`, exact experiment/CFD counterparts `missing-exact-leaf`, repository candidate code/manifests `path-confirmed`, raw media `not-covered`; metric `N/A: qualitative morphology only`; render `blocked`; disposition `fallback-only` through schematic-only to complete omission. No publication-ready claim is permitted. Caption ceiling is unregistered single-frame morphology under named open-loop conditions only.
+
+## Current evidence mismatch
+
+The author reports that an apparatus schematic and single-frame experimental/CFD pairs exist, but targeted inspection did not locate those exact files. CelerisLab contains CFD action, streakline and vorticity generation code plus manifests that point to generated computational images; the referenced image/checkpoint files were absent from the inspected output directories. No physical execution/deployment script or experiment-media provenance package was located. Thus no panel is presently publication-ready, and no manifest path may substitute for the author's exact corresponding files.

docs/JFM_WYQ/My_JFM/FLMguide.tex

@@ -0,0 +1,393 @@
+\documentclass[lineno]{JFM-FLM_Au}
+
+\usepackage{amsmath,amssymb,bm,booktabs}
+\newcommand{\dd}{\mathrm{d}}
+\newcommand{\vect}[1]{\boldsymbol{#1}}
+\newcommand{\draftfigure}[3]{%
+  \begin{figure}
+    \centering
+    \fbox{\parbox[c][0.22\textheight][c]{0.88\textwidth}{}}
+    \caption{#2}
+    \label{#3}
+  \end{figure}
+}
+
+\lefttitle{Y. Wang, F. Ren, X. Wang, H. Bai, H. Tang and S. Chen}
+\righttitle{Symbolic control laws for hydrodynamic cloaking and illusion}
+
+\title{Interpretable control laws for active hydrodynamic cloaking and illusion of the fluidic pinball}
+
+\author{Yanqi Wang\aff{1,3}, Feng Ren\aff{2}, Xu Wang\aff{1}, Honglei Bai\aff{4}, Hui Tang\aff{1}, \and Shiyi Chen\aff{3} }
+
+\affiliation{\aff{1}Department of Mechanical Engineering, The Hong Kong Polytechnic University, Hong Kong SAR, PR China
+\aff{2}School of Marine Science and Technology, Northwestern Polytechnical University, Xi'an 710072, PR China
+\aff{3}Eastern Institute for Advanced Study, Eastern Institute of Technology, Ningbo 315200, PR China
+\aff{4}School of Aeronautics and Astronautics, Sun Yat-Sen University, Shenzhen Campus, Shenzhen 518107, PR China}
+
+\corresau{Hui Tang, \email{h.tang@polyu.edu.hk}}
+
+\begin{document}
+\maketitle
+
+\begin{abstract}
+Hydrodynamic cloaking and illusion require active control of the wake transported beyond a body, rather than optimization of an integrated force alone. We investigate these declared-reference objectives using a fluidic pinball of three independently rotating cylinders. A proximal-policy-optimization controller maps the forces on the cylinders and velocities from three downstream probes to rotational commands, while registered velocity fields provide an independent measure of the resulting wake. For periodic K\'arm\'an cloaking, the controlled mean and phase-resolved fields approach the target substantially more closely than the uncontrolled zero-rotation wake. Post-hoc symbolic regression reveals an executable control law organized around persistent counter-rotation of the rear cylinders, supplemented by a smaller time-dependent correction. Closed-loop deletion tests identify the persistent rear rotation as the dominant tested component. A complementary four-flow analysis shows that fixing the cylinders at the mean learned action accounts for 93.42\% of the measured reduction in scalar mean-field error. The remaining time-dependent field organization is examined separately through action-correlated correction-field modes, linking low-rank wake structure to front, rear-symmetric and rear-antisymmetric actuation coordinates. The framework also accommodates an event-relative Vortex case and a non-zero-target Illusion case, demonstrating how the control problem changes with its temporal coordinate and prescribed wake signature. A separate steady-flow analysis connects persistent rear rotation to a low-deficit endpoint and tests its relation to a prescribed-circulation outer flow. Together, these results show how learned rotational control can be evaluated against declared wake references and distilled into a compact, physically readable K\'arm\'an control structure.
+\end{abstract}
+
+\begin{keywords}
+hydrodynamic cloaking, flow control, symbolic regression, deep reinforcement learning, fluidic pinball
+\end{keywords}
+
+\section{Introduction}
+\label{sec:introduction}
+
+A hydrodynamic cloak seeks to make the flow around a body approach a declared background, while a hydrodynamic illusion seeks a declared non-zero substitute signature. Transformation optics established complementary routes to controlling linear fields through material design \citep{PendryEtAl2006Controlling,Leonhardt2006OpticalMapping}. Active source-based cloaking has also been formulated for specified linear equations \citep{Miller2006PerfectActiveCloaking,VasquezEtAl2009ActiveExterior}. In a distinct role, active constructions have retargeted exterior fields towards non-zero illusion signatures \citep{MaEtAl2013ActiveCloaking,LinEtAl2021ActiveAcousticIllusions}. These branches distinguish redirecting a field, generating compensation and prescribing a substitute signature. They assume linear governing relations and prescribed material or source responses, whereas a viscous wake evolves with the actuation and body-generated vorticity; their role here is conceptual rather than predictive. A multilayer hydrodynamic cloak in free flow has been reported \citep{ChenEtAl2024MultilayerHydrodynamicCloak}. Separated finite-Reynolds-number flow nevertheless poses a distinct problem even relative to active hydrodynamic-cloak formulations \citep{UrzhumovSmith2012ActiveCloaks}: nonlinear advection couples forcing to the evolving wake, no-slip bodies generate shear layers and separation, and viscosity and confinement alter the response. The attainable match depends on the declared target and comparison, especially when coherent shedding carries information beyond local measurements. We therefore consider controlled approach to a specified background or substitute wake, assessed beyond the control observations, rather than universal invisibility or exact cancellation in this setting.
+
+The desired outcome is an extended Navier--Stokes field, whereas actuation is confined to compact boundaries. The fluidic pinball---three independently rotating cylinders in channel flow---is a useful model because a small multi-input plant generates rich nonlinear wakes and admits established open- and closed-loop control studies \citep{DengEtAl2020PinballBifurcations,CornejoMacedaEtAl2021PinballControl,RaibaudoEtAl2020PinballControl}. Transient and post-transient pinball-force dynamics have been treated with reduced modelling \citep{DengEtAl2021PinballForceModel}, supporting integrated load as a useful intermediary between local rotation and the wake. Localized actuation can reorganize bluff-body shedding, and deep reinforcement learning has been applied to pinball force control \citep{FengWang2010SyntheticJetCylinder,FengEtAl2023PinballForceControl}. Conventional active flow control commonly asks how actuation changes shedding or integrated load. The present cloak/illusion objective instead adds a separately acquired reference and an extended-field test: neither scalar drag reduction nor force control establishes that the downstream velocity field approaches that reference. The resulting problem is whether three rotation commands can use force-mediated feedback to approach a declared hydrodynamic signature while leaving the downstream field independently testable.
+
+The difficulty increases when the reference contains periodic shedding, an incoming event or a non-zero target wake. Deep reinforcement learning has enabled closed-loop decisions in nonlinear flows under limited observations \citep{RabaultEtAl2019DRLFlowControl,ParisEtAl2021RobustFlowControl}, including rotating-cylinder actuation in towing experiments and simulations \citep{FanEtAl2020BluffBodyRL}. Related bluff-body work has framed learned active control as hydrodynamic stealth \citep{RenEtAl2021HydrodynamicStealthDRL}, although its observer metric is not the target-relative field comparison used here. Unlike a scalar optimization objective, reference following must distinguish what the controller observes from what is later evaluated. The present policy receives the two force components of each cylinder together with sparse downstream velocities. Integrated forces report body--flow interaction, while remote measurements sample the emerging wake; neither implies full observability. Sparse-probe utility is flow- and task-dependent, including sensitivity to Reynolds number and blockage \citep{LiZhang2022ConfinedWakeRL}. This interface motivates three tests: whether the policy realizes its objective, whether the velocity field approaches the target beyond the reward and sensor locations, and whether recurrent action and field structures can be reduced to forms that remain checkable in closed-loop execution.
+
+These tests define a layered analysis. Learned execution is first assessed within each case's own configuration, target roles and temporal coordinate. The periodic K\'arm\'an cloak is then evaluated independently by registering target, controlled and physical-zero velocity fields, while temporal signal comparison remains complementary rather than a substitute for field agreement. On the action side, symbolic regression compresses the policy into a candidate control law tested by closed-loop deployment, following the principle that readable learned feedback should be judged in execution \citep{GautierEtAl2015ClosedLoop,CornejoMacedaEtAl2021PinballControl}. Section~\ref{sec:karman-law} distinguishes this post-hoc policy-map regression from direct genetic-programming control and sparse dynamical-system identification \citep{LiEtAl2017GeneticProgrammingControl,BruntonEtAl2016SINDy}; PySR is the symbolic-regression implementation \citep{Cranmer2023PySR}, not evidence that the resulting expression controls the flow. This distinction prevents a compact expression from being promoted from representation to explanation before deployment. On the field side, the persistent mean response is separated from a centred residual, and action-correlated correction-field modes describe organization without assigning causality. Sparse-to-field and reduced-order precedents motivate separating online information from fuller-state interpretation \citep{LoiseauEtAl2018SparseROM,LiEtAl2022FlowEstimation}. The action and field analyses answer different questions and do not validate one another.
+
+We therefore ask whether force-plus-sparse-observation rotational control can approach declared cloak and illusion references under separately specified channel-flow formulations, and how far the periodic K\'arm\'an cloak can be evaluated and interpreted. Rotational feedback was executed for the periodic K\'arm\'an cloak, the non-zero Illusion target and the event-relative Vortex case; only K\'arm\'an supports a target-relative field-proximity claim, while Illusion and Vortex provide execution and morphology evidence. For Erase, the available target and comparator roles do not yet permit the same performance comparison. Registered target, controlled and physical-zero fields show improved target proximity for the periodic K\'arm\'an cloak in the parabolic-inflow/no-slip-wall channel configuration. Closed-loop action-law tests further identify a K\'arm\'an control skeleton: dominant persistent rear-cylinder rotation accompanied by a smaller dynamic correction, with the associated field organization described by action-correlated modes. This deeper account is possible for K\'arm\'an because policy execution can be followed through registered field comparison, executable law reduction and organized mean and residual fields. The separate steady analysis instead shows that the corresponding inviscid closure fails, and does not extend the learned periodic result. The next section defines the configurations, actuation, observations, target roles and independent evaluation criteria used to keep these comparisons distinct.
+
+\section{Problem formulation, control framework and evaluation}
+\label{sec:formulation}
+
+\subsection{Flow configurations and numerical method}
+\label{subsec:configurations}
+The controlled body is a two-dimensional fluidic pinball comprising three circular cylinders of diameter $D$. A front cylinder is followed by upper and lower rear cylinders, and all three rotate independently. Streamwise and transverse coordinates are $(x,y)$, and the computational body order is $(F,U,L)=(\text{front},\text{rear}_{y+},\text{rear}_{y-})$. Lateral confinement bounds cross-stream motion and provides a repeatable environment in which rotation can alter the integrated forces and wake. Two channel configurations are used for different parts of the study. The parabolic-inflow/no-slip-wall channel supplies the periodic K\'arm\'an field comparison and the detailed action- and field-side analyses. The uniform-inflow/free-slip-wall channel supplies the broader controlled cases and the separate steady analysis. Their results are interpreted within their own physical and numerical settings rather than combined as realizations of one plant. A case-dependent upstream disturbance or target body defines the requested wake, with its dimensions specified for that case.
+
+Velocities are normalized by a configuration-specific reference speed $U_{ref}$, lengths by $D$, and time by $D/U_{ref}$. The article-wide Reynolds number is
+\begin{equation}
+ \begin{aligned}
+  Re_D &= \frac{U_{ref}D}{\nu},\\
+  U_{max} &\simeq 1.5U_{ref}
+  \quad\text{for parabolic inflow}.
+ \end{aligned}
+\end{equation}
+where $\nu$ is the kinematic viscosity. For uniform inflow, $U_{ref}$ is the inlet speed; for parabolic inflow, it is the bulk/mean inlet speed. This distinction matters when comparing inherited case labels: a label based on $2D$ can be halved only if its velocity scale is also $U_{ref}$. Historical parabolic records do not always identify $U_0$ as centreline or bulk/mean speed, so no physical Reynolds number is inferred from such labels here. Both configurations are advanced with a D2Q9 lattice-Boltzmann method using multiple-relaxation-time collision and curved moving-body treatment. Inlet, lateral-wall and outlet implementations follow the stated configuration. The detailed K\'arm\'an analyses use data from the parabolic/no-slip realization, whereas the control formulation described next is implemented in the uniform/free-slip configuration; no numerical equivalence between the two implementations is assumed.
+
+\subsection{Rotational actuation, observations and PPO training}
+\label{subsec:ppo-framework}
+For a radius vector $(r_x,r_y)$ measured from a cylinder centre, angular velocity $\omega$ imposes the wall velocity
+\begin{equation}
+ \begin{aligned}
+  (U_w,V_w) &= (-\omega r_y,\omega r_x),\\
+  s_i &= \frac{\omega_iR}{U_{ref}}.
+ \end{aligned}
+\end{equation}
+where $R=D/2$, positive $\omega$ is counter-clockwise, and $s_i$ is the nondimensional surface-speed command for cylinder $i\in\{F,U,L\}$. The historical symbolic-regression records use $\alpha_i=\omega_i/U_0$ instead. The two descriptions are related by $s_i=\alpha_iR(U_0/U_{ref})$ and coincide as $s_i=\alpha_iR$ only when $U_0=U_{ref}$. Thus $\alpha_i$ is not dimensionless unless a radius-one convention is also specified. In the canonical control formulation, the policy commands rotation without a preset steady-derived bias, and successive commands are smoothed before application. This choice defines the computational interface; it is not a comparison of bias strategies or a map to an experimental actuator. Prior pinball work has treated force dynamics with reduced models and used learned control for force objectives \citep{DengEtAl2021PinballForceModel,FengEtAl2023PinballForceControl}. This motivates integrated force as a compact actuation-coupled signal, without making it a downstream-field measure.
+
+The observation vector contains the two force components of each cylinder and both velocity components from three circular wake probes. The probes are centred $10D$ downstream of the signature source at $(y-y_0)/D=0,\pm2$ and have radius $0.25D$. Each probe contributes area-averaged streamwise and transverse velocity. Forces and wake velocities describe complementary aspects of the controlled flow: periodic loads can carry phase information, while sparse wake measurements provide downstream coordinates \citep{NairEtAl2021PhaseFlowControl,MarraEtAl2024ActuationManifold}. The centre probe is intended to sample coherent-shedding phase, and the off-centre pair to sample lateral deflection and alternating cross-wake motion. Comparable downstream sensing has been used to characterize and control fluidic-pinball wakes \citep{RaibaudoEtAl2020PinballControl,CornejoMacedaEtAl2021PinballControl,LiEtAl2022FlowEstimation}; sparse wake information is also physically accessible in other wake-sensing settings \citep{BeemTriantafyllou2015WakeSensing}. These precedents motivate the measurement classes, not the exact present geometry.
+
+Sensor placement is inseparable from its use and from the time scales carried by the chosen observable. In periodic wakes, a phase-bearing signal can encode dominant shedding dynamics in the regime for which its model is constructed \citep{GongEtAl2020VortexEstimation}. For feedback, however, moving a sensor downstream can increase developed-wake information while also increasing convective lag, so a location that is effective for estimation need not be best for closed-loop control \citep{JinEtAl2022SensorActuatorPlacement}. Sparse instantaneous measurements can be augmented with temporal histories, but such lifting increases input dimension and redundancy \citep{WangEtAl2024DynamicFeatureDRL}. These results motivate retaining spatially distributed and temporally resolved information, but do not choose the present geometry. The three-probe arrangement is consequently a compact wake monitor rather than a claim of optimal placement or full observability. Solver averaging, fixed physical pre-scaling and online observation whitening are applied as distinct operations. Because whitening parameters form part of the policy input map, evaluation uses the normalizer saved with the selected policy.
+
+PPO updates a stochastic policy by maximizing a clipped surrogate objective relative to the preceding policy \citep{SchulmanEtAl2017PPO}. Clipping limits the incentive for a large policy change, but does not guarantee convergence or successful control. The canonical network has two hidden layers of 64 units with sinusoidal activation. Training uses 2048-step rollouts, batches of 64, ten update epochs, discount factor $0.995$ and learning rate $3\times10^{-4}$. These settings specify the reusable canonical training procedure. Reproducing an individual trained policy additionally requires its selected checkpoint, matching normalizer, target signals and run-specific objective settings; the canonical defaults alone do not determine that policy.
+
+\subsection{Target construction and online objective}
+\label{subsec:targets-objective}
+Every comparison distinguishes how its signals and fields are acquired. The target or reference is the wake requested by the case; controlled denotes the trajectory generated with the policy; physical zero or uncontrolled denotes an independently acquired zero-rotation flow; and constant denotes a separately acquired fixed-actuation comparator where one exists. A cloak target is a declared background flow, whereas an illusion target is a separately generated non-zero wake, including a target-body acquisition when required. The target trajectory and the physical-zero trajectory therefore play different roles even when both contain no policy actuation. Target-body size and upstream geometry are specified case by case so that the requested signature is not silently transferred between configurations.
+
+During control, the policy minimizes discrepancies in the same compact information space that it observes. We write the online objective as
+\begin{equation}
+ \begin{aligned}
+  \mathcal{J}_{online}
+  ={}& w_F\,\mathcal{D}_F(\mathbf F,\mathbf F_t)\\
+  &+w_S\,\mathcal{D}_S(\mathbf q,\mathbf q_t).
+ \end{aligned}
+\end{equation}
+where $\mathbf F$ collects the selected body-force components, $\mathbf q$ contains the six probe velocities, and subscript $t$ denotes the corresponding target signal. The force and sensor terms allow rotation to respond both to the body's integrated interaction with the flow and to the developing wake. Their channel aggregation, normalization, orientation and weights are selected for each case. Because this objective samples forces and three wake regions rather than the complete velocity field, reward improvement establishes signal-space execution only. Field matching is tested separately.
+
+\subsection{Independent signal- and field-level evaluation}
+\label{subsec:evaluation}
+Dynamic time warping (DTW) uses dynamic programming to find a minimum-cost monotone path through the pairwise discrepancies of two sampled sequences \citep{SakoeChiba1978DynamicProgramming,Muller2015FundamentalsMusicProcessing}. Declared endpoint, window and step constraints permit limited local time reparameterization without reducing the comparison to a single global phase shift. For each result, we state whether DTW is a distance or higher-is-better similarity, whether channels are native or normalized, and the time window, scaling, aggregation and path constraints. These conventions are project-specific rather than supplied by the generic DTW references. Because excessive warping can hide timing errors, DTW is not interpreted as a physical delay, causal relation, observability test or field-equivalence measure.
+
+Field agreement is evaluated over a registered downstream rectangle,
+\begin{equation}
+ \Omega=\left\{(x,y):
+ \begin{aligned}
+  -6 &\leq \frac{x-x_s}{D} \leq 14,\\
+  \left|\frac{y-y_0}{D}\right| &\leq 5
+ \end{aligned}
+ \right\}.
+\end{equation}
+where $x_s$ is the sensor-plane streamwise location and $y_0$ is the channel centreline. For $r\in\{\text{controlled},\text{physical zero}\}$, the complete-cycle mean-field error is
+\begin{equation}
+ \begin{split}
+  E_{mean,r}=\Biggl[\frac{1}{|\Omega|}\int_{\Omega}
+  &\frac{\|\overline{\mathbf u}_r
+  -\overline{\mathbf u}_t\|_2^2}{U_{ref}^2}\\
+  &\,\mathrm d\Omega\Biggr]^{1/2}.
+ \end{split}
+\end{equation}
+The fixed $20D\times10D$ region is registered to the sensor plane and centreline so that each role is compared over the same downstream extent. Saved fields do not preserve solver-exact masks, however, so the comparison uses a geometry guard rather than claiming an exact common-fluid mask. A comparator-relative reduction is formed only when target, controlled and physical-zero fields share the same acquisition definition, region, weighting, normalization and averaging window.
+
+The same spatial error is evaluated at eight canonical phase slots for the periodic K\'arm\'an case. Each role is phased independently from its own centre-probe transverse velocity, and each slot uses the nearest saved snapshot. The resulting phase diagnostic samples one canonical cycle; it is not a repeated-cycle ensemble or an unrestricted phase optimization. Other cases require different temporal coordinates: Vortex uses offsets from a declared event, Erase admits a mean-field comparison only when its roles are resolved, and the steady analysis uses a late field or profile. These quantities are not pooled because they answer different physical questions. DTW measures temporal signatures close to the controller's information space, whereas mean and phase errors interrogate the extended velocity field. Either can improve without the other, making their separation central to the evaluation.
+
+\subsection{Compact CFD qualification}
+\label{subsec:cfd-qualification}
+The numerical method is assessed through selected observables in two CelerisLab benchmark configurations. In the rotating-cylinder Kan99b K2 case, Strouhal number, mean drag and force-fluctuation amplitudes lie within their prescribed bands. The mean-lift sign differs because the CFD and comparison conventions use opposite force directions, so the overall K2 comparison remains a \emph{partial pass} despite the conventional origin of that sign difference. In the confined-cylinder Sah04 S2 case, the Strouhal number meets its prescribed band. Only these two completed simulations enter the qualification. They support the named frequency and force observables under their benchmark conditions, but do not assess PPO training, target construction, symbolic regression, correction-field interpretation or hydrodynamic cloaking.
+
+With the configurations, controller interface, target roles and independent measures defined, the following sections compare the controlled cases without combining results across configurations or treating signal and field agreement as interchangeable.
+
+\draftfigure{Two configuration schematics showing inlet and wall conditions, fluidic-pinball body order and rotation sign, the case-dependent disturbance or target body, three downstream two-component probes, and the registered evaluation region.}{Planned control configurations and measurements. The parabolic-inflow/no-slip-wall and uniform-inflow/free-slip-wall channels will be shown separately, together with body order $(F,U,L)$, counter-clockwise-positive rotation, and the three circular probes at $x/D=10$, $(y-y_0)/D=0,\pm2$ with radius $0.25D$. The schematic defines geometry, boundaries and observations only; it does not imply configuration equivalence, optimal sensor placement or full observability.}{fig:configurations-sensors}
+
+\section{From steady cloaking to wake retargeting}
+\label{sec:progression}
+
+Rotational control is considered through four comparisons of increasing temporal and target complexity: an approximately steady uniform profile, a periodic K\'arm\'an street, an isolated-vortex event and a non-zero wake target. Because the configuration, clock and measure change between cases, their ordering describes a progression of control tasks rather than a common performance ranking. The steady case asks whether a simple downstream profile can be recovered. K\'arm\'an cloaking then makes the desired wake periodic, the Vortex case replaces phase by an event-relative coordinate, and Illusion changes the desired wake itself.
+
+The steady case uses the uniform-inflow/free-slip-wall channel configuration. At $x/D=10$, the controlled endpoint follows the uniform target more closely than the stationary pinball in both velocity components: the streamwise profile approaches the uniform level and the transverse departure is reduced. This comparison concerns one downstream profile at one endpoint, not a periodic mean or full-field identity. It establishes neither optimality nor stability or mechanism. The actuation magnitude, settling behaviour and physical interpretation are treated with the steady-flow evidence later; here the result provides the simplest instance in which rotation moves measured velocity components towards the target.
+
+Periodic K\'arm\'an cloaking provides the most complete comparison. In the parabolic-inflow/no-slip-wall channel configuration, the target, frozen-policy controlled and physical-zero roles define the alternating-street problem. For both the complete-cycle mean and the separate eight-phase comparison, the controlled downstream velocity field is closer to the target than physical zero, with a dimensionless error reduction of approximately 82\% in each case. The velocity errors are normalized by the simulation speed $U_0$ used in these data. Its relation to article-wide $U_{ref}$ remains unresolved, but no conversion is needed for either relative reduction. Reward and visual resemblance do not establish the field result; Section~\ref{sec:karman-results} separates the mean, phase, signal and action comparisons in detail.
+
+The Vortex case removes periodic phase and instead compares the target, controlled and physical-zero flow morphology at one event-relative coordinate. At that instant the three fields describe how the wake is organized relative to the passing vortex. Without a field-error scalar, one snapshot cannot establish event evolution or improvement over physical zero. The comparison is therefore morphological: it extends the progression from a repeating wake to an isolated interaction while leaving quantitative event response to the cross-case analysis.
+
+Illusion changes the desired field rather than only its temporal coordinate. In the uniform-inflow/free-slip-wall configuration, one acquisition contains the $0.75L_0$ non-zero target, the seed-43 PPO-controlled trajectory and the physical-zero trajectory under the same sampling and phase procedure. The policy was executed toward that non-zero target, and the controlled morphology at the common literal phase is consistent with wake retargeting. Without a scalar field comparison, this morphology does not establish approach to the target or improvement over physical zero. It also neither constructs a target action nor revives the negative Illusion symbolic-regression result.
+
+The progression therefore changes one principal feature at a time: the target evolves from a uniform background to a periodic street, the temporal coordinate changes from phase to an isolated event, and the target finally becomes a different non-zero wake. K\'arm\'an cloaking receives detailed treatment because it combines repeated learning histories with target, controlled and physical-zero fields, sparse downstream signals and applied rotations. This broader set of comparisons does not make K\'arm\'an representative of the other tasks. It instead permits two bounded questions: over which tested K\'arm\'an conditions were high-performing policies found, and how closely did the controlled wake approach its target?
+
+\draftfigure{A K\'arm\'an-dominant progression from the steady profile to periodic target/controlled/physical-zero fields, with one event-relative Vortex comparison and one non-zero-target Illusion comparison.}{Planned progression of controlled wake definitions. The K\'arm\'an comparison will dominate the composition; the Vortex panels will show target, controlled and physical-zero morphology at one stated event offset, and the Illusion panels will show execution toward the retained non-zero target. Scales and acquisition coordinates will be stated within each comparison. Only the K\'arm\'an panels support a target-relative field-error result; the changed-task panels are morphology comparisons without a pooled efficacy measure.}{fig:capability-progression}
+
+\section{K\'arm\'an-wake cloaking}
+\label{sec:karman-results}
+
+\subsection{Learning across tested K\'arm\'an conditions}
+\label{subsec:karman-learning}
+
+The uniform-inflow/free-slip-wall series samples $Re_D=30$, 50, 100 and 200 and disturbance-radius ratios 0.75, 1.0, 1.5 and 2.0. These discrete points delimit the tested conditions; they do not define a continuous Reynolds-number or geometry law. Only the $Re_D=50$ reference condition has five stochastic training realizations, whereas each other condition has one deterministic policy demonstration. The two outer disturbance-radius cases also used a learning rate of $10^{-4}$ instead of $3\times10^{-4}$, so their differences cannot be attributed to geometry alone. Reward and six-sensor dynamic-time-warping (DTW) similarity are reported as separate, higher-is-better quantities: reward measures the learned objective, whereas DTW measures similarity of the selected downstream signals. Neither is treated as a fitted trend or combined score.
+
+The condition series compares one deterministic policy at each tested condition, whereas the learning histories show how five stochastic searches developed at the reference condition. They are not paired before-and-after observations. At $Re_D=50$, all five 500-iteration histories progress from low reward to high-performing checkpoints. Their best rewards are 0.9266, 0.9302, 0.9160, 0.9222 and 0.9412, reached at iterations 471, 315, 360, 465 and 447, respectively. High-performance policy discovery therefore recurred in all five tested realizations rather than depending on a single selected run.
+
+The recurrence applies to these five searches only: it provides no uncertainty for other seeds or hyperparameters and establishes neither asymptotic convergence nor closed-loop stability. The deterministic policy used for the condition comparison is one deployment drawn from this set, not a sixth realization or a continuation of training reward. Moreover, high reward establishes success under the online objective, which combines force and sparse downstream-signal terms; it does not by itself establish proximity of the downstream velocity field.
+
+That distinction also separates the condition survey from the field analysis below. The survey uses the uniform-inflow/free-slip-wall configuration, whereas the detailed reference wake uses a parabolic-inflow/no-slip-wall configuration. Their numerical case labels therefore do not supply a common physical Reynolds number. The field comparison instead asks directly whether the controlled wake approaches its target in the velocity field and downstream signals.
+
+\subsection{Fields and trajectories in the reference wake}
+\label{subsec:karman-fields}
+
+In the parabolic-inflow/no-slip-wall reference wake, the target, frozen-PPO controlled and physical-zero roles define the field comparison. The controlled wake reproduces the target's alternating organization more closely than the physical-zero pinball. These roles are separate trajectories in the same physical configuration, not paired realizations of one trajectory. Section~\ref{sec:karman-law} uses these same target and physical-zero fields to compare symbolic-regression control, but its controlled trajectory is different from the PPO-controlled trajectory here. Coincident target and physical-zero values therefore come from shared comparator fields, not independent agreement between the two controllers.
+
+The complete-cycle mean tests the persistent velocity organization after periodic variation has been averaged out. With velocity normalized by the simulation speed $U_0$ used in these data, its error is $E_{mean}^{(U_0)}=0.085771$ for the controlled wake and $0.475206$ for physical zero, giving a zero-relative reduction of 0.819508. Thus the mean controlled field lies substantially closer to the target over the registered $20D\times10D$ downstream region. The result applies to this downstream region, which lies beyond the solid geometry; without a saved solver-exact fluid mask it is not a claim of pointwise identity throughout the domain.
+
+The eight-slot diagnostic retains periodic organization instead of averaging it away. It gives $E_{phase8}^{(U_0)}=0.112680$ for controlled and $0.630878$ for physical zero, a zero-relative reduction of 0.821392. Each slot is the nearest saved snapshot after the target, controlled and physical-zero trajectories have been phased independently; the slots are not repeated-cycle ensembles and no global phase minimization is applied. The mean and phase comparisons give the same target-relative ordering but are not replicate measurements or uncertainty estimates. Both use $U_0$, so their relative reductions are dimensionless even though the relation between $U_0$ and article-wide $U_{ref}$ remains unresolved.
+
+Sparse downstream observations provide a complementary trajectory-level comparison. Target-normalized six-channel DTW similarity is 0.932899 for controlled and 0.625127 for physical zero; unlike spatial error, a larger value denotes closer signal evolution. DTW does not imply full-field equivalence, but its ordering is consistent with the smaller controlled spatial errors. Over samples 96--145, corresponding to $tU_0/D=38.4$--58.0 in the simulation scaling and approximately three target cycles, the centre-sensor trajectory shows the repeating downstream response while the three applied PPO rotations show the simultaneous periodic actuation. The target has no action trajectory, so no target-action surrogate is inferred.
+
+Together, the condition survey and learning histories show repeated policy discovery over the tested training set, while the reference-wake analysis shows target-relative proximity in mean fields, phase-resolved fields and sparse signals. These results imply neither mechanism nor a universal optimal law. Similar numbers in the later symbolic-regression or correction-field analyses do not provide validation because those calculations use different controlled trajectories and measures. The remaining control-side question is whether the periodic action produced by the neural policy can be represented by a compact executable law.
+
+\draftfigure{Tested K\'arm\'an condition points, five reference-condition learning histories, and the parabolic/no-slip target, PPO-controlled and physical-zero field and signal comparisons.}{Planned K\'arm\'an learning and field evidence. Separate panels will report deterministic reward and six-sensor DTW similarity over the finite uniform/free-slip condition series, the five stochastic learning histories at $Re_D=50$, and target-relative mean, phase and sparse-signal comparisons for the distinct parabolic/no-slip reference wake. The configurations and estimands will remain separate; points do not define a continuous parameter law, and reward or DTW does not establish field agreement.}{fig:karman-learning-fields}
+
+\section{A compact control law for K\'arm\'an cloaking}
+\label{sec:karman-law}
+
+The compressed action has a simple physical organization: the rear cylinders maintain persistent counter-rotation, rear-lift feedback modulates this background motion, and the front cylinder receives a weaker drag-asymmetry correction.  The question is whether this organization can replace the PPO policy with an executable observation-to-action surrogate for the same periodic K\'arm\'an cloak in the parabolic-inflow/no-slip-wall channel configuration.
+
+Direct symbolic controller search evaluates explicit feedback expressions in the control loop, as in genetic-programming studies of separation and vibration control \citep{GautierEtAl2015ClosedLoop,DebienEtAl2016GeneticProgrammingRamp,RenEtAl2019MachineLearningVIV}.  Comparisons of compact genetic-programming laws and neural policies also separate readability from executed behaviour \citep{CastellanosEtAl2022FewSensorMLControl}.  By contrast, sparse identification of nonlinear dynamics (SINDy) identifies parsimonious dynamical equations from data \citep{BruntonEtAl2016SINDy}, while SINDy-RL learns sparse dynamics and reward models jointly with a policy \citep{ZolmanEtAl2025SINDyRL}.  Both address different mathematical objects from a fixed-policy map for the present controlled wake problem.
+
+Here symbolic regression was applied after PPO training to recorded state--action trajectories, making the expression a fixed-policy map rather than a directly searched controller, identified wake equation or jointly learned reinforcement-learning model.  PySR supplies the search method and software identity \citep{Cranmer2023PySR}, not evidence of control.  Each post-step state at index $i$ was paired with the PPO action at $i+1$.  Agreement on PPO-visited states is an offline diagnostic; after deployment, the surrogate chooses its own actions and visits a different trajectory.  Closed-loop execution against the target is therefore the relevant test, and the cited methods do not validate the present map or its performance.
+
+Before specifying the coefficients, we impose a reflection organization consistent with the centreline-symmetric geometry and cloaking objective.  For a measured state $x$, $Gx$ denotes its centreline-reflected state: upper and lower sensor values are exchanged, streamwise velocity is unchanged, transverse velocity changes sign, front drag is unchanged, front lift changes sign, and the upper and lower force pairs are exchanged with drag even and lift odd.  For actions ordered as front, upper rear and lower rear, the corresponding action map is
+\begin{equation}
+ \begin{aligned}
+  G_\alpha(\alpha_F,\alpha_U,\alpha_L)
+  &=(-\alpha_F,-\alpha_L,-\alpha_U),\\
+  \alpha_F(x)&=\frac{h_F(x)-h_F(Gx)}{2},\\
+  \alpha_U(x)&=h_R(x),\\
+  \alpha_L(x)&=-h_R(Gx).
+ \end{aligned}
+\end{equation}
+The front action is odd, while the rear actions share one mirrored generator.  This construction replaces three separately parameterized heads with an odd front head and one rear generator.  The reflection map was imposed on the surrogate after the PPO trajectories had been collected; it is not evidence that the PPO policy or controlled plant is reflection-equivariant.
+
+The law uses signed native solver forces in the streamwise and transverse coordinate directions, so positive lift is the positive-$y$ force.  For cylinder $i$, the fitted inputs are
+\begin{equation*}
+ \begin{aligned}
+  C_{d,i}&=\frac{2f_{x,i}}{\rho U_0^2D}, &
+  C_{l,i}&=\frac{2f_{y,i}}{\rho U_0^2D},
+ \end{aligned}
+\end{equation*}
+with $\rho=1$ in the retained lattice data.  Thus
+\begin{equation*}
+ \begin{aligned}
+  C_{d,\mathrm{rear},a}&=\frac{C_{d,U}-C_{d,L}}{2}, &
+  C_{l,\mathrm{rear},s}&=\frac{C_{l,U}+C_{l,L}}{2}.
+ \end{aligned}
+\end{equation*}
+Within this imposed action space, the executable map is
+\begin{equation}
+ \begin{aligned}
+  \alpha_F&=\operatorname{odd}
+  (-0.381391\,C_{d,\mathrm{rear},a}),\\
+  \alpha_U&=1.307782\,C_{l,\mathrm{rear},s}-3.431209,\\
+  \alpha_L&=-\alpha_U(Gx).
+ \end{aligned}
+\end{equation}
+where $\operatorname{odd}$ denotes $f(x)\mapsto[f(x)-f(Gx)]/2$.  The constant $-3.431209$ sets the persistent upper-rear rotation and, through reflection, the opposite lower-rear rotation.  Rear-mean lift adjusts this pair, while rear drag asymmetry supplies the front correction.  These terms describe the surrogate action, not a governing law for the fluid.
+
+The historical output is $\alpha=\omega/U_0$, and deployment applies $\omega=\alpha U_0$; hence $\alpha$ has inverse-length units in dimensional notation.  The corresponding surface-speed command is $s=\omega R/U_{ref}=\alpha R(U_0/U_{ref})$.  It cannot be reduced to $s=\alpha R$ because the historical parabolic source does not establish $U_0=U_{ref}$.
+
+Closed-loop execution tests whether the compact action remains useful away from PPO-visited states.  This deployment test is also the selection boundary: expression simplicity and offline action fit can nominate a readable map, but only the executed surrogate determines the trajectory on which its control performance is measured.  No external controller-search or sparse-identification result supplies that evidence for this case.
+
+In the reference run, 480 control intervals were used to establish the trajectory, followed by 160 recorded post-step boundaries spanning eight complete cycles.  The mean native dynamic-time-warping (DTW) similarity was $0.942357$, where higher is better.  This rolling comparison uses a 30-boundary window: one circular lag is applied before the six channel scores are averaged.  The lag aligns signals for comparison and is not a physical delay.
+
+The same run also supplies two spatial comparisons with the same-case target and physical-zero trajectories from the shared acquisition lineage.  Both errors are two-component velocity RMS differences over the geometry-guarded downstream region, normalized by the historical inlet scale $U_0$: $E_{\mathrm{mean}}$ compares complete-cycle arithmetic mean fields, whereas $E_{\mathrm{phase8}}$ is the RMS over eight independently phase-referenced nearest-boundary snapshots.  The values were $E_{\mathrm{mean}}=0.125404$ for SR and $0.475206$ for physical zero, and $E_{\mathrm{phase8}}=0.156040$ for SR and $0.630878$ for physical zero.  The surrogate-controlled wake is therefore closer to the target under both measures.  Each phase entry is the nearest saved snapshot on a phase coordinate determined separately for that flow, and the region has no stored solver-exact mask.  These SR-run values answer a different question from the independently recorded PPO-controlled field errors in Section~\ref{sec:karman-results} and are not repeat measurements of them.
+
+Term deletion asks which parts of the executed map matter most among the tested variants.  In each of four source cases, the full map and a deletion variant were run for the same 40 control steps.  Six-channel native DTW similarity was evaluated after one shared circular-lag alignment, and $\Delta S=S_{deleted}-S_{parent}$.  Deleting the rear constant gave $-0.106537$, $-0.093869$, $-0.178525$ and $-0.191082$; deleting rear-lift feedback gave $-0.028755$, $-0.041660$, $-0.061363$ and $-0.048948$; deleting the front correction gave $-0.006520$, $+0.000579$, $-0.005179$ and $-0.025194$.
+
+Removing the persistent rear rotation produces the largest degradation in every tested case.  Removing rear-lift feedback has a smaller but non-zero effect, while removing the front correction is weak or mixed over this window.  A deletion changes the actions and hence the states subsequently visited, so this ranking does not identify causal or necessary terms in the flow-control process.  It establishes a tested action hierarchy and motivates a separate field question: how do the persistent reference action and the remaining time-dependent organization appear in the mean and centred velocity fields?
+
+\draftfigure{Closed-loop symbolic-controller action traces and target/SR/physical-zero wake fields, accompanied by a fixed-window parent-relative term-deletion plot.}{Planned compact K\'arm\'an law and deployment evidence in the parabolic-inflow/no-slip-wall configuration. The action traces will show persistent opposite rear rotation and smaller modulation; the field panels will compare the exact target, SR-controlled and physical-zero roles; and deletion will report $\Delta S=S_{deleted}-S_{parent}$ over the stated 40-step window. The deletion ranks tested action terms but does not establish causality, necessity, stability or transfer.}{fig:karman-sr-law}
+
+\section{Mean-field correction and action-correlated organization}
+\label{sec:field-ccd}
+
+The field analysis begins with four flows: the target $T$, physical zero $0$, constant control $C$ fixed at the mean action of the PPO policy, and PPO-controlled flow $D$.  For the periodic K\'arm\'an cloak in the parabolic-inflow/no-slip-wall configuration, each mean averages the same 360 post-step boundaries, indices $[480,840)$.  The comparison covers $34\leq x/D\leq54$, $|y/D|\leq5$, on the 80,200 fluid points shared by all four simulations.  With coordinate weights $w_k$ and both velocity components, the target error of flow $r$ is
+\begin{equation}
+ E_r=\left[
+ \frac{\displaystyle\sum_{k\in K}w_k
+ \|\overline{\boldsymbol{u}}_r(k)
+ -\overline{\boldsymbol{u}}_T(k)\|_2^2}
+ {\displaystyle\sum_{k\in K}w_k}
+ \right]^{1/2}.
+\end{equation}
+Here $\boldsymbol{u}_r(k)=(u_{x,r}(k),u_{y,r}(k))$ is the two-component velocity at point $k$, and $\overline{\boldsymbol{u}}_T$ is the target mean, not an instantaneous field.  This four-flow comparison uses different data, averaging windows, masks and weighting from the PPO-controlled comparison in Section~\ref{sec:karman-results} and the SR comparison in Section~\ref{sec:karman-law}; the resulting numbers are not repeat measurements.
+
+The streamwise means give the first physical picture.  Across much of the wake region, $\overline u_{x,0}-\overline u_{x,T}$ is negative: the physical-zero pinball has a mean streamwise deficit relative to the target.  The difference $\overline u_{x,D}-\overline u_{x,0}$ is largely opposite in sign and shows where PPO control adds or removes streamwise velocity relative to physical zero.  This pattern is compatible with adding downstream velocity where the physical-zero flow produces a deficit, but it does not establish momentum restoration.  These fields show only the streamwise component; the errors below use both velocity components.
+
+The exact errors are $E_0=0.4694352273$, $E_C=0.1110140739$ and $E_D=0.0857787860$.  Their ordered differences give
+\begin{equation}
+ \frac{E_0-E_C}{E_0-E_D}=93.42\%,\qquad
+ \frac{E_C-E_D}{E_0-E_D}=6.58\%.
+\end{equation}
+Thus constant control accounts for $93.42\%$ of the measured reduction from the physical-zero error to the PPO-controlled error, and the change from constant control to PPO control accounts for the remaining $6.58\%$.  These percentages divide changes in one scalar distance from the target mean.  They are neither energy fractions nor measurements of the norm of the residual defined next, and they do not assign shares to terms in the SR law.
+
+The time-dependent comparison requires a velocity field rather than a difference between scalar errors.  Mean-field descriptions of natural and actuated cylinder wakes use shift modes to represent changing means and their coupling with fluctuations \citep{TadmorEtAl2010CylinderMeanField}.  That context motivates keeping the mean correction and fluctuation organization as distinct objects here; it does not identify the present constant-control subtraction with a shift-mode model.  The PPO-controlled run contains 19 cycles resolved into ten phase bins.  A separate 19-cycle constant-control run supplies the two-component phase field $\widehat{\boldsymbol{u}}^C_b(k)$.  After subtracting each flow's own mean, the centred velocity residual is
+\begin{equation}
+ \begin{aligned}
+  \boldsymbol{u}^{\mathrm{res}}_{c,b}(k)
+  ={}&[\boldsymbol{u}^D_{c,b}(k)-\overline{\boldsymbol{u}}_D(k)]\\
+  &-[\widehat{\boldsymbol{u}}^C_b(k)-\overline{\boldsymbol{u}}_C(k)],\\
+  &c=0,\ldots,18,\qquad b=0,\ldots,9.
+ \end{aligned}
+\end{equation}
+Every PPO-controlled cycle is compared with the constant-control ensemble template at phase bin $b$; individual cycles from the two runs are not paired.  No cross-run lag, phase wrapping or interpolation is introduced.  The residual is therefore a separately centred comparison, not a matched counterfactual response, and its magnitude is not the 6.58\% scalar share.
+
+Canonical correlation decomposition (CCD) asks whether this residual varies with the executed actions.  The actions are the same-boundary, counter-clockwise-positive cylinder speeds in native solver units.  From front, upper and lower actions $(a_F,a_U,a_L)$, the analysis forms $a_F$, the rear-symmetric coordinate $(a_U+a_L)/\sqrt{2}$ and the rear-antisymmetric coordinate $(a_U-a_L)/\sqrt{2}$.  It removes the measured mean action and then centres the samples, without rescaling or whitening them.  A rank-3 decomposition relates these three action coordinates to the centred velocity residual.  Its field vectors are termed \emph{action-correlated correction-field modes}.  Their singular strengths measure cross-correlation, not field energy, and each mode may be multiplied by $-1$ without changing the decomposition.  The modes describe action association rather than a causal response of the flow.  This placement has a broader correlation-oriented ancestry: extended POD relates fields to correlated events, observable-ranked decompositions order velocity structures by their linear observability in another signal, and point--field correlations have been used for descriptive field reconstruction \citep{Boree2003ExtendedPOD,JordanEtAl2007ObservableJetModes,DiscettiEtAl2018CorrelatedFieldEstimation}.  Sparse sensor-to-field modelling provides a related observable-associated context \citep{LoiseauEtAl2018SparseROM}.  These precedents motivate the distinction from energy ranking; they do not validate CCD, import another decomposition, or establish causality.
+
+Proper orthogonal decomposition (POD) provides the field-only comparison \citep{SchlegelEtAl2012LeastOrder}.  It orders velocity-field variance under the same weights without using the action coordinates.  Here POD and CCD use exactly the same centred residual snapshots, rank, weights and spatial region.  Their rank-3 field subspaces are essentially the same.  CCD therefore does not provide a superior or uniquely physical velocity basis; it adds coordinates that associate the shared low-rank field organization with front, rear-symmetric and rear-antisymmetric motion.
+
+Together, the mean fields show that constant rear-cylinder rotation accounts for most of the measured scalar error reduction.  A separately centred velocity residual has descriptive action-correlated organization.  This relationship does not identify SR terms with CCD coordinates or independently confirm the action analysis.  It applies only to this K\'arm\'an cloak, these four flows and the stated spatial and temporal comparison.
+
+\draftfigure{Four role means and signed streamwise differences with the scalar error ledger, followed by the separately centred residual construction and representative action-correlated phase organization.}{Planned mean-field and correction-field analysis for the periodic K\'arm\'an cloak. Target, physical-zero, constant and PPO-controlled means will precede the two-component scalar error accounting, for which constant control accounts for 93.42\% of the measured zero-to-PPO reduction. Separate panels will define the non-paired centred residual and its action-correlated correction-field modes. The scalar shares are not residual norms or energy fractions, and the modal association is descriptive and noncausal.}{fig:mean-field-ccd}
+
+\section{Extensions under changed temporal and target contracts}
+\label{sec:extensions}
+
+The periodic K\'arm\'an case provides a reference for changing the control problem, not a numerical or interpretive template for the cases considered here.  The incoming vortex replaces periodic phase by event-relative time; Illusion replaces background matching by a declared non-zero target; and Erase compares with a clean-flow target while an upstream disturbance remains in the PPO-controlled and physical-zero flows.  The policy maps its current force-plus-sparse-observation state to cylinder actions in each case, but the clocks, target fields and physical comparisons differ.  Field morphology can therefore be interpreted only within each case.
+
+In the central Taylor-vortex scene, the target, PPO-controlled and physical-zero flows are sampled at offsets relative to the detected incoming event, together with the action history.  The offset indexes the event, not a periodic phase or a calibrated physical time.  At offset $+10$, the three vorticity fields show distinct event-relative morphologies.  Their common event coordinate places the incoming structure, its passage around the pinball and the downstream wake in one spatial comparison, so differences can be assigned to role without treating the offset as a phase.  No scalar comparison is reported; only morphology is considered.  Because the policy acts from the current observation state, the event motivates a hypothesis of reactive control from local information for a comparable-scale disturbance.  This single scene demonstrates neither control across vortex families, amplitudes or offsets nor observability or physical-flow memorylessness.
+
+Illusion changes the desired wake signature rather than the clock.  In the uniform-inflow/free-slip-wall configuration, the retained $0.75$ target-size, seed-43 policy was executed toward a declared non-zero target.  Unlike background matching, that target contains a wake pattern, and the PPO-controlled morphology is consistent with retargeting toward it.  No scalar change in target-relative error is reported.  The historical negative Illusion symbolic-regression result remains specific to that method: this PPO execution supplies neither a positive Illusion symbolic law nor a transfer of the K\'arm\'an surrogate.
+
+Erase changes both the physical scene and the temporal comparison.  In the parabolic-inflow/no-slip-wall configuration, its clean target is one late field, whereas the PPO-controlled and physical-zero flows contain the upstream disturbance and contribute eight phase-referenced fields.  The clean, PPO-controlled and physical-zero morphologies are visible, but one late target and an eight-snapshot set are not the same temporal object.  The historical rolling reward compares an evolving same-rollout history rather than the clean target field.  No scalar field evaluation is reported; averaging the retained snapshots would first remove phase-dependent variation, whereas aggregating snapshot-wise errors would retain that variation in the error.  These operations define different estimands, neither of which is used here.
+
+Together, the fields show PPO execution under changed temporal or target definitions.  For Illusion, the bounded result is execution toward a declared non-zero target with morphology consistent with retargeting; Vortex and Erase remain morphology-only comparisons, and reactive-control or target-erasure interpretations remain hypotheses.  The distinct configurations and temporal comparisons admit neither a common efficacy measure nor transfer of the K\'arm\'an SR/CCD interpretation.  A separately obtained steady endpoint can therefore provide only non-equivalent physical context.
+
+\draftfigure{Target, PPO-controlled and physical-zero fields for one stated Vortex event offset and the retained non-zero-target Illusion acquisition, with acquisition-specific coordinates and scales.}{Planned changed-task morphology. Vortex fields will be compared at a stated event-relative offset, never a periodic phase, and Illusion fields will show policy execution toward the declared non-zero target. These panels describe role identity and morphology only: no scalar efficacy, K\'arm\'an-law transfer, common metric or causal response is inferred.}{fig:changed-task-morphology}
+
+\section{Steady control as physical context and a theory boundary}
+\label{sec:steady-theory}
+
+Steady rear-cylinder rotation was the historical discovery seed, and the periodic K\'arm\'an analysis later identified persistent rear counter-rotation as the dominant tested symbolic-regression term, with a smaller dynamic correction.  Those findings belong to the parabolic-inflow/no-slip-wall configuration.  Here a separately acquired uniform-inflow/free-slip-wall endpoint at $Re_D=50$ supplies a distinct physical comparison.  It is not a continuation, amplitude match, replication or validation of the periodic control chain, and no cross-configuration arithmetic is meaningful.
+
+With body order $(\text{front},\text{rear}_{y+},\text{rear}_{y-})$, the accepted action is $[0,+\Omega,-\Omega]$ and the surface-speed ratio is $s=\Omega R/U_\infty$.  At $x/D=10$, define the two-component profile error by
+\begin{equation*}
+ \begin{aligned}
+  E_{\infty,\mathrm{vector}}
+  &(\boldsymbol u_{\mathrm{ctl}},\boldsymbol u_{\mathrm{in}})\\
+  &=\frac{1}{U_\infty}\max_{y\,\in\,\mathcal C}
+  \left|\boldsymbol u_{\mathrm{ctl}}(y)
+  -\boldsymbol u_{\mathrm{in}}(y)\right|,
+ \end{aligned}
+\end{equation*}
+where $\mathcal C$ is the common-fluid cross-section and $|\cdot|$ is the Euclidean norm of the streamwise and transverse velocity differences.  Thus this $L^\infty$ profile metric uses a cross-sectional maximum, not quadrature weighting.  The settling-qualified endpoint at $s=3.55$ has $E_{\infty,\mathrm{vector}}=0.02317$.  Stationary flow is an unsteady reference, while the reverse endpoint at $s=5.2$ differs in both sign and amplitude.  The accepted velocity field is therefore a distinct low-deficit endpoint, not an optimum, a stable branch or a universal law.
+
+The wall law $(U_w,V_w)=(-\omega r_y,\omega r_x)$ fixes the tangential surface velocity: for the accepted rear-cylinder signs, both gap-facing cardinal points move downstream and both outer cardinal points move upstream.  Relative to the stationary and reverse flows, the accepted vorticity and streamline fields show a different inter-cylinder and near-wake organization.  These endpoint observations motivate a viscous hypothesis: imposed wall motion may alter wall shear and vorticity production, shear-layer development, inter-cylinder transport, separation and recirculation, and hence downstream mean transport.  The sequence is not a demonstrated mechanism because no amplitude-matched intervention, complete same-control-volume momentum or mechanical-energy budget, force attribution, or stability analysis isolates its links.
+
+Rotating-cylinder studies make this separation necessary.  Wake modification by auxiliary rotating cylinders varies with Reynolds number, gap, rotation rate and dimensionality \citep{Mittal2001RotatingControlCylinders,ChanEtAl2011CounterRotatingCylinders}; even a strongly suppressed viscous wake that resembles a potential doublet does not thereby become an inviscid solution \citep{ChanEtAl2011CounterRotatingCylinders}.  Viscous rotating-cylinder flows can also occupy multiple steady or cyclic branches, including bistable regimes \citep{SierraEtAl2020RotatingCylinderBifurcations}, while low-Reynolds-number two-cylinder analysis requires matched inner Stokes and outer Oseen descriptions \citep{Watson1996TwoRotatingCylinders}.  These results motivate an outer reference as a deliberately restricted test of streamline organization, not as a parameter transfer, branch identification or mechanism for the present endpoint.
+
+The rearranged outer streamlines raise a specific closure question: can the finite-Re organization be explained by the circulation degrees of freedom that remain after an inviscid flow satisfies impermeability?  To test that proposition without assigning viscous physics to the outer model, consider the irrotational family \citep{Crowdy2006MultipleCylinders}
+\begin{equation}
+ \begin{aligned}
+  \boldsymbol{u}(\boldsymbol{x})
+  &=\boldsymbol{u}_{N}(\boldsymbol{x})
+  +\sum_{j=1}^{3}\Gamma_j\boldsymbol{h}_j(\boldsymbol{x}),\\
+  \nabla\!\cdot\boldsymbol{h}_j
+  &=\nabla\!\times\boldsymbol{h}_j=0,
+  \qquad \boldsymbol{h}_j\!\cdot\boldsymbol{n}=0,\\
+  \oint_{B_k}\boldsymbol{h}_j\!\cdot\mathrm{d}\boldsymbol{\ell}
+  &=\delta_{jk}.
+ \end{aligned}
+\end{equation}
+Here $\boldsymbol{u}_N$ is the zero-period Neumann solution and the counter-clockwise circulations $\Gamma_j$ are prescribed periods spanning the harmonic freedom left by impermeability; cylinder spin does not select them.  This freedom reorganizes the outer streamlines, but it cannot impose rotating no slip; viscous rotating-cylinder theory retains that tangential boundary condition under its own low-Reynolds-number assumptions \citep{UedaEtAl2003RotatingCylinders}.  For one cylinder,
+\begin{equation}
+ \begin{aligned}
+  u_\theta(R,\theta)
+  &=-2U_\infty\sin\theta+\frac{\Gamma}{2\pi R}\\
+  &\neq R\Omega
+  \quad\hbox{for all }\theta\hbox{ when }U_\infty\neq0.
+ \end{aligned}
+\end{equation}
+For the normalized rear-opposite basis, define the dimensionless coefficient $g$ by
+\begin{equation*}
+ \frac{(\Gamma_1,\Gamma_2,\Gamma_3)}{U_\infty D}
+ =\frac{g(0,1,-1)}{\sqrt{2}},
+\end{equation*}
+with positive circulation geometrically counter-clockwise.  The outer strip objective selects $g_{\mathrm{opt}}=6.148$, whereas the descriptive projection of the accepted finite-Re endpoint gives $g_{\mathrm{fit}}=-12.20$.  The no-slip incompatibility and opposite sign reject wall-spin-to-circulation closure: the outer family organizes geometry but does not explain the viscous endpoint.
+
+The periodic K\'arm\'an fields and actions contain a persistent rear-rotation reference plus a smaller correction under their own configuration.  The separate steady velocity field adds a low-deficit endpoint and motivates the viscous sequence above, but does not establish its links.  Distinguishing those links requires amplitude-matched interventions, held-out endpoint and convergence tests, wall-vorticity flux and separation diagnostics, and a closed same-control-volume momentum/mechanical-energy budget.  Until those measurements are available, the one-circle no-slip incompatibility and the opposite signs of $g_{\mathrm{opt}}$ and $g_{\mathrm{fit}}$ leave the inviscid circulation closure rejected.
+
+\draftfigure{Stationary, accepted and reverse steady endpoint fields and profiles, together with prescribed-circulation outer-reference streamlines and the opposite-sign closure comparison.}{Planned steady endpoint and theory boundary in the uniform-inflow/free-slip-wall configuration. The finite-$Re_D$ panels will distinguish the settling-qualified accepted endpoint from the unsteady stationary reference and the non-amplitude-matched reverse endpoint. Outer-reference panels will compare $g=0$, $g_{\mathrm{opt}}$ and the descriptive fitted $g$; no-slip incompatibility and the opposite signs of $g_{\mathrm{opt}}$ and $g_{\mathrm{fit}}$ reject the proposed wall-spin-to-circulation closure. The composition provides context, not a viscous mechanism or cross-configuration validation.}{fig:steady-theory-boundary}
+
+\section{Qualitative open-loop experiment}
+\label{sec:experiment}
+% AUTHOR-OWNED PLACEHOLDER: replace only after provenance review; this is not a reported result.
+\textit{[Author-supplied qualitative open-loop experiment text will be inserted here after provenance review.]}
+
+\draftfigure{Author-supplied apparatus schematic and, only after provenance review, selected experimental and separately generated CFD single-frame views for named open-loop conditions.}{Planned qualitative open-loop experiment figure. The apparatus and command/acquisition path will be identified from author-supplied records. Any retained experimental and CFD images will be labelled as separate, unregistered, acquisition-specific single-frame views for qualitative morphology only. No temporal change, velocity, force, field matching, quantitative agreement or CFD validation will be inferred.}{fig:open-loop-experiment}
+
+\section{Discussion and conclusions}
+\label{sec:discussion-conclusions}
+
+\subsection{Discussion}
+\label{subsec:discussion}
+
+Three independently rotating cylinders can alter wakes defined by different temporal and target conditions. In the uniform-inflow/free-slip-wall configuration, repeated K\'arm\'an training runs produced high-performing policies across the tested realizations. In the separately documented parabolic-inflow/no-slip-wall configuration, the controlled K\'arm\'an velocity field is closer to its declared target than physical zero in both complete-cycle mean and phase-resolved comparisons. The first result concerns repeatable policy discovery under the online objective; the second concerns the downstream field under a different configuration. Together they locate the periodic K\'arm\'an cloak as the case in which learned rotation is accompanied by a direct target-relative field comparison, without treating the two configurations as one continuous data set.
+
+Persistent counter-rotation of the rear cylinders is the clearest recurring feature of the periodic control. Removing that term degrades the compact surrogate more than removing the tested modulations. In the mean velocity field, the physical-zero pinball leaves a mean streamwise deficit relative to the target, while the controlled-minus-physical-zero field is largely opposite in sign. Holding the cylinders at the mean DRL action produces a constant-control field that accounts for most of the measured scalar mean-field error reduction; the remaining constant-to-DRL scalar increment is distinct from the centred velocity residual. That residual subtracts separately centred DRL and constant-control phase fields without cycle pairing, and its action-correlated modes describe phase organization rather than its size or cause. The surrogate ranking, scalar error increments and residual modes are therefore compatible but non-equivalent: they support persistent rear-cylinder rotation as an organizing component, not a unique decomposition or mechanism.
+
+Changing the task changes what can be inferred. For the incoming-vortex case, the policy was executed on an event-relative clock and the retained fields show target, controlled and physical-zero morphology at the selected event offset, without a scalar efficacy result. For Illusion, the policy was executed toward a declared non-zero wake target, and the retained role fields are consistent with that bounded morphology comparison; no scalar target-relative improvement is claimed. The separate steady case adds physical context: prescribed rear-cylinder counter-rotation accompanies a low-deficit mean-flow endpoint, while the wall kinematics point toward viscous shear-layer and transport effects. The prescribed-circulation outer flow cannot select circulation from cylinder spin, and its opposite sign rejects the proposed inviscid closure. Persistent rotation is therefore a useful kinematic coordinate across the stated cases, but not a transferred control law or causal explanation.
+
+\subsection{Conclusion}
+\label{subsec:conclusion}
+
+Rotational control was executed for periodic, event-relative and non-zero-target wake definitions. The periodic K\'arm\'an cloak provides the strongest field-level result: in the parabolic-inflow/no-slip-wall configuration, its controlled registered velocity field is closer to the declared target than physical zero. The Vortex and Illusion cases extend the tested task definitions through event-relative execution and bounded non-zero-target morphology, without supplying a common efficacy measure.
+
+For the periodic K\'arm\'an case, persistent rear-cylinder counter-rotation links the dominant tested surrogate term to the constant-control mean field, while the centred velocity residual describes time-dependent organization separately. This physical reading is deliberately limited: action ranking, scalar mean-field increments and residual modes answer different questions. Hydrodynamic cloak and illusion here denote control relative to a declared reference for specified observations and comparisons, not universal invisibility, configuration-independent transfer or an unrestricted physical law.
+
+\section*{Acknowledgements}
+The authors acknowledge the computational resources and discussions that supported this work.
+
+\section*{Funding}
+Funding information will be supplied before submission.
+
+\section*{Declaration of interests}
+The authors report no conflict of interest.
+
+\section*{Author ORCIDs}
+Author ORCID information will be supplied before submission.
+
+\bibliographystyle{jfm}
+\bibliography{../ROUND5_BOTTOM_UP/R5_REFERENCES}
+
+\end{document}

docs/JFM_WYQ/My_JFM/jfm.bib

@@ -0,0 +1,210 @@
+@article{Urzhumov2012,
+  author = {Urzhumov, Y. A. and Smith, D. R.},
+  title = {Fluid flow control with transformation media},
+  journal = {Phys. Rev. Lett.},
+  volume = {107},
+  pages = {074501},
+  year = {2011}
+}
+
+@article{Rabault2019,
+  author = {Rabault, J. and Kuchta, M. and Jensen, A. and R{\'e}glade, U. and Cerardi, N.},
+  title = {Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control},
+  journal = {J. Fluid Mech.},
+  volume = {865},
+  pages = {281--302},
+  year = {2019},
+  doi = {10.1017/jfm.2019.62}
+}
+
+@article{Paris2021,
+  author = {Paris, R. and Beneddine, S. and Dandois, J.},
+  title = {Robust flow control and optimal sensor placement using deep reinforcement learning},
+  journal = {J. Fluid Mech.},
+  volume = {913},
+  pages = {A25},
+  year = {2021},
+  doi = {10.1017/jfm.2020.1170}
+}
+
+@article{Tang2020,
+  author = {Tang, H. and Rabault, J. and Kuhnle, A. and Wang, Y. and Wang, T.},
+  title = {Robust active flow control over a range of {R}eynolds numbers using an artificial neural network trained through deep reinforcement learning},
+  journal = {Phys. Fluids},
+  volume = {32},
+  pages = {053605},
+  year = {2020},
+  doi = {10.1063/5.0006492}
+}
+
+@article{CornejoMaceda2021,
+  author = {Cornejo Maceda, G. Y. and Li, Y. and Lusseyran, F. and Morzy{\'n}ski, M. and Noack, B. R.},
+  title = {Stabilization of the fluidic pinball with gradient-enriched machine learning control},
+  journal = {J. Fluid Mech.},
+  volume = {917},
+  pages = {A42},
+  year = {2021},
+  doi = {10.1017/jfm.2021.301}
+}
+
+@article{Deng2020,
+  author = {Deng, N. and Noack, B. R. and Morzy{\'n}ski, M. and Pastur, L. R.},
+  title = {Low-order model for successive bifurcations of the fluidic pinball},
+  journal = {J. Fluid Mech.},
+  volume = {884},
+  pages = {A37},
+  year = {2020},
+  doi = {10.1017/jfm.2019.959}
+}
+
+@article{Loiseau2018Galerkin,
+  author = {Loiseau, J.-C. and Brunton, S. L.},
+  title = {Constrained sparse {G}alerkin regression},
+  journal = {J. Fluid Mech.},
+  volume = {838},
+  pages = {42--67},
+  year = {2018},
+  doi = {10.1017/jfm.2017.823}
+}
+
+@article{Loiseau2018,
+  author = {Loiseau, J.-C. and Noack, B. R. and Brunton, S. L.},
+  title = {Sparse reduced-order modelling: sensor-based dynamics to full-state estimation},
+  journal = {J. Fluid Mech.},
+  volume = {844},
+  pages = {459--490},
+  year = {2018},
+  doi = {10.1017/jfm.2018.147}
+}
+
+@article{Brunton2016,
+  author = {Brunton, S. L. and Proctor, J. L. and Kutz, J. N.},
+  title = {Discovering governing equations from data by sparse identification of nonlinear dynamical systems},
+  journal = {Proc. Natl Acad. Sci. USA},
+  volume = {113},
+  pages = {3932--3937},
+  year = {2016},
+  doi = {10.1073/pnas.1517384113}
+}
+
+@article{Gautier2015,
+  author = {Gautier, N. and Aider, J.-L. and Duriez, T. and Noack, B. R. and Segond, M. and Abel, M. W.},
+  title = {Closed-loop separation control using machine learning},
+  journal = {J. Fluid Mech.},
+  volume = {770},
+  pages = {442--457},
+  year = {2015}
+}
+
+@article{Li2017,
+  author = {Li, R. and Noack, B. R. and Cordier, L. and Bor{\'e}e, J. and Harambat, F.},
+  title = {Drag reduction of a car model by linear genetic programming control},
+  journal = {Exp. Fluids},
+  volume = {58},
+  pages = {103},
+  year = {2017}
+}
+
+@article{Schulman2017,
+  author = {Schulman, J. and Wolski, F. and Dhariwal, P. and Radford, A. and Klimov, O.},
+  title = {Proximal policy optimization algorithms},
+  journal = {arXiv preprint arXiv:1707.06347},
+  year = {2017}
+}
+
+@article{Cranmer2023,
+  author = {Cranmer, M.},
+  title = {Interpretable machine learning for science with {PySR} and {SymbolicRegression.jl}},
+  journal = {arXiv preprint arXiv:2305.01582},
+  year = {2023}
+}
+
+@article{Boree2003,
+  author = {Bor{\'e}e, J.},
+  title = {Extended proper orthogonal decomposition: a tool to analyse correlated events in turbulent flows},
+  journal = {Exp. Fluids},
+  volume = {35},
+  pages = {188--192},
+  year = {2003}
+}
+
+@article{Williamson1996,
+  author = {Williamson, C. H. K.},
+  title = {Vortex dynamics in the cylinder wake},
+  journal = {Annu. Rev. Fluid Mech.},
+  volume = {28},
+  pages = {477--539},
+  year = {1996},
+  doi = {10.1146/annurev.fl.28.010196.002401}
+}
+
+@article{Brunton2020,
+  author = {Brunton, S. L. and Noack, B. R. and Koumoutsakos, P.},
+  title = {Machine learning for fluid mechanics},
+  journal = {Annu. Rev. Fluid Mech.},
+  volume = {52},
+  pages = {477--508},
+  year = {2020},
+  doi = {10.1146/annurev-fluid-010719-060214}
+}
+
+@article{Raibaudo2020,
+  author = {Raibaudo, C. and Zhong, P. and Noack, B. R. and Martinuzzi, R. J.},
+  title = {Machine learning strategies applied to the control of a fluidic pinball},
+  journal = {Phys. Fluids},
+  volume = {32},
+  pages = {015108},
+  year = {2020},
+  doi = {10.1063/1.5127202}
+}
+
+@article{Li2022,
+  author = {Li, S. and Li, W. and Noack, B. R.},
+  title = {Machine-learned control-oriented flow estimation for multi-actuator multi-sensor systems exemplified for the fluidic pinball},
+  journal = {J. Fluid Mech.},
+  year = {2022},
+  doi = {10.1017/jfm.2022.908}
+}
+
+@article{Marra2024,
+  author = {Marra, L. and Cornejo Maceda, G. Y. and Meil{\'a}n-Vila, A. and Guerrero, V. and Rashwan, S. and Noack, B. R. and Discetti, S. and Ianiro, A.},
+  title = {Actuation manifold from snapshot data},
+  journal = {J. Fluid Mech.},
+  year = {2024},
+  doi = {10.1017/jfm.2024.593}
+}
+
+@article{Jin2021,
+  author = {Jin, B. and Illingworth, S. and Sandberg, R.},
+  title = {Optimal sensor and actuator placement for feedback control of vortex shedding},
+  journal = {J. Fluid Mech.},
+  year = {2021},
+  doi = {10.1017/jfm.2021.948}
+}
+
+@article{Gong2020,
+  author = {Gong, J. and Monty, J. and Illingworth, S.},
+  title = {Model-based estimation of vortex shedding in unsteady cylinder wakes},
+  journal = {Phys. Rev. Fluids},
+  volume = {5},
+  pages = {023901},
+  year = {2020},
+  doi = {10.1103/PhysRevFluids.5.023901}
+}
+
+@article{Nair2021,
+  author = {Nair, A. G. and Taira, K. and Brunton, B. W. and Brunton, S. L.},
+  title = {Phase-based control of periodic flows},
+  journal = {J. Fluid Mech.},
+  year = {2021},
+  doi = {10.1017/jfm.2021.735}
+}
+
+@article{Ishar2019,
+  author = {Ishar, R. and Kaiser, E. and Morzy{\'n}ski, M. and Fernex, D. and Semaan, R. and Albers, M. and Meysonnat, P. and Noack, B. R.},
+  title = {Metric for attractor overlap},
+  journal = {J. Fluid Mech.},
+  year = {2019},
+  doi = {10.1017/jfm.2019.447}
+}
+

docs/JFM_WYQ/My_JFM/lineno-FLM.sty

@@ -0,0 +1,113 @@
+%\usepackage[switch]{lineno}
+%\newcounter{lasttextlineno}
+\usepackage[pagewise,mathlines]{lineno}
+\runninglinenumbers %% for continous line numbering regardless of box and minipage but only on left side in single/double column
+%\linenumbers %% properly in single/double column but boxes and minipage linenos not getting continuous
+%\pagewiselinenumbers%%
+\newif \ifLineNumbers%
+\linenumberdisplaymath
+
+\def\linenumberfont{\normalfont\fontsize{8}{10}\selectfont}
+\if@twocolumn
+\setlength\marginparwidth{.75cm}
+\setlength\marginparsep{19.5\p@}
+\else
+\setlength\marginparwidth{3.0cm}
+\setlength\marginparsep{24\p@}
+\fi
+\newcommand\linenomathWithnumbersforams{%
+  \ifLineNumbers
+  %\ifx\@@par\@@@par\else
+    \ifnum\interlinepenalty>-\linenopenaltypar
+      \global\holdinginserts\thr@@
+      \advance\interlinepenalty \linenopenalty
+     \ifhmode                                   % v4.3
+      \advance\predisplaypenalty \linenopenalty
+     \fi
+      %\advance\postdisplaypenalty \linenopenalty
+      \advance\interdisplaylinepenalty \linenopenalty
+    \fi
+  \fi
+  \ignorespaces
+  }
+
+\newcommand\linenomathWithnumbersformultline{%
+  \ifLineNumbers
+  %\ifx\@@par\@@@par\else
+    \ifnum\interlinepenalty>-\linenopenaltypar
+      \global\holdinginserts\thr@@
+      \advance\interlinepenalty \linenopenalty
+     \ifhmode                                   % v4.3
+      %\advance\predisplaypenalty \linenopenalty
+     \fi
+      %\advance\postdisplaypenalty \linenopenalty
+      \advance\interdisplaylinepenalty \linenopenalty
+    \fi
+  \fi
+  \ignorespaces
+  }
+
+
+
+\newcommand*\patchAmsMathEnvironmentForLineno[1]{%
+\expandafter\let\csname old#1\expandafter\endcsname\csname #1\endcsname
+\expandafter\let\csname oldend#1\expandafter\endcsname\csname end#1\endcsname
+\renewenvironment{#1}%
+{\def\linenomath{\linenomathWithnumbersforams}%added
+ \@namedef{linenomath*}{\linenomathNonumbers}%added
+\linenomath\csname old#1\endcsname}%
+{\csname oldend#1\endcsname\endlinenomath}}%
+\newcommand*\patchBothAmsMathEnvironmentsForLineno[1]{%
+\patchAmsMathEnvironmentForLineno{#1}%
+\patchAmsMathEnvironmentForLineno{#1*}%
+}%
+
+\AtBeginDocument{%
+%\patchBothAmsMathEnvironmentsForLineno{equation}% deleted
+%\patchBothAmsMathEnvironmentsForLineno{align}%
+\patchBothAmsMathEnvironmentsForLineno{flalign}%
+\patchBothAmsMathEnvironmentsForLineno{alignat}%
+\patchBothAmsMathEnvironmentsForLineno{gather}%
+%\patchBothAmsMathEnvironmentsForLineno{multline}%
+%\patchBothAmsMathEnvironmentsForLineno{split}%
+\patchAmsMathEnvironmentForLineno{array}%
+%\patchAmsMathEnvironmentForLineno{split}%
+}
+
+%%\AtBeginDocument{%
+  \let\LN@align\align
+  \let\LN@endalign\endalign
+  \renewenvironment{align}%
+     {\linenomath\LN@align}%
+     {\LN@endalign\endlinenomath}%
+%%}
+
+
+\renewenvironment{multline}{\linenomathWithnumbersformultline%
+  \start@multline\st@rredfalse
+}{%
+  \iftagsleft@ \@xp\lendmultline@ \else \@xp\rendmultline@ \fi
+  \ignorespacesafterend
+\endlinenomath}
+\renewenvironment{multline*}{\linenomathWithnumbersformultline\start@multline\st@rredtrue}{\endmultline\endlinenomath}
+
+\def\insplit@{%
+  \global\setbox\z@\vbox\bgroup
+    \Let@ \chardef\dspbrk@context\@ne \restore@math@cr
+    \default@tag % disallow use of \tag here
+    \ialign\bgroup
+      \hfil%
+      \strut@
+       $\m@th\displaystyle{##}$% 
+     &$\m@th\displaystyle{{}##}$%
+      \hfill % Why not \hfil?---dmj, 1994/12/28
+      \crcr\noalign{\global\advance\c@linenumber\@ne}%
+}
+
+\makeatletter
+\setlength\marginparsep{12\p@}
+\setlength\marginparpush{0\p@}
+\setlength\marginparwidth{2.6pc}
+\makeatother
+
+\endinput

docs/JFM_WYQ/ROUND1_BOTTOM_UP/01_DRL_CONTROL_AND_CASES.md

@@ -0,0 +1,207 @@
+# Round-1 bottom-up evidence index: DRL, control and cases
+
+> Evidence index only; not JFM manuscript prose. Repository facts are bounded by authority documents, executable code, manifests and retained artifacts. LegacyCelerisLab and modern V5/CelerisLab are separate chains and are never pooled silently.
+
+## Executive map
+
+- **Active package:** `src/drl_pinball/**`; modern V5 is the executable 2000×600 CelerisLab PPO workflow. Legacy PPO and acquisition code remains the historical compatibility path.
+- **Scene families:** Legacy includes steady, Kármán cloak, vortex (Lamb/Taylor), Illusion and Erase, plus reduced-observation experiments. V5 retained scratch cases are 11 canonical cases: 4 Kármán Re/geometry groups, 4 Illusion target-size groups, 3 additional Kármán Re cases; 15 runs total.
+- **Primary control contract:** PPO, Sin MLP `[64,64]`, three cylinder rotations, six downstream velocity channels plus six force channels; Illusion adds two reconstructed target-force channels. V5 uses calibrated native-v2 physical scales plus frozen-at-inference SB3 `VecNormalize`; Legacy uses fixed Phase-0 norms.
+- **Reward rationale:** force terms impose low resultant force for cloak or target-force matching for Illusion; six-channel downstream DTW supplies an affordable temporal wake-signature objective. Full-field L2 is independent validation, not training reward.
+- **Principal evidence:** Kármán `Re100` is code-Re 100, equivalent to single-cylinder `Re_D=50` under `U0=0.01`, `D=20`, `nu=0.004`. V5 has seeds 41–45; retained deterministic evaluation selects seed 45. The current retained V5 package supports repeated policy discovery but seed-dependent late retention, not robust convergence.
+- **Field validation:** `wake_l2.py` schema `drl-pinball-wake-l2-v2`; canonical summary is 23 rows (15 V5 controlled, 7 Legacy periodic, 1 Legacy steady). Periodic `E_mean` is complete-cycle mean-field RMS; `E_phase8` is an eight nearest-snapshot diagnostic. Current matrix improves `eta_mean` in every row; V5 `kar_re400` seed 43 worsens `E_phase8` relative to zero.
+- **Status:** no new CFD or training run was performed here. `src/drl_pinball/data/reproduction` is external/symlink-backed in this checkout and not discoverable through the normal workspace glob; manifests and repository docs remain the provenance authority.
+
+## Source ledger
+
+| Layer | Authority / source | What it establishes | Boundary |
+|---|---|---|---|
+| Workflow | `.cursor/rules/jfm-research-workflow.mdc`; `ROUND1_BOTTOM_UP/README.md` | evidence-first workflow; Legacy/V5 separation; claim boundaries | not numerical authority |
+| Package map | `src/drl_pinball/README.md`, `CLAIMS.md`, `RESULTS.md`, `DATA_DICTIONARY.md`, `HISTORY_AND_LESSONS.md`, `WRITING_HANDOFF.md` | scope, result lineage, safe interpretation, artifacts | reader layer; artifacts win |
+| Legacy code | `legacy_env/*.py`, `legacy_train/*.py`, `legacy_test/core/dtw_metrics.py`, `legacy_test/metrics.py`, `legacy_test/acquire.py` | historical scene/action/obs/reward/DTW contracts | distinct solver and norm chain |
+| V5 code | `train/env_karman.py`, `train/env_illusion.py`, `train/case_registry.py`, `train/TRAIN_PIPELINE.md`, `train/DESIGN_DECISIONS.md` | executable calibration, PPO, reward, normalization, checkpoint contracts | retained runs predate some current contracts |
+| V5 result authority | `train/output/*/meta.json`, `train/results/latest/tables/latest_results_summary.json`, `latest_eval_summary.csv` | best values, iterations, selected seed, retained evaluation | retained historical; not post-fix retraining |
+| TB export | `train/results/training_csv/training_iterations.csv`, `training_csv/manifest.json` | 7,000 rows, 14 conflict-free runs | auxiliary; Re60 excluded fail-closed |
+| Field evaluation | `eval/wake_l2.py`, `eval/tests/test_wake_l2.py`, `WAKE_L2_AND_SENSOR_PLACEMENT_RESEARCH_NOTES.md` | ROI, formulas, phase/frequency diagnostics, exclusions | geometry-guarded, not solver-mask-proven |
+| Plot navigation | `data/reproduction_plots/manifest.json`, `plot_flow_fields.py`, `plot_reproduction_summary.py`; SR plot manifest/README | 30 V5/Legacy plots; 11 partial Legacy-SR plots | derived views; phase slot 0 default |
+| Solver setting | `configs/config_lbm_karman_2000x600*.json`, `configs/config_lbm_pinball.json`, `CELERISLAB_VALIDATION_DOSSIER.md` | V5/Legacy settings and CelerisLab qualification | validation anchors do not validate DRL mechanism |
+| Experimental navigation | `knowledge.md` §16.5; repository search | water tunnel/towing/force sensor/PIV are described as experimental setup/demos | no full experimental validation package located |
+
+File hashes recorded during this audit (SHA-256, bytes): `latest_results_summary.json` `4d5ef8f1cb1dbcbb9d5528bd0478c2c4815d4b49c92e0253b6ea54acac110617` (36582); training manifest `e5a7b096fb9dd737365fe37e37e748ab732697686d0b45ed08b2f3a980469acd` (1036); reproduction plot manifest `448251183cc80716c7b1b5b96ffdb5206eecaa2cd725fb67fbf67c09ac1bc32c` (45663); SR plot manifest `a5a4588219d658fed9e259aec91171643566c9472938317b8d0f2cbbf214ed4f` (54677); `eval/wake_l2.py` `6b3a9b093b427b06cb5dafcc2ad3a648971a647d5bbfb7a47fc21ee9dd00ebd7` (15975); L2 research note `f97f7044603e0ee3f52cd21c4a514a7250b188dd6e7ece40cb8947aacb7506cb` (39396).
+
+## Chronology and package evolution
+
+1. **Legacy PPO era:** original `LegacyCelerisLab` workflow established steady cloak, Kármán cloak, Illusion, vortex, Erase and reduced-observation experiments. Legacy model inventory and scene notes are consolidated in `knowledge.md`; exact old contracts live in `legacy_env` and `legacy_test`.
+2. **Steady contextual start:** near-uniform/constant rotation was explored before periodic Kármán control. It is contextual calibration and not active DRL robustness evidence.
+3. **Kármán Re100 anchor:** the original periodic cloak chain became the main DRL reference. Code name `re100` means `U0*(2D)/nu=100`; physical one-cylinder `Re_D=50`.
+4. **Transfer and scene expansion:** Legacy transfer was used for vortex and Erase; Illusion introduced target-cylinder force/harmonic matching; reduced-observation experiments probed hardware simplification.
+5. **V5 migration:** CelerisLab changed the solver/API and moved to a 2000×600 grid, uniform/free-slip setting, explicit case registry, calibration-v2/native units, zero bias and `VecNormalize`. V5 was intentionally changed for extensibility and robustness information, not as a numerically identical continuation of Legacy.
+6. **V5 scratch matrix:** 15 retained runs across 11 cases. Re100 has five seeds; all other canonical cases have one. `kar_d075` and `kar_d2` use `1e-4` learning rate versus mostly `3e-4`, confounding a pure radius ablation.
+7. **Evidence consolidation:** training, deterministic policy evaluation, reproduction acquisition and offline wake-L2 evaluation were separated. Result figures/tables are reader-derived products; no current-contract complete retraining is established.
+8. **Current indexing:** plots, manifests, field-role metadata, phase diagnostics and claim ledgers were frozen for downstream agents. SR/CCD/OID are interfaces only here; Illusion SR is historical/negative and not a DRL positive claim.
+
+## Case taxonomy and contracts
+
+### Legacy scene matrix
+
+| Scene / IDs | Geometry and target | Re convention / role | Action mapping (normalized action `a`) | Observation / interval |
+|---|---|---|---|---|
+| steady | three-cylinder pinball; target and constant-control contextual comparison | Legacy steady; late snapshot | scene-specific legacy constant role | nonperiodic; `late_field.npz` |
+| Kármán cloak: `karman_re50/100/200/400` | upstream disturbance cylinder + 3 pinball cylinders; target is disturbance wake | code Re uses `2D`; Re100 ↔ `Re_D=50` | `omega/U0 = a*8 + [0,-4,+4]` | 12 policy channels; `SI=800`, `FIFO=150`, `CONV=30`, `MAX_STEPS=500` |
+| vortex: `vortex_lamb`, `vortex_taylor` and offsets | transient Lamb dipole or Taylor monopole passes through pinball; target is evolving sensor signal | Legacy transfer from Re100 documented; event roles | `omega/U0 = a*4 + [0,-4,+4]` | 12 channels; `SI=800`, `MAX_STEPS=150`; event-relative fields, not periodic phase |
+| Illusion: `illusion_075L/1L/15L` | standalone target cylinder at target-size ratio plus 3 sensors; pinball imitates target | target force and sensor signal; Legacy target body and pinball are separate runs | `omega/U0 = a*8 + [0,-2,+2]` | 14 channels: 12 force/sensor + target `Cd,Cl`; `CONV=36`; target harmonics are 5-frequency/channel reconstruction |
+| Erase | clean inflow target, disturbance-carrying pinball scene; target is mean/clean sensor state | nonperiodic/diagnostic semantics | `omega/U0 = a*8 + [0,-8,+8]` | special 14-channel raw layout; `SI=600`, `CONV=36`; enhanced DTW reward |
+| reduced observation | Kármán layout with force/torque/sensor subsets | hardware simplification experiments | generally cloak scale/bias | S_DIM varies 3–9; historical, not main chain |
+
+Legacy geometry in lattice cells is `L0=20`, `NX=1280`, `NY=512`, `y0=255.5`; Kármán disturbance center `x=200`, pinball front `x=600`, rear `x=626`, sensors `x=800` at `y0+40,y0,y0-40`; pinball radius 10 and sensors radius 5. Illusion shifts pinball to x=380/406, target to x=400, sensors to x=600. The raw Kármán observation is `[dist force(2), sensors(6), pinball forces(6)]`; training slices `[2:14]`, then normalized policy order is **forces first, sensors second**, despite older prose saying otherwise. Illusion raw is sensors then forces, with target force channels appended after normalization.
+
+### V5 canonical matrix
+
+| Family | Canonical IDs | seeds | SI | geometry / Re |
+|---|---|---:|---:|---|
+| Kármán baseline | `kar_re100` | 41,42,43,44,45 | 800 | 2000×600; code Re100, `Re_D=50`; disturbance radius default |
+| Cross-Re | `kar_re60`, `kar_re200`, `kar_re400` | 43 | 800,500,400 | matching viscosity configs; code-Re labels |
+| variable disturbance | `kar_d075`, `kar_d15`, `kar_d2` | 44,45,45 | 800 | radius ratios 0.75, 1.5, 2.0; not pure LR-controlled geometry comparison |
+| Illusion | `ill_075L`, `ill_1L`, `ill_15L`, `ill_2L` | 43 | 1100,1200,1200,1200 | target radius ratios passed as `target_diam*L0`; physical radii 15/20/30/40, diameters 30/40/60/80 |
+
+V5 geometry: `D=20`, front `(1000,299.5)`, rear `(1026,314.5)/(1026,284.5)`, sensor plane `x=1200` for Kármán; Illusion front/rear `(380,299.5)/(406,314.5)/(406,284.5)`, sensor/target plane `x=600`. V5 config uses `D2Q9`, FP32, MRT, double-buffer, uniform regularized inlet, `neq_extrap` outlet with backflow clamp and free-slip y walls. `config_lbm_pinball.json` is Legacy/new compatibility context, not the active V5 2000×600 training configuration: it is 1280×512, parabolic Zou-He and bounce-back walls.
+
+## Formula and implementation contracts
+
+### Legacy normalization and reward
+
+Legacy Phase-0 normalization is frozen per scene/model and must be paired with the saved `norm.json`; fresh recomputation is invalid for reference comparisons.
+
+- Standard Kármán/Illusion/Vortex: `force_norm_fact = 6*max(|forces|)`; `sens_deviation=mean(sensor)`; `sens_norm_fact=5*max(|sensor-sens_deviation|)`.
+- Erase uses special force/sensor factors 100/10; reduced-observation has model-specific factors.
+- Policy input is `hstack([forces_norm(6), sens_norm(6)])`, clipped to `[-1,1]`; Kármán starts from raw `obs[2:14]`.
+- Legacy action persistence is built-in exponential smoothing: `a_applied=(1-0.1)*a_previous+0.1*a_target`. V5 has an explicit equivalent `ActionSmoother(weight=0.1)`.
+
+For standard Legacy Kármán, with normalized forces `cd=(f_x^front+f_x^bottom+f_x^top)/3` and similarly `cl`, `sim=mean_i(1-DTW_i/CONV_LEN)`,
+`r_cd=exp(-|20 cd|)`, `r_cl=exp(-|80 cl|)`, `r_sim=exp(-10|sim-1|)`, and `r=min(0.3 r_cd+0.4 r_cl+0.3 r_sim,1)`. Vortex uses weights `(0.2,0.3,0.5)`. Illusion uses summed three-pinball force against reconstructed target force, weights `(0.3,0.3,0.4)`, and sensitivities 10/10. Erase uses `0.4 r_u+0.4 r_v+0.2 sim`, where `r_u=exp(-40|sensor_u-target_u|)`, `r_v=0.7 exp(-20|amp_v|)+0.3 exp(-20|sensor_v-target_v|)`, and enhanced similarity is 80% max-cost DTW + 10% amplitude ratio + 10% mean similarity.
+
+### Legacy DTW exact implementation
+
+`legacy_test/core/dtw_metrics.py` computes an unweighted absolute-cost dynamic program:
+`D[i,j]=|target[i-1]-state[j-1]|+min(D[i-1,j],D[i,j-1],D[i-1,j-1])`, `D[0,0]=0`, other boundaries infinity; `sim=1-D[n,m]/n`. Kármán lag is `argmax` of cross-correlation of middle-sensor Uy (`target[:,1]` vs state); six channels use `roll(target,-lag)[CONV:2CONV]` versus the last `CONV` state samples and are averaged. Illusion estimates lag from target force/sensor offset channel `target[:,3]` against state middle Uy `state[:,1]`, then target sensor channels are `target[:,i+2]`. Vortex has no lag and rolls the last window by `current_step+1`. DTW lag is a signal alignment index, not physical delay or causality.
+
+`legacy_test/metrics.py` separately provides offline target-channel max-abs normalized cycle DTW: each target channel scale is `max(abs(target[:,i]))` (floor `1e-12`), both arrays are divided by that scale, then the same Legacy cycle DTW is computed. This is not the native reward formula. It records the exact `scale`, lag channel and `conv_len` in `dtw_summary.json`.
+
+### V5 DTW and reward
+
+V5 `calc_dtw_sim` uses the same absolute-cost DP but `sim=max(0,1-D[n,m]/(n*dtw_norm_scale))`. `dtw_norm_scale` is calibration-specific, not a universal constant. Examples from calibration artifacts: Kármán Re100 `0.0026153846153846153` in the canonical calibration and retained run-local calibration `0.204` (the latter demonstrates why calibration identity must accompany numbers); Re400 `0.0039743589743589745` canonical and retained run-local `0.31`; Illusion canonical `ill_075L 0.0023846153846153848`, `ill_1L 0.0032307692307692306`; retained run-local values are `0.186`, `0.252`, `0.355`, `0.433`. The code and artifact schema, not a prose approximation, decide which value applies to a result.
+
+V5 computes lag from target/state middle-sensor channel index 3 in the six-channel `[s0 ux,uy,s1 ux,uy,s2 ux,uy]` convention, selects target `[CONV:2CONV]` and state tail `[-CONV:]`, averages six DTW values, interpolates through calibration `SIM_BP`→`SIM_VAL`, and computes:
+`r_cd=EMA(exp(-cd_norm² K_CD),0.2)`, `r_cl=EMA(exp(-cl_norm² K_CL),0.2)`, `reward=W_CD r_cd+W_CL r_cl+W_SIM r_sim-floor_penalty`, clipped at zero. Kármán force aggregation is the **mean** of three controlled-cylinder forces; Illusion is the **sum** matched to one target-cylinder force. Canonical Kármán constants are `K_CD=50,K_CL=100`; Illusion `12,25`; weights are `0.3,0.3,0.4`; floor thresholds are 0.1 and floor penalty 0.05. V5 normalized physical observation is `[forces/SCALE, sensors/SCALE]`, with Illusion target `Cd,Cl` appended; VecNormalize whitens observations only (`norm_reward=False`, `clip_obs=10`, `gamma=.99`).
+
+### Sin activation and PPO protocol
+
+Both retained model inventory and V5 source specify custom `Sin.forward(x)=sin(x)` in an MLP `Linear→Sin→Linear→Sin→Linear`, hidden widths 64×64. The historical rationale is periodic/vortex spectral fidelity; it is a design rationale, not an independently demonstrated activation ablation. V5 defaults: PPO SB3, `n_steps=2048`, batch 64, 10 epochs, learning rate `3e-4`, `gamma=.995`, deterministic evaluation; action range `[-1,1]`, `omega=-(action*12)*U0/RADIUS`, EMA 0.1. Each logged training iteration is `learn(2048)` plus up to 360 deterministic evaluation steps; scoring tail is last 180. Environments do not terminate Gym episodes. Best retention saves `best_model.zip` together with same-selection `best_vecnormalize.pkl`; per-iteration model/normalizer pairs exist in source workflow but intermediate/final streams are not released. Resume is continuation, not exact replay. Seeds distinguish stochastic runs, not deterministic GPU replay. Symmetry wrapper exists but active default is off; historical configured probability 0.5 did not yield effective transition augmentation under the reset timing audit.
+
+## Result ledger: V5 retained training and evaluation
+
+Authority: `train/results/latest/tables/latest_results_summary.json` and `latest_eval_summary.csv`; table hash above. Values below are retained historical deterministic evaluation, 360 steps, last-180 scoring tail, `legacy-policy-v1` compatibility, not current-contract retraining.
+
+| case | seed(s) / n | eval reward | DTW | r_cd | r_cl | r_sim | status / failure mode |
+|---|---:|---:|---:|---:|---:|---:|---|
+| `kar_re100` | selected 45 / 5 | 0.931258 | 0.918458 | 0.981787 | 0.980641 | 0.856324 | main anchor; selected-seed multi-seed result |
+| `kar_d075` | 44 / 1 | 0.949619 | 0.910736 | 0.966266 | 0.974395 | 0.918552 | demonstration; LR 1e-4 confound |
+| `kar_d15` | 45 / 1 | 0.957656 | 0.892390 | 0.993683 | 0.996858 | 0.901234 | demonstration |
+| `kar_d2` | 45 / 1 | 0.909436 | 0.816482 | 0.976047 | 0.988719 | 0.800015 | demonstration; LR 1e-4 confound |
+| `ill_075L` | 43 / 1 | 0.878312 | 0.880188 | 0.935905 | 0.818551 | 0.879939 | high action saturation, 0.8222 any-channel fraction |
+| `ill_1L` | 43 / 1 | 0.766871 | 0.893062 | 0.914360 | 0.706057 | 0.701863 | high action saturation, 0.9611 any-channel fraction |
+| `ill_15L` | 43 / 1 | 0.596248 | 0.877500 | 0.787644 | 0.349086 | 0.638072 | force/lift degradation; saturation 0.8444 any |
+| `ill_2L` | 43 / 1 | 0.579589 | 0.869949 | 0.728111 | 0.173972 | 0.783248 | force/lift degradation; saturation 0.9000 any |
+| `kar_re60` | 43 / 1 | 0.604643 | 0.364287 | 0.935710 | 0.738946 | 0.255616 | weak DTW; excluded from clean TB CSV due conflicting resume history |
+| `kar_re200` | 43 / 1 | 0.498573 | 0.711742 | 0.498072 | 0.314928 | 0.652500 | degraded cross-Re demonstration |
+| `kar_re400` | 43 / 1 | 0.446254 | 0.564630 | 0.611165 | 0.305857 | 0.448598 | degraded cross-Re demonstration |
+
+V5 retained best training rewards: Re100 seeds 41–45 are `0.9266385, 0.9301960, 0.9160065, 0.9221958, 0.9411612`; best iterations `471,315,360,465,447`. Last-50 median retention ratios are `0.6703,0.2004,0.1626,0.8581,0.6373`; this is direct evidence for discovery versus late retention separation. Best canonical training values are in each run `meta.json`; TB export has 7,000 rows from 14 runs, excludes Re60 because 213 conflicting duplicate steps and mixed-config resume history.
+
+## Result ledger: field/L2 artifacts
+
+`wake_l2.py` schema `drl-pinball-wake-l2-v2` uses `D=20`; ROI `-6≤(x-xs)/D≤14`, `|(y-y0)/D|≤5`, inclusive lattice cells. V5 sensor-plane registrations are Kármán `xs=1200`, Illusion `xs=600`; Legacy Kármán `xs=800`, Illusion `xs=600`. `E_mean` is complete-cycle arithmetic mean-field U0-normalized area RMS; `E_phase8` is eight independently phased nearest-boundary snapshots; `eta=1-E_ctl/E_zero`; time is `tU0/NX`, so frequencies are relative within case only. Exact solver masks are not persisted; this is geometry-guarded, not exact common-fluid-mask L2. Vortex/Erase are excluded from the formal matrix; steady is late-snapshot only.
+
+Canonical summary (memory and L2 note; read JSON/CSV before final numerical writing): 23 rows = 15 V5 controlled seeded roles, 7 Legacy periodic roles, 1 Legacy steady constant role. All 23 have positive `eta_mean` (`0.1092..0.8933`). Exception: V5 `kar_re400` seed 43 has `E_phase8_ctl=0.490471` versus zero `0.450626`, `eta_phase8=-0.08842`, while mean-field still improves (`E_mean_ctl=0.177469` versus zero `0.221541`, `eta_mean≈0.1989`). It also has relative frequency mismatch `0.2650` and period CV `0.27346`; phase claim is dynamically weak. Other audit diagnostics include largest actual-phase slot delta `0.10286 rad` (Legacy illusion_075L), largest selected-role phase error `0.08699 rad` (Legacy karman_re100), unique controlled cycle counts 3–8, and no controlled role reused the same field index across slots.
+
+## Kármán Re100 principal evidence map
+
+- **Definition:** V5 `kar_re100`: 2000×600, `nu=.004`, `U0=.01`, code reference `2D=40`; Legacy `karman_re100`: 1280×512, parabolic Zou-He, bounce-back walls, Legacy API. Never call the label unqualified “cylinder Re=100”.
+- **Training:** five retained scratch seeds, 500 iterations; high best reward on all five, but late retention varies strongly. The strongest retained policy selection is seed 45 (`eval reward .931258`, DTW `.918458` under the retained evaluation contract).
+- **Signals:** six sensors at one downstream plane, three probes with `(y-y0)/D=+2,0,-2`, radius `.25D`; forces and sensors are distinct objective channels.
+- **Field:** reproduction plots exist for all five V5 seed views and Legacy Karman Re100; the default plot is phase slot 0. Source phase NPZ, timeseries, metadata and external-root paths are manifest-bound where available.
+- **Interpretation boundary:** Re100 supports a named-condition multi-seed discovery/retention statement and field comparisons; it does not establish universal robustness, causality, stability, response delay or a unique mechanism.
+
+## Cross-condition ability, degradation and failures
+
+- Cross-Re retained evaluation degrades from Re100 to Re60/Re200/Re400 in DTW and reward; n=1 each, so this is named-condition evidence, not a continuous generalization law.
+- Variable disturbance cases retain high reward/DTW but are confounded by learning-rate differences for `d075` and `d2`; no pure geometry ablation.
+- Illusion reward and force components decline as target size increases, while action saturation is high; this documents difficulty/degeneration, not target-size law.
+- Re60 is a specific provenance failure in the clean training-series export: conflicting duplicate TensorBoard values at 213 steps and a mixed-config resume branch. It remains in the retained result table but is not used in the clean CSV.
+- Legacy Illusion DRL/PPO remains a historical target-matching demonstration under its own Legacy contract. Legacy Illusion SR is negative/historical and does not change the PPO role.
+- Vortex and Erase are retained navigation/artifact roles, but their transient/nonperiodic semantics are excluded from current formal L2 matrix; do not convert their event/late plots into periodic phase evidence.
+
+## Evidence and artifact inventory
+
+### Training and evaluation
+
+- `train/output/<case>_seed<seed>/meta.json`, `calibration.json`, `target.npy`, `models/best_model.zip`, matching `best_vecnormalize.pkl` (compact retained bundle; inspect per-run metadata for hashes and provenance).
+- `train/results/latest/tables/latest_results_summary.json`, `latest_eval_summary.csv`, `HOW_TO_READ.md` (authoritative reader tables and limitations).
+- `train/results/training_csv/training_iterations.csv` + manifest (7,000 rows; 14 runs; Re60 exclusion reason/hash).
+- `train/results/latest/figures/fig01*–fig05*` and retained evaluation figures are referenced by `HOW_TO_READ.md`; they are derived plotting products and should not be mistaken for new CFD.
+
+### Reproduction / field material
+
+- `data/reproduction_plots/manifest.json`: 30 plots; fixed vorticity `[-0.001,+0.001]`; V5 seeded and Legacy Kármán/Illusion/steady/vortex/erase role mappings; default selector is phase slot 0, vortex selector is event relative offset `-10`, steady selector is late field index 159.
+- `data/reproduction_plots_sr/manifest.json` and README: 11 plots out of expected 23, explicitly partial; Kármán Target/PPO/SR/Zero and term-deletion variants; generalization cases intentionally Target/SR/Zero with no PPO. Manifest records hashes, sizes and absolute external role paths.
+- Plot code: `plot_flow_fields.py`, `plot_reproduction_summary.py`, `train/plot_latest_results.py`, `train/plot_scratch_results.py`; offline L2 code/tests under `eval/`.
+- Raw role artifacts, when external store is mounted: `phase_fields.npz` (eight slots plus means/phase diagnostics), `event_fields.npz` (five event offsets), `late_field.npz`, `timeseries.csv/.npz`, `metadata.json`, role manifests and source hashes. The current normal workspace did not expose these directories; do not infer absence from that tooling limitation.
+
+### Experimental/demo navigation
+
+`knowledge.md` §16.5 records a water tunnel with towing platform (25 cm width), custom air-bearing/2D force-sensor system with semiconductor strain gauges, low-noise acquisition and planar PIV; repository search found no dedicated current water-tunnel/demo evidence dossier or raw experimental dataset. Status is demo/feasibility navigation only. Required future inventory: experiment date/configuration, scale, towing speed/Re, sensor calibration and sampling, PIV raw/processed fields, uncertainty, synchronization, and exact figure source data.
+
+## Cross-package interfaces and boundaries
+
+- **SR:** Legacy standardized acquisition reuses target/physical-zero/PPO role contracts, sensor/force/action ordering, 480 warm-up control intervals + 160 retained boundaries where documented, and the same Legacy solver. This index only points to `src/SR_analysis/` and `data/reproduction_plots_sr`; SR law/effect/ablation claims belong to the SR agent. Illusion SR is closed as negative/historical.
+- **CCD/OID:** DRL provides controlled/target/zero signals and fields; CCD describes control-vs-constant differences, OID has separate dual-clock and exact-mask contracts. Neither metric is a DRL score or causal proof. Keep field clocks separate from control clocks.
+- **Theory/steady:** steady constant-control and Legacy steady plots are contextual calibration; theoretical rotation interpretations must not be imported into DRL results without matching solver/geometry/action sign contracts.
+- **Solver validation:** CelerisLab qualification is separate: only Kan99b K2 and Sah04 S2 full runs retained; Kan99 lift sign remains unresolved, qualification matrix is partial, and no solver validation proves DRL mechanism. Legacy equivalence is prohibited.
+
+## Claim boundaries, conflicts and gaps
+
+### Authorized / supported at named scope
+
+- V5 retained PPO uses calibrated targets, native-v2 physical scaling, observation-only VecNormalize, zero bias, explicit EMA smoothing and Sin MLP.
+- Re100 has five retained seeds and repeated high-reward checkpoint discovery; late retention is seed-dependent.
+- Retained V5 cross-Re, disturbance-size and Illusion cases are named-condition demonstrations with the result values above.
+- Six-channel sensor DTW and independent field L2 are distinct evidence roles; field metric is exactly specified by `wake_l2.py`.
+- CelerisLab is a GPU D2Q9 MRT LBM implementation with documented sensor-readback validation and partial rotating/confined-cylinder qualification.
+
+### Prohibited without new evidence
+
+Do not claim universal Re generalization, distribution-wide robustness, asymptotic stability, causal wake mechanism, physical response delay from DTW, pure disturbance-radius ablation, post-fix complete retraining, exact Legacy/V5 numerical equivalence, full-state observability from three probes, optimal `x/D=10` or `y/D=±2`, exact solver-mask L2, or statistical uncertainty for n=1 cases.
+
+### Conflicts / gaps to resolve later
+
+1. Retained V5 run-local `calibration.json` values for `dtw_norm_scale` differ from canonical calibration directories; every number needs exact calibration identity and target hash.
+2. `train/results/latest` evaluation uses `legacy-policy-v1` compatibility while current source describes native-v2; this is intentional retained-history boundary but limits direct current-code interpretation.
+3. `src/drl_pinball/data/reproduction` role tree is external/symlink-backed and not exposed by ordinary workspace glob; audit its mounted tree before quoting per-role file hashes or schema values.
+4. Legacy raw and normalized observation descriptions had historical channel-order conflict; executable `legacy_env` and `legacy_test` code wins: normalized policy vector is forces first, sensors second.
+5. Wake L2 has no persisted solver-exact masks; future acquisition should persist common-fluid masks and, if needed, repeated-crossing phase-bin fields.
+6. Vortex/Erase require explicit event/nonperiodic metrics before formal cross-scene field claims; current plots are navigation, not a common periodic matrix.
+7. No dedicated current water-tunnel/demo raw data or synchronized PIV/force package found.
+8. CelerisLab validation dossier records only K2/S2 full qualification, unresolved Kan99 lift sign, no mesh-convergence package, no uncertainty ensemble and no full Kan99 fields.
+9. Exact per-artifact hashes for all V5 model bundles, calibration/target pairs and external raw NPZ should be extracted from their manifests before final figure panels.
+10. Figure inventory is complete at navigation level, but publication-ready panels/data exports do not yet exist; future panel assembly must preserve role, seed, phase/window, comparator and hash.
+
+## Candidate physical logic (not frozen claims)
+
+- The reward deliberately combines global force balance and a sparse downstream wake signature: force terms penalize persistent resultant drag/lift, while sensor DTW asks for temporal target-signature agreement. This division is physically motivated by experimental feasibility and computational cost, but force/DTW reward success does not itself prove full-field cloaking.
+- In coherent Kármán regimes, centre `u_y` provides a phase-sensitive shedding coordinate; upper/lower probes add transverse coverage and streamwise/transverse complementarity. This can explain why sparse history can guide control, while L2 tests what lies between and beyond probes.
+- The repeated Re100 high-best/variable-late pattern suggests a policy-search and retention issue, not automatically a fluid instability. V5 cross-condition degradation and Illusion saturation identify where the retained controller departs from the anchor regime.
+- Steady, periodic and transient cases should be treated as distinct dynamical contracts. Near-constant action structure may be a candidate control skeleton, but establishing that belongs to SR/theory evidence and must retain action signs, solver chain and case identity.
+
+## Quick handoff checklist
+
+For every future numerical claim record: `path`, SHA-256, schema, array shape/keys, units, geometry/Re convention, role/case, seed, comparator, metric formula, evaluation window, normalization/calibration/target identity, and status (`current`, `retained historical`, `diagnostic`, `negative`, or `gap`). Read artifacts before prose; keep this file an index rather than a manuscript.

docs/JFM_WYQ/ROUND1_BOTTOM_UP/02_SR_CONTROL_LAWS.md

@@ -0,0 +1,216 @@
+# Round 1 bottom-up evidence dossier: symbolic-regression control laws
+
+> **Role.** This is an index and evidence dossier for later condensation, not manuscript prose. It records repository facts, bounded observations, interpretations, negative/historical results, and unresolved questions. Repository authority/code/manifests/artifacts outrank memory and old narrative. No immutable artifact was modified; no CFD was launched.
+>
+> **Authority status at 2026-08-10.** Active SR science is LegacyCelerisLab Kármán/cloaking only. `steady` is contextual calibration. Illusion SR is retained historical/negative evidence. Modern CelerisLab V5, DRL, CCD, OID and steady packages are interface/context only and are not pooled into this SR chain.
+
+## 1. Scope and questions
+
+This dossier covers `src/SR_analysis/**`, the Legacy acquisition/deployment interface in `src/drl_pinball/legacy_test/**`, and the relevant Legacy solver boundary in `LegacyCelerisLab/**`. It indexes:
+
+- the real Stage 1 → Stage 2 → Stage 3 workflow and the closed-loop CFD arbiter;
+- state/action pairing, feature construction, normalization, action order, units, Reynolds convention and reflection contracts;
+- canonical formula/tree/coefficient identity and deployment mapping;
+- Kármán training, validation, finite-duration and named pointwise extension cases;
+- ablation/deletion/scaling evidence and its strict interpretation;
+- steady, vortex/other cloak transfer boundaries, Illusion SR’s negative archive status, offline-fit versus closed-loop evaluation, failures, and future SINDy/tree-search boundaries;
+- a path/hash/result ledger, formula table, case matrix, figure/data/script inventory, and gaps requiring verification.
+
+**Evidence labels used below:** `FACT` = directly stated by current authority/code/manifest/artifact; `OBS` = bounded observed output; `INTERPRETATION` = permitted interpretation; `HISTORICAL/NEGATIVE` = retained but not active positive evidence; `UNSUPPORTED` = do not claim without new authority; `GAP` = needs targeted verification. Numerical values are copied only where a current package index or artifact exposes them.
+
+## 2. Sources inspected and authority ordering
+
+### 2.1 Repository sources
+
+| Layer | Paths | Use/status |
+|---|---|---|
+| Cross-repo authority | `README.md`, `src/README.md` | solver separation, current status, evidence hierarchy |
+| SR authority | `src/SR_analysis/README.md`, `PIPELINE.md`, `results/README.md`, `results/INDEX.md`, `HISTORY_AND_LESSONS.md`, `HANDOFF.md` | current scope, contracts, claim ledger/navigation |
+| Executable contracts | `configs.py`, `stage_1_infer.py`, `stage_2_fit.py`, `stage_3_validate.py`; `utils/{data_contracts,feature_builder,g_operator,formula_schema,metrics,provenance}.py` | executable order, alignment, features, G, formula identity, DTW and hashes |
+| Stage 1 evidence | `data/runs/article-joint-data-karman-20260718/**`, `data/runs/article-joint-data-illusion-20260718/**` | manifests/config/result and model/norm provenance |
+| Derived result packages | `results/runs/article2-plotting-package-20260721/**`, `article2-steady-analysis-20260720/**`, `article2-long-timeseries-csv-20260720/**` | derived tables, steady interpretation, long-run manifests/tables |
+| Legacy interface | `src/drl_pinball/legacy_test/README.md`, `acquire.py`; `legacy_test/tests/**`; `LegacyCelerisLab/README.md` | role semantics, warm-up/phase/publication contract, solver boundary |
+| Historical archive | `src/SR_analysis/archive/README.md`, `archive/**` | failed/superseded routes only; not active authority |
+
+### 2.2 External memory and literature routing
+
+- **Nowledge Mem:** used `read_context_bundle` (Working Memory loaded) and a targeted `memory_search` for SR/Legacy/Kármán/Illusion. The targeted search returned zero durable results; Working Memory supplied the relevant prior decisions but was treated only as retrieval context, not scientific evidence.
+- **Undermind:** called orientation, listed workspaces, and inspected the existing `DynamisLab` workspace (`4383d145-2c80-4991-8a9c-7ac2dd683438`). Existing SR notes and literature folders were found, including `SR_analysis/` and `JFM References/03 Symbolic regression and interpretable control/`, `05 Hydrodynamic cloaking and flow illusion/`. No new deep search was started. No external method claim is needed to establish repository facts in this dossier.
+
+## 3. Real pipeline: Stage 1 → 2 → 3 plus standardized roles
+
+```text
+Frozen PPO + Legacy scene/norm
+  └─ Stage 1: infer + record immutable PPO trajectories; replay/order/provenance gates
+       └─ accepted controlled.npz + config/manifest/norm/result/target artifacts
+            └─ Stage 2: broad discovery (per-case and within objective, several seeds)
+                 └─ topology recurrence / deployment safety diagnostics
+                      └─ freeze topology; refit coefficients on declared eligible data
+                           └─ formula JSON + expression hashes + parent/source hashes
+                                └─ Stage 3: static safety → 40-step L2 screen → 200-step L3 Legacy closed loop
+                                     └─ finite 400-step / named unseen points / deletion / steady context
+```
+
+**CFD arbiter (FACT).** Stage 2 offline fit metrics (R²/RMSE/Pareto) diagnose imitation and topology; they do not accept a controller. Stage 3 closed-loop Legacy CFD, finiteness, early termination, schema/order, provenance, declared metric, and comparator arbitrate. One failed case must not be hidden by an average. Failed/rejected/partial telemetry remains evidence.
+
+**Separate August collector (FACT).** `src/drl_pinball/legacy_test/acquire.py` is not Stage-3 replay or a refit pipeline. It reruns frozen `target`, `controlled`/PPO, `sr`, `zero`, and named deletion roles under a common contract: 480 warm-up intervals, 160 retained post-step boundaries, same-run fields/telemetry, independent phase, complete cycles, eight fixed-phase snapshots, complete-cycle means, native and separately named normalized DTW, transactional staging, exact-schema validation and no-clobber publication. It audits late/standardized role behavior and complements the July chain.
+
+## 4. Contracts: pairing, features, normalization, units and symmetry
+
+### 4.1 Data pairing and windows
+
+| Contract | Verified rule | Evidence |
+|---|---|---|
+| Array schema | sensors `(T,6)`, forces `(T,6)`, actions `(T,3)`; Illusion target forces `(T,2)` | `utils/data_contracts.py` |
+| Causal fit pairing | post-state `i` predicts action `i+1`; history uses action `i` and `i-1`; trajectory-local | `CAUSAL_ALIGNMENT`, `build_batch_dataset` |
+| Historical pairing | post-state `i` → same action `i`, with historical lag construction; retained for compatibility, not silently mixed | `HISTORICAL_ALIGNMENT` |
+| Splits | contiguous, non-overlapping train/validation/blind blocks; no trajectory boundary crossing | `contiguous_block_slices`, `PIPELINE.md` |
+| Stage 1 | four trained Kármán cases re-code 50/100/200/400; 200 recorded PPO steps; 197 causally aligned rows per scene | `results/README.md`, plotting table |
+| Stage 3 | L2 gate 40 control steps; L3 standard 200; Article2 long 400; named Kármán extension 200 | results index/plotting README |
+| Standardized periodic roles | 480 warm-up + exactly 160 retained boundaries; ≥4 accepted rising crossings or fail closed | Legacy acquisition README |
+| DTW | `legacy_dtw_v1_abs_n_unclipped`; target cycle `[conv_len:2 conv_len]`, state final `conv_len`; lag channel 1; six-channel mean | `utils/metrics.py`, `configs.py` |
+
+DTW circular lag is alignment metadata, not a physical delay. Native reward DTW and target-max-abs-normalized six-sensor cycle DTW answer different questions and are not merged.
+
+### 4.2 Feature and dimensional contracts
+
+Raw sensor order is `(upper_ux, upper_uy, center_ux, center_uy, lower_ux, lower_uy)` and raw force order is `(front_fx, front_fy, upper_fx, upper_fy, lower_fx, lower_fy)`. Historical aliases in the feature builder are `T=upper`, `C=front/centre sensor`, `B=lower`; do not infer force and sensor insertion order from aliases.
+
+Dimensionless conversion is `u_hat=u/U0`, `Cd,Cl=2F/(rho U0^2 D)` with `U0=0.01`, `D=20`, `rho=1` in the code contract. Symmetry features include `u_m,u_a,u_c,v_a,Cd_tot,Cd_rear,Cl_tot,Cl_diff` and rear symmetric/antisymmetric components. Optional temporal/action-history families include lag and `d(alpha)/dt_c`; Illusion target/error features exist in the runtime pool.
+
+`re_code` uses reference length `2D`; therefore `Re_D=re_code/2` and `mu=2/re_code`. Kármán example manifest `re_code=100` records `Re_D=50`, `sample_interval=800`, `control_dt=0.4`, `action_scale=8`, `action_bias=(0,-4,+4)`, FIFO 150, `conv_len=30`. Illusion `target_diameter` is historically passed to Legacy builder as a radius; stored labels must be kept literal.
+
+### 4.3 Reflection/symmetry contract
+
+The native action order is `(front, upper, lower)`. The corrected action map is
+
+\[
+G(\alpha_F,\alpha_U,\alpha_L)=(-\alpha_F,-\alpha_L,-\alpha_U).
+\]
+
+The canonical mapped-shared deployment is
+
+\[
+\alpha_F(x)=\frac{h_F(x)-h_F(Gx)}{2},\qquad
+\alpha_U(x)=h_R(x),\qquad
+\alpha_L(x)=-h_R(Gx).
+\]
+
+This **imposes** exact reflection symmetry at deployment. It does not establish PPO equivariance. `apply_G_raw12` swaps upper/lower sensor and force slots with lift sign reversal while retaining front drag and reversing front lift; `apply_G_target_force` retains drag and reverses lift.
+
+**Formula identity contract.** A formula is identified by immutable artifact/run ID, source SHA-256, fitted-expression hash, deployment-expression hash, feature names/metadata, output metadata, alignment/dt metadata, G metadata, fit/PySR/data provenance, and (for a pair) front/rear source hashes and pair hash. Term deletion is generated from the parent deployment expression and records parent source/deployment hashes plus term identity and scale. Display labels do not replace these identities.
+
+## 5. Canonical Kármán law and deployment mapping
+
+### 5.1 Canonical formula table
+
+| Role | Canonical deployed law | Mathematical meaning in this index | Deployment |
+|---|---|---|---|
+| front | `alpha_F = odd(-0.381391 Cd_rear,a)` | `Cd_rear,a=(Cd_upper−Cd_lower)/2`; odd projection extracts the reflection-odd component, making front action sign-compatible | `alpha_F=[h_F(x)-h_F(Gx)]/2` |
+| upper/rear shared | `alpha_U = 1.307782 Cl_rear,s − 3.431209` | `Cl_rear,s=(Cl_upper+Cl_lower)/2`; feedback from rear-mean lift plus a persistent negative constant in upper action | `alpha_U=h_R(x)` |
+| lower | not separately fit in the canonical pair | reflected copy of upper action, not an independent coefficient set | `alpha_L=-h_R(Gx)` |
+
+`alpha=omega/U0` is dimensionless surface tangential velocity. Thus the rear constant is approximately `−3.431209 U0` for upper and `+3.431209 U0` for lower, with the front stationary in the perfectly symmetric state. The current formula is “mapped-shared topology A”, not a unique/global symbolic law.
+
+### 5.2 Term interpretation and strict boundary
+
+- **OBS:** deterministic Kármán deletion table uses the same 40-step window and legacy DTW parent comparison. Removing the rear constant (`k_rear1`, leaving rear lift feedback) yields the largest listed degradation across re50/100/200/400 (Δ approximately −0.107/−0.094/−0.179/−0.191). Removing rear-lift feedback (`k_rear0`, leaving constant) degrades less (approximately −0.029/−0.042/−0.061/−0.049). Removing tested front feedback gives small mixed effects (approximately −0.0065/+0.0006/−0.0052/−0.0252).
+- **INTERPRETATION:** within this declared deletion window and four named Kármán cases, the persistent rear counter-rotation is the dominant tested element; rear lift feedback is secondary; the tested front feedback is weak.
+- **STRICT BOUNDARY:** this is term ranking under a fixed topology, fixed coefficients, fixed Legacy solver/contract, fixed case set, fixed 40-step window and declared DTW. It is not universal necessity, causal authority, sensitivity over all state space, or proof that the fitted constant is physically sufficient by itself. “Front weak” does not mean front action is identically zero in all deployments; its imposed odd generator can be active away from symmetry.
+
+### 5.3 Steady correspondence
+
+`article2-steady-analysis-20260720` reports a disturbance-free constant rear-rotation sweep `A=0…6`, with upper `−A`, lower `+A`; best listed sweep point `A=5`, similarity `0.9872116`. Canonical law deployment gives `0.9664944`; tail rear mean is `(-3.430589, +3.431828)`, matching fitted magnitude `3.431209` to rounding. This is **OBS** and a bounded **INTERPRETATION**: steady open-loop magnitude calibrates the order of the rear constant. It does not prove momentum balance, one-to-one deficit-to-rotation conversion, causality, or a mechanism.
+
+## 6. Case matrix and evidence ladder
+
+| Objective | Training/discovery | Validation/extension | Comparator/window | Status |
+|---|---|---|---|---|
+| Kármán | re-code 50/100/200/400; 200 PPO rows each; per-case and joint discovery, multiple seeds | L2 40; L3 200; long 400; named re-code 25/70/150/300, 200 each | target reference vs controlled role; native Legacy DTW; fixed formula | active, finite and pointwise |
+| Illusion | trained 0.75L/1L/1.5L; feature pools actual/target/error; discovery/refit alternatives | L2/L3 historical; long 400; named size packages; standardized long role acquisition | target similarity alone is insufficient; compare physical zero | historical/negative |
+| steady | no SR fitting objective; constant rear sweep | A=0…6 and canonical law deployment, 200-step contextual runs | disturbance-free downstream similarity | contextual |
+| Vortex/Lamb/Taylor | archived exploratory route | no current accepted transfer | excluded archive | unsupported |
+| Other cloak transfer | no authoritative SR transfer chain in current package | no pooled evidence with V5/CCD/OID | requires separate solver/contract | gap/unsupported |
+
+**Generalization boundary:** named unseen Kármán points are one realization per point with frozen coefficients. They support pointwise deployment observations, not continuous interpolation law, distributional robustness, uncertainty, or universal Reynolds behavior. Illusion size labels and old short target-similarity values do not establish analogous claims.
+
+## 7. Illusion SR: negative/historical result
+
+The Illusion runtime supports target forces, target sensors, harmonics and error features; Stage 2 search profiles included actual-only, actual-plus-error and actual-plus-target families. Yet selected retained formulas omitted target/error terms. That omission is evidence about the selected historical fit, not proof target information was unnecessary.
+
+The decisive boundary is the standardized long-window Legacy result: after warm-up, retained SR action decayed near physical zero; native and target-normalized SR comparisons were effectively equal to the physical-zero role and the mean field was nearly zero-role. Therefore old 200/400-step target-similarity numbers cannot be called Illusion SR efficacy, target tracking, cross-size generalization, term necessity or mechanism. Illusion topology B is separately rejected by a non-finite 40-step gate at 1L/1.5L; failed crossings/incomplete roles remain diagnostics. PPO/DRL Illusion capability is a separate supported runtime statement.
+
+## 8. Offline fit versus closed loop
+
+Offline action RMSE is evaluated on causally aligned PPO-visited states (197 rows/case). The canonical Kármán aggregate RMSE is 1.4435, 1.0895, 2.2860 and 2.1699 for re50/100/200/400 respectively; these are dimensionless `alpha` diagnostics. They do not measure state visitation, stability, wake matching or target performance. Closed-loop Stage 3 can visit states outside the PPO trajectory and is the acceptance arbiter. This distinction is a central SR lesson and must remain explicit in any later figure/table.
+
+## 9. Figure, data and script inventory
+
+| Item | Path | Role/status |
+|---|---|---|
+| Main executable scripts | `src/SR_analysis/stage_1_infer.py`, `stage_2_fit.py`, `stage_3_validate.py` | active Stage 1→2→3 |
+| Contracts/checks | `src/SR_analysis/checks/`, `utils/`, `tests/` | replay/order/formula/data/metric/provenance gates; CPU-verifiable |
+| Accepted inputs | `src/SR_analysis/data/runs/article-joint-data-karman-20260718/**`; corresponding Illusion family | immutable Stage 1 parents; Illusion historical |
+| Formula/result families | `src/SR_analysis/results/runs/article-*`, `article2-*` | immutable; exact family purpose in `results/catalog.json` |
+| Derived publication package | `src/SR_analysis/results/runs/article2-plotting-package-20260721/**` | five figures, tables, presentation pages; derived, no refit |
+| Key tables | `tables/offline_action_rmse.md`, `tables/term_deletion.md`, `ablation_summary.csv`, `scaling_summary.csv`, `term_contributions_*.csv` | diagnostics/derived views; trace to parents |
+| Long telemetry exports | `src/SR_analysis/results/runs/article2-long-timeseries-csv-20260720/**` | 400-step SR/PPO/target exports and manifests |
+| Standardized collector | `src/drl_pinball/legacy_test/acquire.py` | common Target/PPO/SR/Zero role acquisition; serial Legacy CFD |
+| Legacy runner/interface | `src/drl_pinball/legacy_test/`, `LegacyCelerisLab/` | historical plant compatibility; not V5 |
+| Archive | `src/SR_analysis/archive/**` | excluded/superseded/diagnostic; may not execute |
+
+**Figure interpretation limits:** plotting outputs are derived. The flow-field package selects one exact same-run stable-cycle field by phase; it is a single snapshot, not ensemble/uncertainty or full-field identity. No figure alone upgrades a result to mechanism.
+
+## 10. Artifact/path/hash/result ledger
+
+| Evidence node | Identity/paths | Hash/provenance status | Result/claim status |
+|---|---|---|---|
+| Stage 1 Kármán | `data/runs/article-joint-data-karman-20260718/**` | re100 manifest binds Git `ca8ee5f…`, model SHA `148240c1…`, GPU UUID, config, norm source; other scenes have per-scene manifests | accepted input; four trained cases, 200 steps, 197 causal rows |
+| Stage 1 Illusion | `data/runs/article-joint-data-illusion-20260718/**` | re1L binds model SHA `570fbdb0…`, Legacy target radius contract, Git `ca8ee5f…` | immutable historical input; not active positive claim |
+| Canonical formula | `results/runs/article-refit-karman-topology-a-20260718` | exact formula artifact/source and expression hashes are required; catalog binds run ID | active bounded Kármán law |
+| Mixed article manifest | `results/runs/article-joint-sr-final-20260718/evidence_manifest.json` | contains hashes and historical absolute paths; do not rewrite | authority index only; mixed Illusion wording superseded by current scope |
+| Kármán L2/L3 | `article-L2-karman-20260718`, `article-L3-karman-20260718` | run IDs are immutable; manifests are provenance parents | 40-step screen; 200-step standard closed loop |
+| Kármán long/gen | `article2-long-karman-20260720`, `article2-gen-karman-20260720` | named run identities; derived exports bind input telemetry | finite 400-step duration; named pointwise 200-step extensions |
+| Kármán deletion | `article-ablation-*`, `article-scaling-L1-v2-20260718`, plotting `term_deletion.md` | variants retain parent formula/deployment hashes and term identities | bounded term ranking only |
+| Steady | `article2-steady-analysis-20260720/**` | derived JSON/CSV; interpretation states source families | contextual magnitude calibration |
+| Illusion long | `article2-long-illusion-20260720/**`, standardized `legacy_test` reproduction mapping | long manifests bind 400 rows; exact standardized storage/hash must be checked in situ | negative near-zero/zero-equivalent interpretation |
+| Failure archive | rejected topology B, crossing-gate failures, incomplete campaigns, `src/SR_analysis/archive/**` | retained; no-clobber/immutable rules | diagnostic and boundary evidence |
+
+**Hash ledger limitation:** the checked-in human index and catalog bind many IDs, but this dossier did not recompute every SHA-256 or enumerate every external reproduction-storage object. The exact hashes are therefore to be read from each manifest/catalog at audit time, not inferred from truncated entries above.
+
+## 11. Transfer and adjacent-method boundary
+
+- **Steady:** contextual calibration only (Section 5.3).
+- **Vortex/Lamb/Taylor:** explicitly excluded archive route; no current SR transfer claim.
+- **Other cloak scenes / modern V5:** no silent pooling. A transfer result would require its own solver, geometry, normalization, action, Re, case/seed, metric/window and immutable artifact chain.
+- **DRL:** supplies frozen PPO roles/trajectory source and baseline interface; this dossier does not reproduce DRL training or claim DRL generalization.
+- **CCD:** may supply descriptive cross-analysis only; it does not validate SR causality or mechanism.
+- **SINDy/tree-search:** future work may compare libraries, sparsity/topology recurrence, derivative/state definitions and closed-loop selection under the same contracts. It must not silently refit or overwrite the canonical formula, mix alignment conventions, or treat an offline sparse law as accepted without Legacy CFD. A fair comparison needs predeclared feature/library, split, coefficient freeze, formula identity, safety gate, closed-loop metric and baseline.
+
+## 12. Gaps and prohibited claims
+
+### Must re-check before manuscript use
+
+1. The round README exists at `docs/JFM_WYQ/ROUND1_BOTTOM_UP/README.md` and is the round-specific authority for requirements; preserve its evidence-index and non-manuscript constraints when this dossier is condensed.
+2. Recompute and record full SHA-256 values for canonical formula files, formula pair, all parent/deletion variants, manifests, and external standardized role outputs.
+3. Resolve storage mapping and enumerate August Target/PPO/SR/Zero role manifests, including exact standardized Illusion failure/near-zero artifacts.
+4. Verify exact canonical formula file paths and formula-pair hash; the current catalog gives the immutable run ID and displayed law but not every leaf filename.
+5. Reconcile all `article-*`/`article2-*` consumers and absolute historical paths before any relocation or figure regeneration.
+6. Audit the complete Kármán 200/400 metrics by scene, comparator, window and realization; this dossier intentionally does not reconstruct unexposed tables.
+7. If a physical mechanism claim is later needed, consult existing Undermind literature notes/PDFs and label literature evidence separately; do not use it to upgrade local SR observations.
+
+### Explicitly unsupported here
+
+No claim of universal Reynolds/size behavior, distribution-wide robustness, uncertainty, unique/global optimality, term necessity outside the declared deletion window, PPO equivariance, causal wake mechanism, momentum equality, physical DTW delay, Illusion tracking/efficacy/generalization, Vortex/other cloak transfer, or modern-V5 substitution is authorized by this dossier.
+
+## 13. Verification status
+
+- **Existing derivation:** equations/contracts as implemented and stated by current SR authority/code.
+- **Independent check:** this dossier cross-checked order, `Re_D=re_code/2`, dimensional conversion, `G` action/raw mappings, canonical deployment algebra, DTW window semantics, deletion arithmetic and steady constant correspondence by source inspection and arithmetic reading. No CFD and no new scientific conjecture were performed.
+- **New conjecture:** none.
+
+Recommended non-CFD follow-up: run focused SR CPU contract tests and `git diff --check` after this Markdown is reviewed. Do not launch CFD solely to verify documentation.
+
+## 14. Bottom-up handoff summary
+
+The index supports one compact active result: a hash/provenance-bound, symmetry-mapped Legacy Kármán policy surrogate whose tested closed-loop evidence is consistent with dominant persistent rear counter-rotation, secondary rear-lift feedback and weak tested front feedback over a named 40-step deletion window. Steady provides magnitude context; named unseen cases and 400-step runs remain finite/pointwise; Illusion SR is a negative historical route because long-window behavior collapses toward physical zero and does not establish target efficacy. The principal remaining uncertainty is not formula syntax but completeness of leaf-level hash/role ledgers and exact round/storage reconciliation.

docs/JFM_WYQ/ROUND1_BOTTOM_UP/03_FIELD_CCD_OID.md

@@ -0,0 +1,436 @@
+# Round 1 bottom-up evidence index — field / corrected CCD / OID
+
+> 状态：evidence index / dossier，非 manuscript prose。编制范围：`src/CCD_analysis/**`、`src/OID_analysis/**`，并索引其直接依赖的 Legacy acquisition、`q_target/q_blk/q_ctl`、constant-mean、mask/ROI/phase/action artifacts，以及与 SR、steady theory 的受限接口。
+>
+> 事实权威：package authority 文档、可执行代码/测试、canonical manifest/artifact。Nowledge Mem 只作定位线索；Undermind 只检查已有 DynamisLab workspace 的方法文献/笔记，不启动 deep search，不以文献替代本地证据。
+
+## 0. 读取结论与 fail-closed 总览
+
+| 分析线 | 当前状态 | 本轮允许的科学层级 | 明确禁止升级 |
+|---|---|---|---|
+| corrected Kármán CCD | current / completed / bounded | inclusive wake ROI 均值估计量；DRL 相对 constant phase template 的 phase-coherent descriptive co-variation；稳定 rank-3 子空间；CCD/POD 子空间近等价 | causal response、paired counterfactual、mechanism、response time、explained variance、CCD 优于 POD、独立 realization uncertainty、global stability |
+| OID active package | active method core / claim-free / CPU-only | 加权 snapshot POD、Schlegel LR/LE map、generated-range/right-inverse 数学、alignment/artifact contracts | active scene result、action importance、control authority、controllability、causal/mechanistic interpretation |
+| July OID study | archive / conditional historical | 可报告 provenance、合同演进、传感器 gate 通过与 force gate 失败的历史事实 | 任何 importance 排名、force positive result、action result、paper conclusion、`COMPLETE` |
+
+**总边界：** CCD 只能描述共变，不给出 causality/response/mechanism。OID 当前不承载正面科学结果。archive 只用于解释撤回路线和失败原因，不能复活结果。
+
+## 1. Sources inspected and authority map
+
+### 1.1 Repository authority
+
+**CCD current reader entry**
+
+- `src/CCD_analysis/README.md`
+- `RESULTS_INDEX.md`：number → artifact → figure lookup
+- `MATHEMATICAL_DERIVATION.md`、`karman_dynamic/CORRECTED_MATH_CONTRACT.md`
+- `CLAIMS.md`、`FINAL_RESULTS.md`、`WRITING_HANDOFF.md`
+- `FIGURES_AND_DATA.md`
+- `EXECUTION_CHECKPOINT.json`、`artifact_catalog.json`
+- code: `karman_dynamic/cycle_template_ccd.py`, `domain_comparison.py`, `mode_diagnostics.py`, `pod_baseline.py`, `phase.py`, `contracts.py`, `artifacts.py`; acquisition code under `acquisition/`; tests under `tests/`.
+
+**OID current method entry**
+
+- `src/OID_analysis/README.md`
+- `METHOD.md`
+- `SCIENTIFIC_RESET.md`, `SCIENTIFIC_RESET_CHANGESET.json`
+- `PHYSICS_CONTRACT.md`, `HANDOFF.md`
+- active implementation: `v2/analysis.py`, `v2/alignment.py`, `v2/artifacts.py`
+- active tests: `tests/test_schlegel_oid.py`, `test_alignment_artifacts.py`, `test_physics_contract.py`.
+
+**OID historical/conditional source**
+
+- `archive/studies/2026-07-two-scene-conditional/STATUS.md`
+- its `presentation/docs/PIPELINE.md`, `RESULTS.md`, source scripts, audits, freeze and migration manifests
+- `archive/legacy_oid_v1/` and `archive/v2_development_code/`: provenance only.
+
+### 1.2 Machine canonical roots and publication sets
+
+**CCD canonical root:**
+`/home/frank14f/optane/DynamisLab/ccd/karman-dynamic/canonical/`
+
+Canonical entries from `src/CCD_analysis/artifact_catalog.json`:
+
+| role | path | schema | manifest SHA-256 | status |
+|---|---|---|---|---|
+| ROI mean | `.../results/roi-mean` | `ccd-karman-roi-mean/v2` | `f7f6141042678e52314a4d8f76d589108da8e844c8ce504f8f523a6f88a17e73` | current authority |
+| corrected wake CCD | `.../results/cycle-template-ccd-review-fix-per-cycle-20260808T1736+0800` | `ccd-karman-cycle-template/v1` | `b114fbcc5f821034829ca3771ac5ba13df4b75df312b0615b662f81a9e5737d8` | current authority |
+| preferred pinball-inclusive CCD/POD | `.../results/pinball-domain-20260809T1925+0800` | `ccd-karman-pinball-domain/v1` | `7e107a895364ed3e34abf2be513a3ce9e12736766ec6383c5b831891802d5828` | preferred derived result |
+| preferred reader publication | `.../publication-pinball-domain-20260809T1925+0800` | `ccd-karman-pinball-publication/v1` | `31888f8eeb6c70dd9f0ff3449fa879f44abf381338ec261b0e45f69603691fdd` | reader view |
+| predecessor | `.../results/pinball-domain-20260809T1651+0800` | `ccd-karman-pinball-domain/v1` | `8b952d5dd8cf9813a241da31295497e5c79db6df7dc4ed12b95a6706f99e37b6` | superseded; retain, do not promote |
+
+The corrected primary publication transaction is `.../publication-review-fix-per-cycle-20260808T1736+0800`, bound to `roi-mean`, corrected cycle-template result, and per-cycle DRL/constant phase products. Its review-fix result manifest is the above `b114...`; `EXECUTION_CHECKPOINT.json` also records publication manifest SHA `a6f0c55515b0e4a3b8980174988bf6bbea3617d60f8648aab25a7ffa0a2a24cd` for the validated review-fix publication. Do not collapse the primary wake-only result and its pinball-inclusive derived sibling into one artifact.
+
+**Repository reader view:** `src/CCD_analysis/data/karman-dynamic/canonical/publication-pinball-domain-20260809T1925+0800/`; dense source remains external. PNG/PDF are presentation assets; `plot_data.npz/json` is the redraw interface, not replacement authority.
+
+**OID canonical/freeze identities:**
+
+- Final historical audit: `runs/active-two-scene-v2/OID_PHASE4_AUDIT_20260723T1510Z.json`, SHA-256 `9358105c7199793a2fd810885044295350da165a0f8e9026734ca1a51fc52ed8`.
+- Conditional freeze: `provenance/freeze/OID_CONDITIONAL_FREEZE_20260723T173300Z.json`; this is an inventory, not strict `COMPLETE`.
+- Relocation/migration binding: `lineage/CURRENT_STATE_MIGRATION_MANIFEST.json`; `archive/RELOCATION.md` and `archive/RUNS_RELOCATION_MANIFEST.json` govern external payload recovery. The reset moved 516 historical files; no historical payload is thereby active.
+
+## 2. Estimands, field variables, roles, and chains
+
+### 2.1 General three-field notation
+
+For the Legacy correction-field vocabulary:
+
+\[
+q_{in}=\text{no-pinball inlet/reference field},\quad
+q_{blk}=\text{passive pinball field},\quad
+q_{ctl}=\text{controlled pinball field}.
+\]
+
+\[
+\Delta q_{blk}=q_{blk}-q_{in},\quad
+\Delta q_{ctl}=q_{ctl}-q_{blk},\quad
+\Delta q_{tot}=q_{ctl}-q_{in}=\Delta q_{blk}+\Delta q_{ctl}.
+\]
+
+This decomposition is an estimand contract, not a causal decomposition. In OID, `q_ctl-q_blk` is a same-parent paired closed-loop branch difference. Feedback alters future state and action together; it is not an isolated commanded-action intervention.
+
+For corrected CCD the primary fields are role means and phase snapshots from `q_target`, `q_zero`, `q_constant_mean`, and `q_DRL`; the corrected dynamic field is not the old generic `dq_ctl` route. Constant and DRL cycles are not paired.
+
+### 2.2 Field vector and mask
+
+- Nondimensional Cartesian coordinates `x_D,y_D`; persisted velocity is already `q/U0`, so no second division by `U0`.
+- For selected fluid points `K`, `M_p=|K|`; component-major vector:
+  \[
+  q=[u_x(k_1),\ldots,u_x(k_{M_p}),u_y(k_1),\ldots,u_y(k_{M_p})]^T,
+  \quad M=2M_p.
+  \]
+- Weight is coordinate quadrature `w_xw_y`; duplicated for both velocity components: `W=diag(w_1,...,w_Mp,w_1,...,w_Mp)`.
+- CCD wake-only: ROI `34<=x/D<=54`, `|y/D|<=5`, exact intersection of the four solver-fluid masks, `M_p=80,200`, `M=160,400`.
+- CCD pinball-inclusive sibling: `29<=x/D<=54`, `|y/D|<=5`, `99,264` common fluid points, `M=198,528`.
+- Plot array convention: stored `(x-index,y-index)`; plotting uses transpose; front/upper/lower body order; fixed regions geometry `29–32D`, near wake `32–40D`, middle `40–47D`, far wake `47–54D`.
+
+### 2.3 Mean-field estimands
+
+For role `r`, coordinate-weighted target error:
+
+\[
+E_r=\left[\frac{\sum_{k\in K}w_k\|\bar q_r(k)-q_T(k)\|_2^2}{\sum_{k\in K}w_k}\right]^{1/2}.
+\]
+
+The exact mean identity is:
+
+\[
+\bar q_D-q_T=(\bar q_0-q_T)+(\bar q_C-\bar q_0)+(\bar q_D-\bar q_C).
+\]
+
+The reported scalar error reductions are differences in `E`, not energy reductions or causal effects.
+
+**Canonical corrected Kármán ROI numbers** (`roi-mean` artifact; four-role common mask):
+
+| role | `E_r` |
+|---|---:|
+| zero | `0.4694352273` |
+| constant_mean | `0.1110140739` |
+| DRL | `0.0857787860` |
+
+Derived additive increments:
+
+- zero→constant mean: `0.3584211534`;
+- constant mean→DRL: `0.0252352879`;
+- zero→DRL total: `0.3836564413`;
+- mean share `0.3584211534/0.3836564413 = 93.42%`;
+- remaining dynamic increment share `6.58%`.
+
+**Allowed wording strength:** within this exact Legacy Kármán case, ROI, common mask, mean estimator and window, constant mean accounts for most of the measured zero-to-DRL target-error reduction; the DRL dynamics add a smaller measured increment. This does not identify the responsible actuator, prove downstream positive-velocity generation, prove deficit compensation as a mechanism, or establish that feedback is unnecessary. “Positive downstream velocity” can be used only as a bounded description of a plotted/field difference if the exact field artifact, sign convention, ROI and comparator are stated. “Deficit compensation” is at most a candidate physical interpretation consistent with the mean-field pattern and SR/steady context—not a CCD/OID-proven mechanism. Do not write “the controller restores momentum” or “causes compensation”.
+
+### 2.4 Phase/dynamic estimand
+
+There are 19 retained DRL cycles and 10 bins (`N=190`), with sample index `j=10c+b`. Constant template:
+
+\[
+\hat q^C_b=\frac1{19}\sum_cq^C_{c,b}.
+\]
+
+Each role is centered over its own retained ensemble:
+
+\[
+U_{c,b}=(q^D_{c,b}-\bar q_D)-(\hat q^C_b-\bar q_C).
+\]
+
+`U` is a phase-conditioned reference residual. It is not paired DRL/constant counterfactual evidence; no interpolation, lag, wrap, temporal PSD, or cross-role cycle pairing is used.
+
+Action observable: same-boundary effective action in native order `(front, upper, lower)`; exact retained DRL physical action mean removed, followed by empirical centering. Requested normalized action, requested physical action, EMA-effective action and field-time effective action must not be silently interchanged.
+
+## 3. Corrected CCD operator, modes, symmetry, phase
+
+### 3.1 Operator and dimensions
+
+With field snapshots as columns `U∈R^(M×N)` and action observables `P∈R^(3Q×N)`, `Q=1`:
+
+\[
+A=\frac{P_c(W^{1/2}U_c)^T}{N\sqrt{3Q}},\quad A=L\Sigma V^T.
+\]
+
+No whitening, standardization, lag, wrap, field-energy normalization, or POD preprojection. Physical modes `Φ=W^{-1/2}V`, so `Φ^TWΦ=I`; coefficient `c_ij=φ_i^TWU_j`. Singular values are cross-correlation strengths, not field energy, explained variance or reconstruction fractions.
+
+### 3.2 Current numbers
+
+Primary wake-only result (`N=190,Q=1,M=160400`):
+
+- singular strengths: `0.0174141920`, `0.0069694895`, `0.0010500356`;
+- numerical rank: 3;
+- least-favourable retained-boundary bin/origin check: rank-3 projector cosine `0.9947903592`;
+- largest relative singular-value shift `0.0766020533`;
+- leave-one-cycle-out leading projector cosine `0.9997247674`;
+- alternatives: 8 bins spectrum `[0.0172112894,0.0067331096,0.0009696007]`, cosine `0.9947903592`; 12 bins `[0.0170742270,0.0071347287,0.0011180038]`, cosine `0.9977713810`; half-shift 10 bins `[0.0175962424,0.0068663334,0.0010454157]`, cosine `0.9982476661`.
+
+Only the rank-3 subspace is stability-supported; individual vectors can flip sign and rotate within near-degenerate subspaces. A global sign flip `(φ_i,c_i)→(-φ_i,-c_i)` is identical.
+
+### 3.3 Pinball-inclusive CCD/POD comparison
+
+Preferred derived result keeps the same `19×10` samples, action coordinates, centering and `Q=1`, but extends domain to include bodies (`29–54D`). Full-fit weighted POD uses `G=U_c^TWU_c` and is not CCD-guided. Rank-3 principal cosines between CCD and raw POD field subspaces:
+
+`[0.9999616609, 0.9998971005, 0.9976439804]`; declared minimum bin/cycle stability cosine `0.9974651867`.
+
+Interpretation: CCD and raw POD span essentially the same field subspace in this tested object/domain. CCD’s bounded added information is action-coordinate association (`front`, `rear-symmetric`, `rear-antisymmetric`), not a distinct field decomposition and not evidence that CCD is superior to POD. Procrustes-aligned POD is visual registration only.
+
+### 3.4 Three action-associated modes
+
+- Left observable coordinates are native `front`, `upper`, `lower`; publication re-expresses them as `front`, `rear-symmetric`, `rear-antisymmetric`.
+- Mode 1 and 2 are predominantly antisymmetric (`0.972`, `0.930` diagnostic parity measures); mode 3 predominantly symmetric (`0.833`).
+- Phase reconstructions use bins nearest `0,π/2,π,3π/2`; harmonic amplitudes/phases are circular diagnostics over each cycle/bin array, not temporal PSD frequencies. Phase maps are registered coordinate partitions, not discovered wake boundaries.
+- Figures/data: `01_domain_estimand`, `02_ccd_mode_identity`, `03_phase_reconstruction`, `04_raw_pod_vs_ccd`; source fields include `ccd_mode_velocity`, `ccd_mode_rotation_proxy`, `left_action_physical_coordinates`, representative phase arrays, `plot_data.npz/json`.
+- Body-local rotation proxy is a Gaussian localized signed angular-momentum-like visualization; it is not wall vorticity. Complete-stencil solver-fluid vorticity is auxiliary.
+
+**Candidate interpretation (not established):** symmetry-separated rear action coordinates may organize the antisymmetric/symmetric field components, while phase-resolved reconstructions may be compatible with a periodic correction pattern. This is a descriptive organizational hypothesis only; no lag/phase offset is a response time or causal path.
+
+## 4. Acquisition and provenance contracts
+
+### 4.1 Legacy acquisition boundary
+
+Corrected CCD source chain is LegacyCelerisLab. The package contract records case `karman_re100`, `Re_code=100` based on `2D`, physical `Re_D=50`, `U0=.01`, `nu=.004`, body order and geometry, source/config/model bindings, decoder identity, masks, coordinates, timeline, DDF/FIFO state, action order/units, and hashes. Legacy and modern CelerisLab V5 are separate chains; no V5 replay may replace the Legacy CCD evidence.
+
+`q_target`, `q_blk`, `q_ctl` are role contracts. `q_target` is a desired/reference trajectory with case-specific target bodies, not same-checkpoint counterfactual. `q_blk` is passive zero-action pinball; `q_ctl` is frozen-policy controlled branch. Illusion `q_ctl` requires strict +11D history/replay compatibility before production; the CPU compatibility certificate is runtime compatibility only, not accuracy/stability/phase/mechanism validation.
+
+Acquisition must fail closed on missing/moved roots, duplicate/non-increasing times, nonfinite arrays, mismatched schema/hash/coordinates/masks, incomplete exact joins, stale telemetry or wrong action lineage. No silent trim, nearest-time join, interpolation, timestamp tolerance or block wrap.
+
+The active acquisition contract passed static CPU checks, but published/smoke status must be read from current evidence; rejected production paths remain rejected. No CFD/CUDA was run for this index.
+
+### 4.2 q roles and direct-dq distinction
+
+OID canonical products use same-parent branches per realization:
+
+1. `controlled` = PPO `q_ctl`;
+2. `zero_baseline` = zero-action `q_blk` used for subtraction;
+3. `baseline_twin` = independent numerical twin used only for reproducibility gate.
+
+Exact pairing requires equal `sample_ids`, lattice steps and times; field clock is distinct from control clock. Historical canonical crop is `x=360:1000`, all y, stride 8; reduced shape `80×64`; C-order `ux` then `uy`; exact Legacy solid rasterization; uniform cell quadrature, zero solid weights; Kármán fluid/solid counts `5025/95`, Illusion `5024/96`. These are historical OID product contracts, not current active scientific claims.
+
+### 4.3 Phase/action artifacts
+
+For CCD, phase artifacts are the immutable `phase/{drl,constant_mean,target,zero-failure}` families and review-fix siblings listed in `EXECUTION_CHECKPOINT.json`. The primary zero phase product is `FAILED_CLOSED`; it must not be used as a positive role. Corrected CCD uses retained independent phase alternatives rather than Fourier-resampling the primary field.
+
+For OID, exact/delayed pairing is implemented as `b(t+τ)=C(τ)a(t)` only where the future sample exists in the same declared segment; no padding, nearest match, wraparound or cross-segment pairing. Circular shifts occur only inside the explicitly labelled null. Delay is a declared analysis parameter, not physical causality.
+
+## 5. OID method core: definitions, valid mathematics, reset
+
+### 5.1 Active mathematical objects
+
+For centered state coefficients `a` and observable POD coordinates `b`, forward map is fitted regression:
+
+\[
+C=Σ_{ba}Σ_{aa}^{+},\qquad \hat b=Ca.
+\]
+
+This is not raw cross-covariance. LR inverse:
+
+\[
+L_R=Σ_{aa}C^T(CΣ_{aa}C^T)^+,
+\]
+
+is the minimum empirical state-reconstruction MSE among exact linear inverses on the generated range. LE inverse:
+
+\[
+L_E=G^{-1}C^T(CG^{-1}C^T)^+,
+\]
+
+minimizes declared positive-definite state-field `G`-norm among exact generated-range inverses. `G` is a project metric, not actuator energy, control cost, controllability Gramian, or Navier–Stokes energy principle.
+
+Valid standard mathematics: weighted POD orthogonality; method of snapshots; least-squares normal equations; right-inverse identities on `range(C)`; sign-invariant MAC/principal angles/projector similarity; exact/delayed pairing; atomic artifact validation. Existing derivation / method theorem is distinct from independent verification in tests; no new physical derivation was performed here.
+
+Literature method map, existing Undermind notes and active `METHOD.md`: Schlegel et al. JFM 697 (2012), DOI `10.1017/jfm.2012.70` for LR/LE construction; Sirovich (1987) for snapshot POD; Adrian (1994) for stochastic estimation; Borée (2003) for EPOD/LSE relation; Everson & Sirovich (1995) for gappy POD boundary; Theiler et al. (1992) for surrogate-data context. These references do not license local importance, mechanism or causality claims.
+
+### 5.2 Why OID was reset
+
+The July two-scene line was reset because:
+
+- sensor observables are local velocity functionals inside the state domain, so high sensor R² is partly self-observation;
+- force is an integrated surface projection with different dimension, symmetry, noise and conditioning; force jobs failed their all-fold gate;
+- action is generated inside feedback, so delayed association does not isolate commanded action from state-dependent policy response;
+- `q_ctl-q_blk` is a closed-loop branch difference, not autonomous perturbation/tangent state or causal intervention;
+- legacy products called PLS/cross-covariance SVD directions OID without fitted Schlegel propagator and distinct LR/LE inverses;
+- earlier contiguous holdout was not independent-realization validation;
+- old observables used absolute controlled telemetry rather than exact paired controlled-minus-blocked deltas;
+- field telemetry was once read at control boundaries, yielding invalid zero intermediate values; corrected lineage requires field-time observation;
+- index/replay resemblance did not establish exact same-parent alignment; canonical products now require exact IDs/steps/times and a separate numerical twin;
+- archived reduced mask was defective and required exact Legacy rasterization rebuild;
+- mixed numerical defaults, fixed-selection controls, and “phase surrogate” wording understated conditioning/selection effects;
+- historical `COMPLETE` cannot be reconstructed from relocated artifacts, so final audit remains conditional.
+
+**Disposition:** all old runners, plots, analyses, freeze builders, contradiction evidence and payloads remain archive/provenance. They are not active instructions and must not be imported or executed as current route.
+
+### 5.3 What remains reusable
+
+Reusable definitions/contracts: weighted snapshot POD; diagonal spatial quadrature and exact solid-zero masks; field/observable metric declarations; LR/LE generalized inverse equations; generated-range restriction; sign-invariant mode/subspace comparisons; same-parent exact pairing; delayed sample semantics; circular-shift null with explicit exclusions; atomic no-clobber artifact/hash validation; physics coordinate/body/action oracle. Reusable numbers are only historical audit facts with conditional status, never positive mechanism results.
+
+## 6. Historical OID ledger (reportable only as negative/conditional)
+
+Final audited conditional package: `runs/active-two-scene-v2/OID_PHASE4_AUDIT_20260723T1510Z.json`.
+
+| scene / observable | fold values (observed R²; p; gate) | status and allowed use |
+|---|---|---|
+| `karman_re100/delta_sensors` | r1 `.964749; .001; pass`; r2 `.967141; .001; pass`; r3 `.973321; .001; pass` | conditional historical association only for three declared realizations; self-observation caveat |
+| `karman_re100/delta_forces` | r1 `.836206; .001; fail`; r2 `.867670; .001; pass`; r3 `.872087; .001; pass` | not promoted because all-fold gate fails; descriptive/negative-control record only |
+| `illusion_1.0L/delta_sensors` | r1 `.936949; .001; pass`; r2 `.928960; .001; pass`; r3 `.939764; .001; pass` | conditional historical association only; not importance or mechanism |
+| `illusion_1.0L/delta_forces` | r1 `.701865; .001; pass`; r2 `.694162; .001; pass`; r3 `.480277; .001; fail` | not promoted; descriptive/negative-control record only |
+
+Other audit facts: 363 Kármán field samples at 317-step spacing (control interval 800); 642 Illusion samples at 257-step spacing (control interval 600); candidates ranks `(4,8,12)`, lags `(0,1,2)`; 999 periodic-excluded circular shifts/fold; fixed `rcond=1e-12`; `H=I`; no regularization; 3 LORO folds. Only three declared realizations exist per scene; this is not broad stochastic independence. No field-mode figure is promoted.
+
+Do **not** put the historical positive sensor values into a paper as “sensor importance”, and do not report the force rows as positive force-OID evidence. The active package has no active LR/LE action result.
+
+## 7. Figure / data / script inventory
+
+### 7.1 Current CCD figures and data
+
+Preferred reader-facing four groups under `src/CCD_analysis/data/karman-dynamic/canonical/publication-pinball-domain-20260809T1925+0800/`:
+
+1. `01_domain_estimand`: domain, bodies, partitions, masks, residual definition;
+2. `02_ccd_mode_identity`: `u_x/u_y`, action coordinates, descriptive body-local rotation proxy;
+3. `03_phase_reconstruction`: raw, CCD rank-2/rank-3, POD rank-3 phase-bin means;
+4. `04_raw_pod_vs_ccd`: CCD/POD modes, reconstruction curves, principal cosines.
+
+Machine redraw fields: `x_D(1280)`, `y_D(512)`, `domain_mask(1280,512)`, `point_weights(99264)`, `ccd_mode_velocity(3,2,1280,512)`, `ccd_mode_rotation_proxy(3,1280,512)`, `ccd_modes(198528,3)`, `ccd_left_functions(3,3)`, `left_action_physical_coordinates(3,3)`, representative phase bins. Do not digitize PNG; preserve source manifest hash, mask, orientation, limits and plotting parameters.
+
+### 7.2 OID current scripts/tests
+
+Active `v2` code is method-only: `analysis.py` (POD, forward regression, LR/LE), `alignment.py` (exact/delayed pairing), `artifacts.py` (atomic/hash validation). Tests are synthetic/source-oracle tests for metrics, rank deficiency, zero weights, exact pairing, artifact integrity, and body/action physics. There is no active scene runner, CUDA path, plotting path, audit publisher or redo command.
+
+Historical source scripts include `run_phase3_5_formal.py`, `rebuild_phase3_canonical.py`, `plot_phase4_summary.py`, audit/freeze builders and archived presentation scripts. Inventory is provenance/navigation only; do not execute as active analysis.
+
+### 7.3 Archive figure/data disposition
+
+Historical OID summary PNG/PDF and NPZ/JSON models are useful to reconstruct what was tried and why it was downgraded. They are not current figures. The relocation manifest and SHA-256 inventory must be checked before any staging restore; restore only into an empty staging destination; never restore over compact active tree; never use archived defective mask as final product.
+
+## 8. Conflict and gap list
+
+1. **Corrected CCD source/action sign conflict:** `PHYSICS_CONTRACT.md` records an archived OID tuple `[0,-5.1,+5.1]` versus stated CCD `[0,+5.1,-5.1]`; current source/doc history also contains disagreement. Treat as unresolved source/documentation conflict, not corroboration. Do not infer sign from labels.
+2. **Old correction-field CCD versus current corrected CCD:** Undermind/repository July notes contain old `dq_ctl/dq_tar`, delayed/reduced CCD and positive mechanism language. Current August package explicitly supersedes that route; no old overlap/R² is current CCD evidence.
+3. **Phase-domain CCD history:** old temporal/phase publication is exploratory/descriptive and downgraded; current corrected cycle-template artifact is authority. Fourier-resampled sensitivities are forbidden for current stability.
+4. **OID status contradiction:** archived `RESULTS.md` contains conditional promoted sensor scores, while active README/reset says no active scientific result. Resolve by status precedence: archive facts may be indexed as conditional historical; no active claim.
+5. **OID `COMPLETE` gap:** relocated payload and conditional freeze bind artifacts but do not reconstruct strict-loader `COMPLETE`; no promotion.
+6. **Zero phase gap:** CCD execution checkpoint marks canonical zero phase product `FAILED_CLOSED`; do not treat it as a successful current role.
+7. **Independent-realization gap:** CCD has 19 cycles within one retained DRL lineage and constant ensemble template; OID has three declared realizations but not broad stochastic population. Neither supports general uncertainty.
+8. **Causal intervention gap:** no matched randomized actuator intervention, open-loop perturbation, or branch design isolates action effect from state-policy feedback.
+9. **Wall/mechanism resolution gap:** body-local proxy is not wall vorticity; available field resolution/diagnostics do not close boundary vorticity, pressure, viscous momentum or energy mechanism.
+10. **Cross-chain gap:** Legacy CCD/OID/SR and modern V5/steady contracts differ; no silent merge or metric transfer.
+
+## 9. Cross-map: SR constant component, steady theory, and non-crossable links
+
+### 9.1 Comparable at bounded level
+
+- Both corrected CCD mean analysis and Legacy SR/steady context point to a persistent rear counter-rotation / near-constant component as a candidate backbone. CCD can say the constant-mean arm dominates the tested ROI error reduction; SR can say the fitted rear constant is the dominant tested term in its deletion window; steady can say an accepted constant-rotation endpoint has low `E_inf_vector` under its own uniform-inlet/free-slip `Re_D=50` contract.
+- These are compatible observations about a recurring constant component, not a matched intervention or equality of physical amplitude.
+- A future same-plant, same-role, same-mask comparison could test whether the mean CCD increment spatially resembles SR/steady correction fields. Current package does not establish that mapping.
+
+### 9.2 Explicitly non-comparable / cannot cross-chain
+
+- CCD ROI mean error `E` versus steady cross-section `E_inf_vector` at `x/D=10`: different domain, norm, target/inlet reference and solver/plant contract.
+- Legacy SR DTW versus CCD mean-field error or CCD singular strength: different observables, windows and estimands; DTW lag is not physical delay.
+- Legacy SR action formula versus CCD left action function: former is an `obs→act` symbolic deployment interface; latter is action-coordinate association with a residual field; neither proves the other’s terms or necessity.
+- Steady `q_ctl` versus `q_in` endpoint versus OID `q_ctl-q_blk`: steady explicitly treats `q_blk` as non-ranking; OID requires paired same-parent subtraction. Do not subtract `q_blk` again in steady.
+- Steady accepted branch `[0,+Omega,-Omega]` and historical opposite-sign OID/CCD tuples are not interchangeable. Current sign oracle and body order win over prose labels.
+- Modern V5 compact policies/results cannot replace Legacy acquisition underlying current CCD/SR evidence.
+
+## 10. Candidate physical logic (hypothesis ledger, not conclusion)
+
+H1. A persistent rear-rotation component may alter the pinball-scale mean wake in a way consistent with reduced target error / downstream deficit compensation. Current evidence supports only “consistent with” under exact comparator; matched intervention and budget closure are missing.
+
+H2. A smaller dynamic correction may be phase-coherent and action-associated while living in a field subspace that is also energy-dominant POD. The CCD result supports the descriptive algebra; whether it corresponds to a physical feedback response remains open.
+
+H3. Symmetry decomposition may separate rear-symmetric and rear-antisymmetric action coordinates from predominantly symmetric/antisymmetric field structures. Current parity numbers are diagnostics and basis-dependent; subspace tests are safer.
+
+H4. A possible organizational chain is observation → action → near-body correction → body-connected wake → downstream signature. SR constrains only observation→action; CCD/OID would need interventional, time-aligned and independent evidence to test later arrows. This chain is a future test design, not a finding.
+
+H5. Steady low-deficit endpoint and Legacy constant component may share a base-bleed/gap-jet family at the level of qualitative physical framing. Potential circulation cancellation failed; no NS mechanism closure or cross-solver parity exists.
+
+## 11. Candidate figures / data / script needs for later rounds
+
+**Reusable current CCD panels:** four preferred groups in §7.1, with source hashes and plot-data contract. The panels are reader-facing derivatives, not new CFD evidence.
+
+**Not reusable as positive panels:** old correction-field panorama, old force/signature OID panels, historical Illusion positive figures, any archived mode plot labelled importance/causality, and any figure without exact source manifest/mask/timeline identity.
+
+**Future evidence needed before stronger claims:**
+
+- matched constant/action interventions with same parent and randomized/controlled action;
+- independent DRL parents or genuinely independent cycle blocks with declared uncertainty;
+- phase and clock parity across roles, exact field-time action history;
+- mask/ROI sensitivity and grid/solver parity;
+- wall-resolved vorticity/pressure/shear or closed control-volume momentum/energy budgets;
+- OID sensor-excluded/upstream-only states, explicit measurement-operator baseline, force symmetry coordinates, independent parents and action interventions;
+- if OID becomes active again, a new study ID/run IDs and complete source/config/model/checkpoint/FIFO/DDF/clock/action/geometry/metric/hash manifest.
+
+## 12. Number / formula / artifact ledgers
+
+### 12.1 Number ledger
+
+| ID | number | definition / scope | authority | status |
+|---|---:|---|---|---|
+| N01 | 0.4694352273 | zero ROI mean target error | CCD `roi-mean` | current |
+| N02 | 0.1110140739 | constant_mean ROI mean target error | CCD `roi-mean` | current |
+| N03 | 0.0857787860 | DRL ROI mean target error | CCD `roi-mean` | current |
+| N04 | 93.42% | share of zero→DRL error reduction from zero→constant | derived exact scalar ledger | bounded current |
+| N05 | 6.58% | constant→DRL dynamic increment share | derived exact scalar ledger | bounded current |
+| N06 | 190 | CCD samples = 19 cycles × 10 bins | CCD manifest | current |
+| N07 | 160400 | wake-only CCD field dimension | CCD manifest | current |
+| N08 | 0.0174141920, 0.0069694895, 0.0010500356 | primary CCD strengths | CCD manifest | current descriptive |
+| N09 | 0.9947903592 | least-favourable rank-3 projector cosine | CCD stability artifact | bounded current |
+| N10 | 0.9999616609, 0.9998971005, 0.9976439804 | pinball CCD/POD rank-3 principal cosines | preferred derived result | bounded current |
+| N11 | 0.9974651867 | pinball-inclusive minimum declared stability cosine | preferred derived result | bounded current |
+| N12 | 0.964749, 0.967141, 0.973321 | historical Kármán sensor R² folds | OID final audit | conditional historical only |
+| N13 | 0.836206, 0.867670, 0.872087 | historical Kármán force R² folds | OID final audit | force not promoted |
+| N14 | 0.936949, 0.928960, 0.939764 | historical Illusion sensor R² folds | OID final audit | conditional historical only |
+| N15 | 0.701865, 0.694162, 0.480277 | historical Illusion force R² folds | OID final audit | force not promoted |
+| N16 | 0.0231672176752314 | steady `E_inf_vector` at `x/D=10`, `s=3.55` | steady current result | contextual, separate chain |
+
+### 12.2 Formula ledger
+
+| formula | role | classification |
+|---|---|---|
+| `E_r` weighted ROI mean target error | CCD performance estimand | existing derivation / canonical result |
+| `U_(c,b)=(q_DRL-c mean)-(constant template_b-c mean)` | corrected phase residual | existing derivation / canonical result |
+| `A=P_c(W^1/2 U_c)^T/(N√3)` | literal weighted CCD operator | existing derivation / code-verified |
+| `Φ=W^-1/2V`, `Φ^TWΦ=I` | physical weighted modes | existing derivation / code-verified |
+| `G=U_c^TWU_c`, principal cosines of `Φ^TWΨ` | raw weighted POD comparison | existing derivation / code-verified |
+| `C=Σ_baΣ_aa+`, `L_R`, `L_E` | OID method core | existing method / tests independently verify finite-dimensional identities |
+| `Δq_ctl=q_ctl-q_blk` | OID paired branch difference | established contract, not causal effect |
+| `Re_D=Re_code/2` for Legacy `2D` reference | scale convention | code/document contract |
+| `(Uw,Vw)=(-ωr_y,ωr_x)` | wall/action oracle | source-derived physics contract |
+
+No independent numerical linear-algebra run was needed for this index. If later run in `pinball_math`, label it **independent check**, not new evidence; any new physical interpretation is **new conjecture**.
+
+## 13. Search and use record
+
+- **Nowledge Mem:** context bundle/Working Memory was read at start; targeted memory search covered corrected Kármán CCD, mean-field boundary, OID reset/history, q roles and evidence authority. Memory was used only for routing and cross-check; repository artifacts overrode it. No durable memory was added because this file is the requested repository index and no new project decision was made.
+- **Undermind:** called orientation, inspected existing `DynamisLab` workspace `4383d145-2c80-4991-8a9c-7ac2dd683438`, browsed existing CCD/OID files and method-reference folders. No deep search started. Existing workspace notes for Schlegel/LR-LE, CCD/PCD and observable decomposition were treated as literature pointers only. The active method bibliography is recorded in `src/OID_analysis/METHOD.md`; no external paper was used to upgrade internal evidence.
+- **No CFD/GPU:** this indexing task ran no CFD, no CUDA, and did not modify evidence artifacts.
+
+## 14. Final fail-closed checklist
+
+- [x] `src/CCD_analysis/**` current authority and archive boundary indexed.
+- [x] `src/OID_analysis/**` active method core and conditional historical study indexed.
+- [x] field variables, estimands, masks, ROI, weights and dimensions recorded.
+- [x] corrected mean numbers, mean/dynamic decomposition and bounded wording recorded.
+- [x] CCD operator, POD comparison, rank/strength/stability, symmetry, phase and figure inventory recorded.
+- [x] explicit noncausal/nonmechanistic CCD boundary recorded.
+- [x] OID reset reasons, reusable definitions and prohibited results recorded.
+- [x] q target/block/control and acquisition/provenance contracts recorded.
+- [x] canonical roots, publication sets and hashes recorded.
+- [x] SR constant component and steady theory comparable/non-comparable map recorded.
+- [x] conflict/gap list and candidate physical logic recorded.
+- [x] no article-style narrative, no CFD, no evidence modification, no commit.
+
+**Handoff boundary:** downstream synthesis may use this file as an index. It must preserve status labels (`current`, `bounded`, `contextual`, `historical`, `conditional`, `failed_closed`, `prohibited`) and must not turn candidate physical logic into manuscript fact.

docs/JFM_WYQ/ROUND1_BOTTOM_UP/04_STEADY_THEORY_AND_CROSSMAP.md

@@ -0,0 +1,250 @@
+# Round 1 bottom-up index — steady theory and cross-map
+
+**Role / status.** This is an evidence-navigation dossier, not JFM prose. Repository contracts, current manifests, code/tests, and immutable artifacts decide project facts; Nowledge Mem and Undermind are navigation/context only. Scope is the active `src/steady_pinball_theory/**` package, remaining `src` cross-navigation, and explicit failed/open gates. No CFD or new evidence was run for this index.
+
+## 1. Scope and question
+
+Primary question: under the canonical uniform-inlet, free-slip-channel, finite-Re steady contract, what is actually established about a rear-only constant-rotation low-deficit endpoint, and what analytical or conservation route could connect wall actuation to the observed profile without overclaiming a mechanism?
+
+Required separation:
+
+- **Outer theory:** incompressible irrotational Euler/Neumann fields with prescribed geometric circulations, unbounded or free-slip strip.
+- **Finite-Re observation:** current-v5 endpoint fields and bounded diagnostics at `Re_D=50`.
+- **Mechanism/stability:** not closed. Potential flow cannot supply viscous no-slip vorticity, separation, drag, torque, base pressure, or global stability.
+- **Cross-project context:** SR/CCD/OID/Legacy/V5 are linked only where their plant, estimator and evidence role match; labels and memory never merge chains.
+
+## 2. Sources inspected and authority map
+
+### 2.1 Package authority (current)
+
+- `src/steady_pinball_theory/README.md`: entrypoint, frozen boundaries, display/reproduction commands, no-CFD verification.
+- `DERIVATION.md`: equation/claim registry, geometry/sign contract, Euler/Hodge proof, one-cylinder counterexample, MFS/strip/Laurent/multipole validation boundary, NS-first stage notes.
+- `RESULTS.md`: current numeric observations, accepted/reverse endpoint roles, failed circulation explanation, partial NS ledger, transient limits, SR/CCD scope.
+- `LITERATURE.md`: source-to-claim ledger and prohibited transfers; primary references include Crowdy 2006, Glass et al. 2019, Ueda et al. 2003, Glauert/Moore 1957, Cornejo Maceda et al. 2021, Chan et al. 2011, Rodrigues-Asensio et al. 2026, Mao–Blackburn–Sherwin 2015 and others.
+- `HISTORY_AND_LESSONS.md`: navigation/history only; records wrong-sign, double-subtraction, incomplete-CV, torque-ownership and immutable-artifact failures.
+- `contract.py`: canonical geometry, `Re_D`, wall law, `s` mapping, body/action oracle.
+- `metrics.py`: direct profile metric, gap/flux, vorticity/near-wall proxies, partial momentum/energy ledgers, no closure unless all terms supplied.
+- `theory.py`: MFS, strip images, prescribed-vortex fields, Laurent cross-check, multipole projection, gap ansatz, root/Jacobian and Kármán-quarantined helpers.
+- `tests/test_core.py`: implementation contracts and mathematical/numerical checks; passing tests are not physical validation.
+- `evidence/INDEX.json`, `evidence/README.md`, authoritative roots referenced by package README/results, and figure manifest/CSV outputs: machine/artifact authority. External Optane paths are part of provenance and must be hash-resolved before reuse.
+
+### 2.2 Repository cross-navigation
+
+- `src/analysis_knowledge.md`: cross-project contracts, `Re_code` versus `Re_D`, `q_in/q_blk/q_ctl`, evidence taxonomy, and steady bounded/open status.
+- `src/analysis_notes.md`: current gates. Steady Gate B requires same-CV NS closure, intervention/independent-realization evidence for causality/stability, and a fresh no-slip-transfer problem.
+- `src/understanding_notes.md`: Legacy/V5 compatibility ledger; useful for Re/action/telemetry separation, not a steady result source.
+- `src/README.md`: package roles and authority order; `pv_plot` is a rendering utility only.
+- `src/SR_analysis/{CLAIMS.md,results/README.md,PIPELINE.md}`: current Kármán/cloaking SR; steady is contextual calibration only.
+- `src/CCD_analysis/{README.md,FINAL_RESULTS.md,CLAIMS.md}`: corrected Kármán descriptive mean/dynamic decomposition; no steady quantitative import.
+- `src/OID_analysis/{README.md,METHOD.md,PHYSICS_CONTRACT.md}`: active claim-free CPU method core; July study is archive/conditional.
+- `src/pv_plot/README.md`: interface confirmation only (`axis_order`, presets, ParaView environment); no physical evidence.
+- `src/archive/**`: historical/superseded by default. No archive item is current evidence unless an active manifest explicitly references it, and even then the manifest/role—not archive location—controls status.
+
+### 2.3 Memory / literature navigation
+
+Nowledge Mem context bundle and targeted memory search were used. Relevant durable context was consistent with repository authority: accepted branch is `[0,+Omega,-Omega]`, centers/order are explicit, `E_inf_vector` is full-cross-section direct `q_in` comparison, MFS validation precedes interpretation, and SR/CCD boundaries remain separate. Memory did not supply project facts or upgrade claims.
+
+Undermind was oriented and the existing `DynamisLab` workspace (`4383d145-2c80-4991-8a9c-7ac2dd683438`) was identified. No new deep search was started. Existing workspace literature is reflected through the repository’s source ledger; no external number or equation was imported here. Full-PDF/equation mapping remains explicitly held where `LITERATURE.md` says PDF unavailable (notably Glauert/Moore and Urzhumov–Smith for this pass).
+
+## 3. Canonical contract / assumption and BC ledger
+
+| Item | Frozen definition | Status / boundary | Source |
+|---|---|---|---|
+| Geometry | `D=20`, `R=10`; centers `(1000,299.5)`, `(1026,314.5)`, `(1026,284.5)` in lattice coordinates; normalized centers `(0,0)`, `(1.3,±0.75)` relative to `D` | established contract | `contract.py`, README/manifest |
+| Body order | `(front, rear_y_plus, rear_y_minus)` | established; semantic IDs beat labels | `contract.py`, sign oracle |
+| Coordinates | x right, y up | established | `contract.py`, `theory.py` |
+| Wall law | `(U_w,V_w)=(-omega*r_y, omega*r_x)` | positive `omega` is geometric CCW | `contract.py`; runtime/source oracle |
+| Accepted action | `[0,+Omega,-Omega]`, `Omega>0` | accepted/current; gap-facing rear cardinals move downstream, base-bleed/gap-jet wall-sense family | contract oracle, current evidence |
+| Reverse action | `[0,-Omega,+Omega]` | reverse/high-deficit role; wrong-sign campaigns and withdrawn production labels must not be re-promoted | `RESULTS.md`, evidence disposition |
+| Spin ratio | `s=Omega R/U_inf`; `omega=s U_inf/R` | established definition; do not silently use `Omega D/U_inf` | `contract.py`, `DERIVATION.md` |
+| Reynolds number | canonical `Re_D=U_inf D/nu=50`; `nu=U_inf D/Re_D`; Legacy `Re_code` convention is separate and often `Re_D=Re_code/2` | established; never call Legacy `re100` simply `Re_D=100` | `contract.py`, `analysis_knowledge.md`, `understanding_notes.md` |
+| Inlet/reference | uniform `q_in`; direct comparison to `q_ctl` | primary steady comparator; `q_blk` non-ranking | README, `RESULTS.md`, `metrics.py` |
+| Outer channel | free-slip walls at declared `H`/half-height; finite strip and unbounded plane are separate mathematical settings | established model assumption; not no-slip transfer | `DERIVATION.md`, `theory.py` |
+| Cylinder walls | current finite-Re solver uses rotating no-slip; potential solver enforces impermeability only | critical interface boundary | README, `DERIVATION.md`, LITERATURE |
+| Station/metric | full two-component `E_inf_vector=max_y ||(q_ctl-q_in)/U_inf||_2` at `x/D=10`; `E_inf_x/y`, explicit L2 and other stations diagnostics | established metric contract | `metrics.py`, RESULTS |
+| Steady qualification | settling/endpoint qualification is finite-horizon and case-specific | does not establish uniqueness, attraction, stability, or causal intervention | RESULTS/evidence |
+| Telemetry | normalized Celeris loads are lattice sums averaged once per accumulated LBM step; runtime manufactured oracle supports torque work sign and `P_act→fluid=sum(omega*tau_read)` | force ownership and full closure remain open | LITERATURE, `metrics.py`, NS ledger |
+
+### Rotation-sign checkpoint
+
+For upper rear (`r_y=+R`), positive omega gives outer-top `U_w=-omega R` (upstream) and gap-bottom `U_w=+omega R` (downstream). For lower rear with negative omega, outer-bottom is upstream and gap-top downstream. Therefore accepted `[0,+,-]` injects/organizes the rear gap jet; it is not boat-tail under the project/literature sign map. Every future campaign should print IDs, centers, action, four cardinal velocities, and direct reversal label before solving.
+
+## 4. Steady CFD sweep and accepted/reverse branches
+
+### 4.1 Authorized numerical observations
+
+Current v5 direct recomputation at `s={3.45,3.55,3.65}` gives approximately:
+
+- `E_inf_vector={0.0232525,0.0231671,0.0312490}` at `x/D=10` (current article-authoritative exact value for `s=3.55`: `0.0231672176752314`).
+- `s=3.55` is settling-qualified / low-deficit endpoint in the canonical free-slip `Re_D=50` setting.
+- Earlier/current displays include stationary, accepted `s=3.55`, and reverse `s=5.2`; these are reader views of authenticated artifacts, not new evidence.
+- Accepted `s=3.55` and reverse `s=5.2` are **not amplitude matched**. Reverse is a distinct stable/high-deficit state candidate, not a same-amplitude causal control pair.
+- A provisional direct-profile minimum around `s≈3.45–3.55` is not a qualified continuous optimum; restarted bracket, continuation, grid/domain/method convergence, and held-out validation gates remain required.
+
+Do not infer from the sweep: unique/global optimum, global attraction, asymptotic/nonlinear stability, cross-BC transfer, causal mechanism, or a universal `s` law.
+
+### 4.2 Withdrawn / reverse state disposition
+
+- `[0,-Omega,+Omega]` campaigns were wrong-sign/withdrawn for the accepted cloak branch; retain their disposition for audit, never use their attractive plots or labels as current evidence.
+- A reverse endpoint can be numerically steady while retaining a poor wake. “Steady” and “low deficit” are independent axes; Chan’s pair result is only qualitative context for this distinction.
+- The current reverse role is high-deficit/stability candidate at `s=5.2`; because amplitude is mismatched against accepted `s=3.55`, no accepted-versus-reverse mechanism or causal ranking follows.
+
+### 4.3 Artifact and figure map
+
+| Need | Current artifact/reader asset | What it supports | Does not support |
+|---|---|---|---|
+| machine authority | `src/steady_pinball_theory/evidence/INDEX.json` and linked authoritative roots | path/hash/role disposition | prose authority detached from hashes |
+| direct endpoint | `steady-contract-v5-torque-d20-20260808/{qin,accepted-s3.55,reverse-s5.2,stationary}/` | fields/manifests/contract-bound metrics | closure/stability |
+| article result | `steady-article-publication-v3b-20260809/REPORT.md` | exact direct metric and bounded publication status | new CFD |
+| NS-first diagnostics | `ns-ledger-v2-oracle-bound-d20-20260809/analysis.json` plus display CSVs | power, gap flux, station/wake, recirculation, vorticity, force readbacks as partial terms | residual/closed balance, force ownership |
+| potential/strip | `steady-strip-modal-analysis-20260804-v4/analysis.json`; project theory readers | strip modal comparison and failed sign route | no-slip closure / mechanism |
+| display | `figures/manifest.json`, `field_metrics.csv`, `profiles_long.csv`, summary PNG/PDF/CSV | traceable reader visualization | publication-ready independent evidence |
+| tests | `tests/test_core.py` | implementation contract checks | physical proof |
+
+## 5. Equation ledger
+
+| ID | Relation / object | Classification | Validity and gate |
+|---|---|---|---|
+| E1 | `u=u_N+Σ Γ_j h_j`; fixed normal data + all geometric periods imply uniqueness under stated integration hypotheses | existing derivation / proved mathematical result | Euler outer BVP; conditional theorem, not existence/stability |
+| E2 | One circle: `W=U(z+a²/z)-iΓ/(2π)log(z/a)`; `u_r(a)=0`, `u_θ(a)=-2U sinθ+Γ/(2πa)` | existing derivation, literature-supported | impermeability independent of Γ; full rotating trace impossible for `U≠0`; special `U=0, Γ=2πa²Ω` is counterexample |
+| E3 | MFS zero-source-per-body constraints and Neumann collocation | existing implementation/math contract | numerical representation of outer impermeability; zero source is stronger than global flux compatibility |
+| E4 | Strip images: equal-sign source images, opposite-sign vortex images; finite image layers are truncation | existing implementation/derivation | requires dense off-grid and image-layer convergence; not automatic unbounded equivalence |
+| E5 | Independent finite Laurent/Fourier collocation | existing implementation | zero-circulation unbounded cross-check, not Crowdy exact, strip or prescribed-circulation validation |
+| E6 | Rear-opposite prescribed circulation gives log cancellation and dipole coefficient `C_pair=-Γ b/π` | existing derivation / verified mathematical artifact | unbounded outer potential far-field property only |
+| E7 | Affine far coefficient `C(Γ)=C0+Γ dC`; real least-squares cancellation | existing derivation / numerical verification | potential objective only; `g_opt≈+6.147944` in strip route conflicts with finite-Re fitted descriptive `g≈-12.200554` |
+| E8 | `E_inf_vector` direct profile and explicit L2 | existing code + authorized artifact observation | metric definitions; L2 is diagnostic and cannot replace primary metric |
+| E9 | Gap flux `Q_g=∫u·n ds`, mass flux `∫ρu_n ds`, momentum proxy `J_g=∫ρu_n²ds` | existing code / diagnostic | oriented gap observable; does not establish jet causality or full CV balance |
+| E10 | Station mechanical energy flux `∫(p+½ρ|u|²)u_n ds`; dissipation `∫2μS:S dA`; ledger residual `right-left+lateral+dissipation+storage-P_act` | existing derivation/code contract | closure exists only when every signed term is supplied on one CV; current ledger is partial |
+| E11 | `p≈c_s²(ρ-ρ_ref)` | existing code diagnostic | weak-compressible LBM pressure proxy, not incompressible NS pressure |
+| E12 | wall tangential speed, radial tangential-gradient and near-wall sign-change estimators | existing code diagnostics | proxies; current resolution/gates do not prove wall-vorticity flux or separation/reattachment |
+| E13 | source multipoles / low-order projection of high-order MFS | existing derivation/implementation | interpretation aid after solver validation; not a viscous closure |
+| E14 | Kármán point-vortex Hamiltonian/street helpers | quarantined mathematical oracle | separate Kármán chain; no steady pinball transmission claim |
+
+### What is not an equation-level result
+
+`Gamma_eff(s)`, `Gamma(Omega)`, gap-jet-to-deficit causality, boat-tail/base-bleed equivalence across Re, stability boundaries, or any energy/momentum coefficient inferred from incomplete terms are not established relations. They are either deferred empirical closure targets or conjectures.
+
+## 6. Potential-flow / rotation-sign conflict
+
+The potential route is internally coherent as an outer impermeability problem and has useful rear-opposite dipole cancellation. Its proposed physical identification fails because the sign required by the potential strip cancellation (`g_opt≈+6.147944`) is opposite to the accepted finite-Re endpoint’s corrected descriptive circulation fit (`g≈-12.200554`). The accepted CFD nevertheless has a potential-like exterior appearance in figures; resemblance is insufficient to identify the mechanism.
+
+Correct interpretation:
+
+1. potential flow supplies `q_outer(Γ)` counterfactual/reference fields;
+2. rotating no-slip CFD supplies vorticity and wake processes absent from Euler;
+3. comparison must be `q_CFD-q_potential` on preregistered common masks, with contour sign and uncertainty;
+4. then connect residuals to gap flux, shear/separation proxies, recirculation, force and downstream deficit;
+5. no claim survives without held-out endpoint/branch and resolution checks.
+
+## 7. MFS / analytic / multipole status and priority
+
+### Existing capability
+
+`theory.py` implements unbounded and strip MFS with prescribed geometric-CCW circulations, zero-source constraints, finite image layers, dense boundary diagnostics, gap/transpiration ansatzes, independent Laurent collocation, source-multipole reconstruction, affine circulation bases, and sampled stagnation-root/Jacobian tools. Existing project tests include mathematical or implementation checks; the code is not a substitute for a new evidence package.
+
+### Validation priority / gate
+
+Before interpreting any MFS field:
+
+1. dense off-grid no-penetration residuals, separately by cylinder and gap/outer sectors;
+2. rank/conditioning, source radius, MFS order, image-layer convergence (strip);
+3. one-cylinder exact circle oracle including circulation trace;
+4. independent Laurent/Fourier cross-check for unbounded zero-circulation fields;
+5. held-out points/geometry and common-exterior profile convergence;
+6. only then project high-order MFS into low-order per-body dipoles/multipoles plus antisymmetric circulation, showing error versus term count.
+
+Current gate state in `DERIVATION.md` is reference-validation/deferred for dense and parameter-convergence evidence. Collocation residual alone is not validation. The composite semi-analytic route is failed historical evidence (double subtraction, inactive wake term, non-independent gates), not an active model.
+
+## 8. NS-first, energy, momentum, gap-jet and base-bleed chain
+
+### Candidate physical chain (hypothesis, not established mechanism)
+
+`wall rotation sign → gap-facing wall motion / local shear → near-wall vorticity production and separation/shear-layer birth → gap mass/momentum redistribution and base-bleed-like jet → recirculation/base pressure/force changes → downstream deficit reduction`.
+
+A second, weaker candidate is `rear constant rotation → persistent mean-field compensation`, with dynamic feedback as a separate perturbation. SR/CCD can motivate this decomposition but cannot prove the arrows.
+
+### Available observables
+
+- actuator torque/power sign and samples where explicit history exists;
+- signed gap `Q_g`, mass flux and `J_g` by oriented segment;
+- station mass/vector momentum/kinetic/mechanical-energy flux;
+- recirculation, wake deficit proxies, vorticity moments, near-wall tangential speed, radial-gradient and ux-zero-crossing proxies;
+- endpoint field comparison against `q_in` and potential references.
+
+### Why closure is not yet available
+
+The current NS ledger is endpoint/spatial partial: missing or unsupported terms include a common control volume with complete storage, lateral/outlet flux, pressure and viscous traction/wall terms, body-adjacent resolution, and certified force ownership. `p` is a weak-compressible density proxy. The implementation explicitly returns `incomplete_not_closed` unless every signed term is supplied. Therefore “energy balance,” “momentum balance,” “actuator power explains deficit,” and “gap jet causes cloak” are open, not results.
+
+### Required next gate (no CFD authorized by this dossier)
+
+If reopened, use fresh no-clobber artifacts and one frozen CV: matched accepted/reverse amplitude, same inlet/reference and window, pressure/viscous/stress boundary terms, body work, storage, lateral flux, grid/domain/time convergence, and held-out prediction. A matched intervention or independent realization is needed for causality; continuation/eigenvalue/basin/order-parameter evidence is needed for stability. Do not use archive artifacts to fill missing terms.
+
+## 9. Cross-project map
+
+| Package / chain | Connectable item | Explicitly not connectable |
+|---|---|---|
+| SR Legacy Kármán/cloaking | rear-constant term ranking and steady magnitude correspondence as contextual calibration; `Re_code`/`Re_D` conversion only when explicitly documented | V5 steady metric as Legacy SR evidence; SR deletion as physical intervention; necessity, circulation, momentum law, mechanism, universal `s` |
+| corrected CCD Kármán | mean-versus-dynamic separation and descriptive constant-field context | no steady CCD number; no phase co-variation as steady mechanism/causality; different plant, periodic estimator and chain |
+| OID active core | future method vocabulary for force/signature observables and strict alignment | no active scientific result; archived high-R²/action claims; no importance/authority/causality |
+| Legacy / V5 | explicit compatibility audit: action units, object order, telemetry, norm, BC, solver version | numerical equivalence, shared metric, cross-generation pooling without parity artifact |
+| `drl_pinball` | broad control/case navigation; Kármán scope and role boundaries | DRL performance as steady mechanism proof; Legacy controls as V5 steady data |
+| `pv_plot` | `tyx`/`txy` axis-order and vorticity/speed/comparison interface confirmation | physical interpretation, metric authority, CFD validation |
+| `src/archive` | failure chronology, superseded route and negative controls | current evidence, new numbers, mechanism rescue |
+
+## 10. Core short derivations: main-text suitability
+
+The following are short enough to become later main-text equations after evidence-parity review, but this dossier does not draft prose:
+
+- canonical `Re_D`, `s`, wall-velocity mapping and direct `E_inf_vector` definition;
+- one-circle trace showing Euler impermeability does not select circulation and full rotating trace is obstructed for `U≠0`;
+- rear-opposite log cancellation / dipole coefficient as an explicitly labeled outer reference and sign counterexample;
+- energy/momentum ledger identity only as a required closure equation with all terms visibly marked unavailable.
+
+MFS matrix construction, strip images, source constraints, Laurent cross-check, multipole projection, and root/Jacobian algorithms belong in methods/supplement unless the validation gate is completed. No derivation should be written as a finite-Re mechanism until CFD-minus-reference residual and NS closure evidence exists.
+
+## 11. Failed routes, open checks, coverage gaps
+
+### Failed / withdrawn routes
+
+- wrong-sign `[0,-Omega,+Omega]` cloak interpretation and campaigns;
+- prescribed-circulation cancellation as accepted CFD mechanism (`g` sign conflict);
+- composite semi-analytic scorer with double subtraction/inactive wake/non-independent gates;
+- incomplete CV ledger promoted to energy/momentum closure;
+- single switch/return realization promoted to causality, branch switching, basin or stability;
+- direct potential-to-rotating-wall no-slip transfer;
+- archive plots or Legacy/V5 similarity used as current cross-chain evidence.
+
+### Open checks
+
+- dense off-grid MFS residuals by body sector and all convergence dimensions;
+- one-circle exact and Laurent/Fourier independent validation in the active evidence lineage;
+- strip image-layer and finite-strip/unbounded comparison;
+- accepted/reverse amplitude-matched continuation and stability qualification;
+- grid/domain/window convergence and held-out `E_inf_vector`;
+- complete same-CV NS momentum/mechanical-energy closure with certified pressure, viscous, wall and storage terms;
+- empirical `Gamma_eff(s)` only with sign/contour/mask/Re/geometry/confinement/uncertainty and held-out endpoints;
+- no-slip transfer experiment is a new contract, not inherited from free-slip;
+- unresolved CCD steady interface remains separate and cannot backfill steady mechanism.
+
+### `src` cross-navigation coverage
+
+Covered as active entry/role checks: `steady_pinball_theory/**`, `analysis_knowledge.md`, `analysis_notes.md`, `understanding_notes.md`, `src/README.md`, `SR_analysis` claims/results navigation, `CCD_analysis` corrected authority, `OID_analysis` reset authority, `pv_plot` interface, and repository-level archive boundary. No `src/archive/**` file was promoted to current evidence. Remaining limitation: `drl_pinball` active utility/package files and the large `CelerisLab/` external/generated tree were not exhaustively indexed here; DRL agent owns graphic/control use, and CelerisLab is explicitly outside repository closeout. This is the principal residual coverage gap.
+
+## 12. Candidate figure / code / artifact map
+
+- contract/sign panel: `contract.py`, sign oracle tests, current endpoint manifest;
+- accepted/reverse/stationary aligned fields: `figures/plates/**`, raw `fields/**`, `field_metrics.csv`, `profiles_long.csv`;
+- direct metric curve: `figures/summary/profile_Einf_vector.{csv,png,pdf}`;
+- gap/base-bleed diagnostic: `figures/summary/signed_gap_Qstar.{csv,png}` and NS gap CSVs;
+- near-wake diagnostics: `figures/summary/near_wake_diagnostics.{csv,png}`;
+- potential-vs-CFD/error plates: `figures/plates/error/**`, with potential analytical-vorticity-zero convention;
+- MFS diagnostic code: `theory.py`, `metrics.py`, `tests/test_core.py`; no generated display is a new result;
+- article/report and machine map: authoritative external roots and `evidence/INDEX.json`.
+
+## 13. Final evidence verdict for handoff
+
+**Supported:** canonical contracts; conditional Euler/MFS mathematics; outer potential reference; bounded current-v5 `Re_D=50` free-slip accepted low-deficit endpoint; bounded reverse high-deficit steady candidate; descriptive gap/wake/force/power diagnostics where artifacts expose them; SR steady correspondence as contextual only.
+
+**Rejected/withdrawn:** potential-circulation cancellation as the accepted finite-Re mechanism; wrong-sign accepted branch; composite semi-analytic PASS; closed energy/momentum explanation; causality/stability/no-slip transfer claims.
+
+**Interpretation level:** the gap-jet/base-bleed chain is a candidate physical chain, not a project-established mechanism. Main mathematical risks are (i) Euler circulation is independent of rotating no-slip wall speed, (ii) strip versus unbounded asymptotics and finite-image truncation, (iii) sign/body/metric confusion, and (iv) incomplete NS terms and coarse near-wall resolution. Highest-value next work is validation/closure gating, not new narrative or another CFD sweep.

docs/JFM_WYQ/ROUND1_BOTTOM_UP/05_INTRODUCTION_LITERATURE_INDEX.md

@@ -0,0 +1,145 @@
+# JFM Round 1 — Introduction Literature / Material Index
+
+> 资料索引，不是 Introduction prose。标签：`[FACT]` 外部文献直接支持；`[OBS]` 本地 bounded observation；`[INTERP]` 作者叙事选择；`[HIST]` 历史动机；`[OPEN]` 未闭合。阅读层级：`full-PDF`、`full-report`、`abstract/snippet`。
+
+## 0. Executive map
+
+最稳健的问题链是：**在包含分离、涡脱落、对称性破缺和多吸引子的中等 Re bluff-body 流中，能否用有限传感器和紧凑旋转执行器，以 observer-level wake signature 为目标主动控制；若 DRL 找到性能策略，能否通过受原 plant 闭环 CFD 检验的 SR、场分解和稳态/理论分析，将其降解为受限的物理控制骨架？**
+
+候选链：`minimal cloak vocabulary → linear/passive wave analogues → active model-specific hydrodynamic cloaks → AFC/DRL → fluidic pinball → sparse sensing/signature/DTW → SR and field interpretation → bounded gap`。其中线性波到非线性流体是背景对比，不是物理等价链；Urz12 是 active porous cloak，Ren20b 是 DRL cylinder stealth，二者都不等同于本项目旋转 pinball。
+
+## 1. Audit trail
+
+### Nowledge Mem
+已调用 `read_context_bundle`，读取 Working Memory、owner identity、active scope；并定向调用 `memory_search` 检索完整路线、导师意见、Confirmation 原始动机、JFM reconnaissance、hydrodynamic stealth/illusion、fluidic pinball、DRL、sparse sensor/DTW、SR/可解释控制。关键 recalled boundaries：active positive SR 仅 Legacy Kármán/cloaking；Illusion SR 为 historical/negative；CCD descriptive；OID claim-free；steady bounded contextual；Legacy 与 V5 分开；应用需简短前瞻，seals、海洋结构、环境扰动、扬沙等不作主线。`thread_search` 对相同主题无匹配线程，因此 Mem 是路线线索而非外部科学证据。
+
+### Undermind
+已先 `get_orientation`，再 `list_workspaces`；进入现有 DynamisLab workspace `4383d145-2c80-4991-8a9c-7ac2dd683438`。只读检查 root、`Confirmation_report/`、`JFM/`、`JFM References/`，及 folders 01–03、05–06；workspace 有 54 files、14 existing deep searches。检查了已有报告：Hydrodynamic stealth and illusion；Symbolic regression；Interpretable analysis；Minimal wake-field L2；Wake-field L2/sensor rationale；Closed-loop OID；JFM rigor for CCD/PCD；以及各论文清单。批量读取 20 篇最相关 PDF 的正文或原始 PDF 页眉：Rab18b, Ren20b, Mac20, Den18b, Den20, Den21, Li21, Wan23b, Xia23, Loi16, Loi17, Gau14, Par20, Urz12, Vas09, Ma13, Zou19b, Pen06b, Leo06b, Ren20c；其余按摘要/元数据标注。本轮没有启动新 deep search，也没有修改 workspace。一次 moderate-Re cloak `search_papers` 被自动审查拦截为新 deep-search-like discovery，未绕过；只保留已有库和已返回定向结果。
+
+### Repository / historical materials
+已检查 `docs/JFM_WYQ`、round README、根 README、`src/steady_pinball_theory/LITERATURE.md`、DRL/SR/CCD writing handoff/results、`src/understanding_notes.md`，以及 Undermind 的 Confirmation `Conf_1_Intro/2_Description/5_Conclusion` 和 `JFM/Introduction and Background Draft`。当前仓库未发现可直接作为 authority 的 Confirmation `.tex/.bib`；旧报告只用于原始动机与演进，不覆盖当前 artifact/claim ledger。
+
+## 2. Searchable taxonomy
+
+- **A 概念**：visibility/signature；cloak=observer-space 接近 background；illusion=接近非零 target；passive/active；local body vs external signature。
+- **B 线性谱系**：transformation optics/conformal mapping；active exterior cloak；active illusion；acoustic/water-wave cloak；线性叠加、频带、材料/源限制。
+- **C hydrodynamic bridge**：porous/Brinkman active cloak；低 Re/线性化水波；finite-Re separation/wake；compact moving-wall/jet actuation 的未闭合转移。
+- **D AFC/DRL**：rotation/jets；sensitivity/optimal control；genetic-programming MLC；PPO；noise/Re robustness；partial observation/history/action filtering。
+- **E pinball**：MIMO rotating cylinders；Hopf/pitchfork；force/low-order models；multi-attractor/chaos；open/closed-loop control。
+- **F observer/metric**：wake information；sparse velocity/pressure；sensor placement/delay；DTW similarity；sensor vs full-field estimand；zero/target comparator。
+- **G interpretation**：black-box policy；GP/SR；SINDy/sparse Galerkin/feature lifting；POD/OID/CCD；closed-loop redeployment；association ≠ causality。
+
+## 3. Core literature cards
+
+### Cloaking, illusion, hydrodynamic bridge
+
+|key|exact identity|DOI|read|supports / cannot support|
+|---|---|---|---|---|
+|Pen06b|Pendry, Schurig & Smith, “Controlling Electromagnetic Fields”, *Science* 312 (5781), 1780–1782 (2006)|10.1126/science.1125907|full-PDF; formal publisher metadata overrides the erroneous workspace PDF header|[FACT] transformation-optics origin / not nonlinear hydrodynamics|
+|Leo06b|Leonhardt, “Optical Conformal Mapping”, *Science* 312, 1777–1780 (2006)|10.1126/science.1126493|full-PDF; journal pages verified|[FACT] optical genealogy / not fluid transfer|
+|Vas09|Vasquez, Milton & Onofrei, “Active exterior cloaking for the 2D Laplace and Helmholtz equations”, *Physical Review Letters* 103, 073901 (2009)|10.1103/PhysRevLett.103.073901|full-PDF; preprint PDF plus journal metadata|[FACT] finite active sources and quiet exterior / not Navier–Stokes|
+|Ma13|Ma, Mei, Zhu, Jin & Cui, “Experiments on active cloaking and illusion for Laplace equation”, *Physical Review Letters* 111, 173901 (2013)|10.1103/PHYSREVLETT.111.173901|full-PDF; journal first page verified|[FACT] cloak vs nonzero illusion distinction / not wake target tracking|
+|Zou19b|Zou et al., “Broadband Waveguide Cloak for Water Waves”, *Physical Review Letters* 123, 074501 (2019)|10.1103/PhysRevLett.123.074501|full-PDF; journal first page verified|[FACT] engineered bathymetry/mode conversion / not bulk viscous wake|
+|Urz12|Urzhumov & Smith, “Flow stabilization with active hydrodynamic cloaks”, *Physical Review E* 86, 056313 (2012)|10.1103/PhysRevE.86.056313|full-PDF; preprint PDF plus journal metadata|[FACT] active Brinkman cloak suppresses cylinder shedding in finite Re window / not rotating no-slip pinball equivalence|
+|Ren20b|Ren, Wang & Tang, “Bluff body uses deep-reinforcement-learning trained active flow control to achieve hydrodynamic stealth”, *Physics of Fluids* 33, 093602 (2021)|10.1063/5.0060690|full-PDF; journal first page verified|[FACT] WSLB jets + near-wake sensors attenuate cylinder deficit / not global first, field cloak, illusion or causal mechanism|
+
+**Priority hold:** workspace contains `Ren19b`, “Elimination of velocity deficit behind a cylinder using reinforcement learning” (Ren & Tang, 2019), but no PDF/DOI. Do not call Ren20b the global first; next round must resolve whether Ren19b is preprint, predecessor or duplicate.
+
+### Fluidic pinball / JFM nearest neighbours
+
+|key|exact identity|DOI|read|supports / cannot support|
+|---|---|---|---|---|
+|Den18b|Deng, Noack, Morzyński & Pastur, “Low-order model for successive bifurcations of the fluidic pinball”, *Journal of Fluid Mechanics* 884, A37 (2020)|10.1017/jfm.2019.959|full-PDF; preprint PDF plus journal metadata|[FACT] successive Hopf/pitchfork and compact 3-cylinder nonlinear benchmark / not project thresholds/BCs|
+|Den20|Deng, Noack, Morzyński & Pastur, “Galerkin force model for transient and post-transient dynamics of the fluidic pinball”, *Journal of Fluid Mechanics* 918, A4 (2021)|10.1017/jfm.2021.299|full-PDF; journal first page verified|[FACT] structured modal force model / not controlled-flow causality|
+|Den21|Deng, Noack, Morzyński & Pastur, “Cluster-based hierarchical network model of the fluidic pinball – cartographing transient and post-transient, multi-frequency, multi-attractor behaviour”, *Journal of Fluid Mechanics* 934, A24 (2022)|10.1017/jfm.2021.1105|full-PDF; journal first page verified|[FACT] periodic/quasi-periodic/chaotic multi-attractor dynamics / not universal controllability|
+|Mac20|Cornejo Maceda, Li, Lusseyran, Morzyński & Noack, “Stabilization of the fluidic pinball with gradient-enriched machine learning control”, *Journal of Fluid Mechanics* 917, A42 (2021)|10.1017/jfm.2021.301|full-PDF; journal first page verified|[FACT] rotating-cylinder open/closed-loop ML; asymmetric steady + phasor / not PPO or project metric/sign transfer|
+|Rai20|Raibaudo, Zhong, Noack & Martinuzzi, “Machine learning strategies applied to the control of a fluidic pinball”, *Physics of Fluids* 32, 015108 (2020)|10.1063/1.5127202|full-PDF; journal metadata/PDF verified|[FACT] experiment, three hot wires at x/D=10, non-obvious actuator relations / not sensor optimality|
+|Fen23|Feng, Wang, Xiang, Jin & Fan, “How to Control Hydrodynamic Force on Fluidic Pinball via Deep Reinforcement Learning”, *Physics of Fluids* 35(4), 045137 (2023)|10.1063/5.0142949|full-PDF; formal publisher/DOI metadata verified|[FACT, weaker] DRL force min/tracking capability / not JFM, cloak, SR mechanism|
+|Mit01|Mittal, “Control of flow past bluff bodies using rotating control cylinders”, *Journal of Fluids and Structures* 15, 291–326 (2001)|10.1006/JFLS.2000.0337|full-PDF; journal metadata verified|[FACT] rotating controls modify separation/wake / not equal-cylinder pinball|
+
+### DRL, sparse sensing, interpretation
+
+|key|exact identity|DOI|read|supports / cannot support|
+|---|---|---|---|---|
+|Rab18b|Rabault, Kuchta, Jensen, Réglade & Cerardi, “Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control”, *Journal of Fluid Mechanics* 865, 281–302 (2019)|10.1017/jfm.2019.62|full-PDF; journal metadata/PDF verified|[FACT] PPO/DRL cylinder AFC from wake sensing; ~8% drag / not interpretability or field matching|
+|Par20|Paris, Beneddine & Dandois, “Robust flow control and optimal sensor placement using deep reinforcement learning”, *Journal of Fluid Mechanics* 913, A25 (2021)|10.1017/jfm.2020.1170|full-PDF; journal metadata verified, PDF is preprint|[FACT] noise/Re/sensor-layout trade-offs / not project sensor optimality|
+|Ren20c|Ren, Rabault & Tang, “Applying deep reinforcement learning to active flow control in weakly turbulent conditions”, *Physics of Fluids* 33, 037121 (2021)|10.1063/5.0037371|full-PDF; journal first page verified|[FACT] Re=1000 weak turbulence DRL AFC / not pinball universality|
+|Li21|Li & Zhang, “Reinforcement-learning-based control of confined cylinder wakes with stability analyses”, *Journal of Fluid Mechanics* 932, A44 (2022)|10.1017/jfm.2021.1045|full-PDF; journal metadata verified, PDF is preprint|[FACT] reward/reference/stability and few informative probes / not DTW causality|
+|Xia23|Xia, Zhang, Kerrigan & Rigas, “Active flow control for bluff body drag reduction using reinforcement learning with partial measurements”, *Journal of Fluid Mechanics* 981, A17 (2024)|10.1017/jfm.2024.69|full-PDF; journal first page verified|[FACT] partial observation/history/actions matter / not guaranteed reconstruction|
+|Wan23b|Wang, Yan, Hu, Chen, Rabault & Noack, “Dynamic feature-based deep reinforcement learning for flow control of circular cylinder with sparse surface pressure sensing”, *Journal of Fluid Mechanics* 988, A4 (2024)|10.1017/jfm.2024.333|full-PDF; formal publisher metadata overrides the erroneous workspace record/PDF header|[FACT] feature lifting enables sparse-sensor control / not velocity-pinball transfer|
+|Gau14|Gautier, Aider, Duriez, Noack, Segond & Abel, “Closed-loop separation control using machine learning”, *Journal of Fluid Mechanics* 770, 442–457 (2015)|10.1017/jfm.2015.95|abstract/snippet; metadata/journal page verified, PDF is preprint|[FACT] GP finds explicit closed-loop AFC law / not unique causal law|
+|Loi16|Loiseau & Brunton, “Constrained sparse Galerkin regression”, *Journal of Fluid Mechanics* 838, 42–67 (2018)|10.1017/jfm.2017.823|full-PDF; journal first page verified|[FACT] parsimonious constrained sparse models / not policy SR|
+|Loi17|Loiseau, Noack & Brunton, “Sparse reduced-order modelling: sensor-based dynamics to full-state estimation”, *Journal of Fluid Mechanics* 844, 459–490 (2018)|10.1017/jfm.2018.147|full-PDF; journal metadata verified, PDF is preprint|[FACT] sensor lifting/SINDy/field mapping black-to-grey route / not direct controller law|
+|Jin21|Jin, Illingworth & Sandberg, “Optimal sensor and actuator placement for feedback control of vortex shedding”, *Journal of Fluid Mechanics* 929, A13 (2021)|10.1017/jfm.2021.948|full-PDF; journal metadata/PDF verified|[FACT] placement has estimation/control/delay trade-offs / not x/D=10 optimum|
+|Bru19|Brunton, Noack & Koumoutsakos, “Machine Learning for Fluid Mechanics”, *Annual Review of Fluid Mechanics* 52, 477–508 (2020)|10.1146/annurev-fluid-010719-060214|full-PDF; journal metadata/PDF verified|[FACT] ML taxonomy and interpretation cautions / not project evidence|
+
+**DTW:** `Sak78` Sakoe & Chiba, “Dynamic programming algorithm optimization for spoken word recognition”, *IEEE Transactions on Acoustics, Speech, and Signal Processing* 26, 43–49 (1978), DOI `10.1109/TASSP.1978.1163055`; `Mul15` Müller, *Fundamentals of Music Processing*, Springer (2015), DOI `10.1007/978-3-319-21945-5`. Both are full-PDF in Undermind; they support alignment/similarity only. They support alignment/similarity only. Current project authority explicitly forbids interpreting DTW lag as physical delay or causality.
+
+## 4. Citation network / chronology
+
+`Pen06b/Leo06b → Vas09 → Ma13 → Zou19b` = linear/passive → active/illusion → water-wave vocabulary; no nonlinear-wake equivalence.
+
+`Urz12 → Ren20b` = active hydrodynamic stabilization/signature attenuation; porous shell and WSLB jets remain different actuators/metrics.
+
+`Gau14 → Mac20`; `Rab18b → Par20/Ren20c/Li21/Xia23/Wan23b` = ML/DRL-AFC lineage through robustness, turbulence, stability and partial observation.
+
+`Den18b → Den20/Den21/Mac20/Rai20` = pinball bifurcations → force/multi-attractor models → ML control and experiment.
+
+`Loi16 → Loi17 → project SR/OID/CCD` = sparse constrained models → sensor/field representation → project-specific audit; the final arrow is adaptation, not citation proof.
+
+Chronology: 2006 optical; 2009–13 active linear cloak/illusion; 2012 active porous hydrodynamic cloak; 2014–18 interpretable/low-order/DRL foundations; 2019–21 pinball and DRL stealth/control; 2022–23 multi-attractor, stability, partial-observation and sparse-feature JFM control; 2024–26 includes the JFM sparse-feature paper (Wang et al., JFM 988, A4, 2024); project evidence remains bounded.
+
+## 5. Confirmation / old-introduction motivation disposition
+
+### Retain, but demote
+- one sentence on science-fiction/visibility vocabulary;
+- passive/active and linear/nonlinear taxonomy;
+- conventional AFC objectives (drag, lift, separation, vibration);
+- pinball geometry/dynamics and sparse observer signature;
+- DRL as nonlinear policy discovery;
+- SR as an audit question;
+- one short prospective application clause.
+
+### Qualify / quarantine
+- “first explicit attempt” for Ren20b → early/direct precedent; resolve `Ren19b` first;
+- 99.5% → Ren’s exact near-wake velocity-deficit metric only;
+- “complete field restoration” → declared sensor/field space only;
+- Confirmation Re=50 and bifurcation language → audit `Re_code` vs `Re_D` first;
+- seals/whiskers → historical sensing context only, never application driver;
+- target Strouhal matching, cross-size generalization, first illusion, high-frequency explanation → historical/negative archive pending current authority.
+
+### Delete from mainline
+seals hunting; cylindrical marine structures; environmental disturbance/sediment resuspension; defense/low-observability claims; positive Illusion SR; DTW-as-delay; CCD/OID-as-mechanism; pooled Legacy/V5 evidence; universal/optimal/causal wording.
+
+## 6. Candidate argument chains (logic skeleton)
+
+### A — observer signature → active nonlinear cloak → interpretable skeleton (recommended)
+1. Wake carries observer information. 2. Define cloak/background and illusion/nonzero target. 3. Linear literature supplies vocabulary, not Navier–Stokes solution. 4. Urz12/Ren20b give adjacent hydrodynamic precedents. 5. DRL solves policy discovery but leaves observability/interpretability issues. 6. Pinball supplies MIMO actuation plus bifurcation complexity. 7. Ask whether Kármán cloak policy has a compact, CFD-validated law and bounded field structure. 8. Present SR → CCD/field audit → steady context with explicit nonclaims.
+
+### B — nonlinear benchmark → DRL → white-box audit
+1. Open with pinball’s simple geometry/difficult dynamics. 2. Den18b/20/21 explain why scalar force optimization is insufficient. 3. Mac20/Rai20 establish rotating-cylinder ML control. 4. Rab18b/Par20/Li21/Xia23/Wan23b motivate DRL with sparse/partial observations. 5. Gap: policy distillation plus field-level audit in the original plant. 6. Cloak vocabulary is objective, not history chapter.
+
+### C — mean compensation + dynamic correction
+1. Cloak as declared field/signature error decomposition. 2. Literature gives wake stabilization, sparse sensing, low-order/observable tools but no shared decomposition for this plant. 3. DRL → SR action law. 4. CCD tests descriptive phase-coherent low-rank residual. 5. Steady theory is contextual, not mechanism closure. 6. Gap is the bounded triangulation, not a universal law.
+
+## 7. Gaps and next-round deep-search questions
+
+Do not launch these in Round 1: (1) resolve `Ren19b` identity and Tang/Ren priority meaning; (2) peer-reviewed moving-wall/jet finite-Re hydrodynamic cloaks; (3) nonlinear viscous nonzero-target wake illusion; (4) JFM 2021–26 pinball + DRL/PPO/partial observation/SR nearest neighbours; (5) neural AFC policy → SR/GP → held-out closed-loop redeployment; (6) descriptive OID/EPOD/CCA/CCD against POD and causality boundaries; (7) DTW wake-signature papers and timing interpretation; (8) x/D sensor placement trade-offs for pinball/cylinder; (9) same-Re/geometry base-bleed and rotating-cylinder momentum/vorticity evidence; (10) recent JFM introduction narrative patterns.
+
+## 8. Citation priority
+
+**Tier 1:** Den18b, Mac20, Rab18b, Ren20b, Urz12, Par20 or Xia23, Gau14 or Loi17.
+
+**Tier 2:** Den20, Den21, Li21, Wan23b, Rai20, Jin21, Vas09, Ma13, Zou19b, Spe14, Bou11.
+
+**Tier 3:** Pen06b, Leo06b, Tri15, Mit01, Dip07, Loi16, Li16, Zol25, Sak78, Mul15, Fan20b, Ren20c, Bru19 and broad reviews. Use only when a specific transition needs them.
+
+Do not cite unresolved Confirmation BibTeX entries (`dolin1961possibility`, `bugnot2021current`, `corradini2001dynamic`, `krazer1905verhandlungen`) or `Ren19b` without re-verification.
+
+## 9. Boundaries and limitations
+
+- This is a curated traceable first pass, not a systematic bibliometric review; existing Undermind library/reports were prioritized and no new deep search was launched.
+- 20 PDFs were batch-read or checked at the original PDF header; cards now distinguish complete journal-page verification from preprint-only PDF reading. Full-PDF means Undermind paper-level extraction, not independent equation-by-equation audit.
+- Formal publisher metadata takes precedence over conflicting workspace metadata or PDF headers. In this revision, Pen06b, Wan23b and Fen23 were corrected from those conflicts; no identifiers were guessed.
+- External papers cannot establish internal solver, Re convention, metric, seed, artifact or claim status. Legacy and V5 remain separate.
+- Before prose: resolve `Ren19b`, verify Fan20b/Li16/Cas22/Zol25/Sak78 metadata, freeze one chain, and run a whole-manuscript claim audit.

docs/JFM_WYQ/ROUND1_BOTTOM_UP/06_CURRENT_DOCUMENT_COVERAGE_AUDIT.md

@@ -0,0 +1,111 @@
+# Round-1 current-document coverage audit
+
+> **Scope.** DynamisLab JFM Round-1 closure audit, not manuscript writing and not raw-artifact re-audit. Repository state and current package authority outrank Round1 dossiers, memory, plans, and archive material. No CFD/CUDA/training was run; dossiers were not modified.
+
+## 1. Coverage methodology and corrected denominator
+
+I first read the Nowledge Mem context bundle/Working Memory, then read the repository rule and the five existing Round1 dossiers. I then enumerated the six package trees and read every Markdown document retained as current reader/contract/navigation material in the audit scope, plus every selected current authority/index JSON listed below. This pass does not read archive leaf files or raw/generated payload artifacts.
+
+Statuses are now deliberately strict: **`read-full`** means the complete file was read in this pass; **`archive-navigation`** means only a current top-level document/index was read to establish historical routing; **`artifact-deferred`** means the authority document was read but its large/external/raw payload was not. `mapped` is no longer accepted as current coverage. `read-full` is not independent verification and neither status is `artifact-audited`.
+
+The corrected denominator is **63 current Markdown documents** and **8 selected current authority/index JSON documents**. The earlier 65-Markdown/17-JSON statement was a mapped inventory, not a full-read ledger, and overstated the current set. The two files `src/drl_pinball/train/CROSSRE_ANALYSIS.md` and `src/drl_pinball/train/VARDIST_ANALYSIS.md` explicitly identify themselves as archived transfer analyses; they are removed from the current denominator and retained only in archive navigation. Generated result JSON/CSV/NPZ, TensorBoard/model bundles, dense external Optane roots, and hidden raw artifacts are excluded from the denominator even when their manifests are current; their exclusion and authority-led deferral are recorded below.
+
+**Coverage outcome: 63/63 current Markdown documents are `read-full`; 8/8 selected current authority/index JSON documents are `read-full`.** No current reader-document hole remains. This is a full-reading result, not a claim that every scientific artifact or claim has been independently verified.
+
+## 2. Current Markdown ledger
+
+All paths in this section were read in full. The package counts include the repository-level current navigation documents (`src/README.md`, `src/analysis_knowledge.md`, `src/analysis_notes.md`, and `src/understanding_notes.md`) in the total of 63.
+
+| Package / role | Current Markdown read-full | Count |
+|---|---|---:|
+| `src/` cross-package authorities | `README.md`, `analysis_knowledge.md`, `analysis_notes.md`, `understanding_notes.md` | 4 |
+| `src/drl_pinball` | `README.md`, `CLAIMS.md`, `RESULTS.md`, `DATA_DICTIONARY.md`, `ASSET_STATUS.md`, `HISTORY_AND_LESSONS.md`, `WRITING_HANDOFF.md`, `knowledge.md`, `train/README.md`, `train/TRAIN_PIPELINE.md`, `train/DESIGN_DECISIONS.md`, `train/SERVER_DEPLOY.md`, `train/results/latest/HOW_TO_READ.md`, `eval/README.md`, `legacy_test/README.md`, `data/reproduction_plots_sr/README.md` | 16 |
+| `src/SR_analysis` | `README.md`, `CLAIMS.md`, `PIPELINE.md`, `HANDOFF.md`, `HISTORY_AND_LESSONS.md`, `ASSET_STATUS.md`, `results/README.md`, `results/INDEX.md`, `data/README.md`, `checks/README.md`, `tests/README.md`, `tools/README.md`, `utils/README.md`, `results/runs/article2-sr-elements-20260720/sr_key_elements.md`, `results/runs/article2-steady-analysis-20260720/interpretation.md`, `results/runs/article2-percase-refit-summary-20260720/interpretation.md`, `results/runs/article2-plotting-package-20260721/README.md`, `results/runs/article2-plotting-package-20260721/presentation/README.md`, `results/runs/article2-plotting-package-20260721/tables/offline_action_rmse.md`, `results/runs/article2-plotting-package-20260721/tables/term_deletion.md` | 20 |
+| `src/CCD_analysis` | `README.md`, `RESULTS_INDEX.md`, `CLAIMS.md`, `FINAL_RESULTS.md`, `WRITING_HANDOFF.md`, `FIGURES_AND_DATA.md`, `MATHEMATICAL_DERIVATION.md`, `karman_dynamic/CORRECTED_MATH_CONTRACT.md` | 8 |
+| `src/OID_analysis` | `README.md`, `SCIENTIFIC_RESET.md`, `METHOD.md`, `PHYSICS_CONTRACT.md`, `HANDOFF.md` | 5 |
+| `src/steady_pinball_theory` | `README.md`, `RESULTS.md`, `DERIVATION.md`, `LITERATURE.md`, `HISTORY_AND_LESSONS.md`, `figures/README.md`, `evidence/README.md` | 7 |
+| `src/pv_plot` | `README.md` | 1 |
+| **Total** |  | **61** |
+
+The table lists 61 package/root reader files explicitly named in this audit. The remaining two current Markdown files are the current reader-facing report/index documents discovered during the full enumeration: `src/drl_pinball/eval/output/reports/comparison_table.md` and `src/CCD_analysis/evidence/real-ccd-karman-figures-v1/KARMAN_INTERPRETATION.md`; both were read-full but are derived/historical reader views and do not create new authority. This makes the denominator 63.
+
+### Archive navigation (excluded from current denominator)
+
+- `src/drl_pinball/train/CROSSRE_ANALYSIS.md` and `VARDIST_ANALYSIS.md`: **archived transfer analyses**, not current reader docs. They describe `_tr` transfer experiments and deprecated calibration/training recommendations; current contracts are in `train/README.md`, `TRAIN_PIPELINE.md`, and `DESIGN_DECISIONS.md`. Their historical observations remain navigation only.
+- `src/drl_pinball/train/old/`, `train.local.bak/`, and `eval/archive/`: old/backup/deprecated navigation, excluded without reading leaf files.
+- `src/SR_analysis/archive/` and `src/CCD_analysis/original_ccd/`: archive/superseded routes. Current package READMEs and asset ledgers establish their boundaries; archive leaves were not read.
+- `src/OID_analysis/archive/`: the July two-scene conditional study is archive evidence only. The active OID package is claim-free and has no current scientific result.
+- Superseded steady publication/display routes are navigated through `steady_pinball_theory/evidence/README.md` and `INDEX.json`; superseded leaves were not read.
+
+## 3. Selected current authority/index JSON ledger
+
+The following **8 JSON documents were read in full** and retained as selected current authority/index documents:
+
+- `src/drl_pinball/train/results/training_csv/manifest.json`
+- `src/drl_pinball/data/reproduction_plots/manifest.json`
+- `src/drl_pinball/data/reproduction_plots_sr/manifest.json`
+- `src/SR_analysis/results/catalog.json`
+- `src/CCD_analysis/EXECUTION_CHECKPOINT.json`
+- `src/CCD_analysis/artifact_catalog.json`
+- `src/OID_analysis/SCIENTIFIC_RESET_CHANGESET.json`
+- `src/steady_pinball_theory/evidence/INDEX.json`
+
+Generated result tables, plotting manifests, and external artifact manifests not in this selected set remain excluded from the denominator. In particular, large `training_iterations.csv`, plot CSV/NPZ/PDF/PNG trees, model/TensorBoard bundles, external Optane field roots, and hidden raw data are not “unread current documents”: they are generated/raw artifacts governed by manifests and authority documents. No hash recomputation, raw-field read, or artifact audit was performed.
+
+## 4. New full-read addenda
+
+### A7 — DRL archive boundary is semantic and explicit
+
+`ASSET_STATUS.md`, `train/README.md`, `TRAIN_PIPELINE.md`, and both transfer analyses distinguish active native V5 scratch IDs from archived `_tr` transfer work. `CROSSRE_ANALYSIS.md` and `VARDIST_ANALYSIS.md` are therefore archive navigation, not current reader authority. Their recommendations (generic `SIM_BP`, transfer learning, learning-rate changes, and historical peak/crash explanations) must not be carried into current active claims or counted as current coverage.
+
+The current DRL reader package also makes the evidence boundary sharper: retained V5 results are historical writing inputs, Re100 alone has five seeds, other cases are n=1 demonstrations, and shared train/eval updates `VecNormalize` moments. A best policy is incomplete without its same-selection normalizer; a logged episode is a PPO chunk plus evaluation, not a terminating Gym episode.
+
+### A8 — Current generated reader manifests are partial by contract
+
+The training CSV manifest reports 7,000 rows from 14 of 15 discovered runs and excludes `kar_re60_seed43` fail-closed because of conflicting resumed TensorBoard branches and mixed configuration history. The SR plotting manifest is intentionally partial (11 of 23 expected plots), and its generalization cases intentionally have no PPO role. These are honest manifest states, not current-document coverage failures. The generated plots remain reader views and do not create new CFD evidence.
+
+### A9 — SR current/archive status is a semantic view, not filesystem movement
+
+The full SR read confirms a mixed executable/runtime tree: Kármán is the only active scientific claim; steady is contextual; Illusion runtime and immutable article/article2 paths remain for reproducibility but are historical/negative. `results/catalog.json` says `v2`/`v3` suffixes do not prove supersession and discovery generations remain pending dependency audit. The long standardized Illusion 1L result is a negative control against the older positive interpretation; no target-tracking or cross-size claim may be promoted.
+
+### A10 — CCD is closed descriptively, with canonical choice still pending
+
+The CCD package and checkpoint identify the validated per-cycle sibling as complete and the preferred pinball-inclusive result as a derived reader view. The primary bounded conclusions are mean-control dominance (`93.42%` of the tested ROI reduction), a stable rank-3 descriptive residual subspace, and near-equivalence to raw full-fit POD in the pinball-inclusive domain. The checkpoint still records `canonical_promotion_choice=PENDING_USER_CHOICE` and final review as writing claim closeout; therefore “validated sibling complete” must not be rewritten as “canonical alias frozen.”
+
+### A11 — OID remains method-only; reset JSON records non-destructive preservation
+
+The current OID Markdown and reset changeset agree: the active package is CPU-only, claim-free, and has no active scientific result or redo entrypoint. The archived two-scene study remains conditional; sensor self-observation, failed force gates, and feedback-loop action association prohibit any importance, mechanism, causality, necessity, or authority claim.
+
+### A12 — Steady current index separates authority, supersession, and generated display
+
+`evidence/INDEX.json` and the current steady readers identify v3b and display-v2 as current, while older article/display/causal/return artifacts are superseded or diagnostic. The accepted `[0,+Omega,-Omega]` free-slip `Re_D=50` endpoint is bounded; potential cancellation fails by sign (`g_opt=+6.147944` versus descriptive `g=-12.200554`); NS diagnostics are partial and do not establish force ownership, closure, causality, or stability. The figures manifest is CPU-only derived presentation, not evidence.
+
+### A13 — Cross-package source contradictions remain blockers, not omissions
+
+The full read confirms the OID physics contract’s unresolved sign disagreement (`[0,-5.1,+5.1]` in archived/source material versus stated CCD `[0,+5.1,-5.1]`) and the separation of Legacy/V5 action semantics. These are not solved by the coverage pass. The accepted steady branch is governed by its current wall-velocity/action contract; historical OID material remains quarantined.
+
+## 5. Conflicts and justified deferrals
+
+1. CCD documentation is complete at the validated suffixed sibling, but canonical alias promotion remains user-pending.
+2. OID source/document sign disagreement remains unresolved and must not be called corroboration.
+3. SR current Kármán claims and historical Illusion evidence share mixed immutable paths; use semantic claim status, not physical relocation.
+4. V5, Legacy SR, CCD, OID, and steady results remain separate evidence chains; no numerical pooling is licensed.
+5. Large/external generated artifacts are authority-led deferrals only. This audit did not read raw fields, NPZ/NPY payloads, model bundles, TensorBoard streams, or archive leaf files.
+
+## 6. Completeness verdict
+
+**Document reading completeness: PASS.** All 63 current Markdown documents in the corrected denominator and all 8 selected current authority/index JSON documents were read-full. Archive/old/superseded material was removed from the current denominator and listed as navigation; the two DRL transfer analyses are explicitly archived rather than mapped-only current documents.
+
+**Information completeness: BOUNDED / not yet synthesis-clean.** The full read added A7–A13 and confirms that current reader coverage is complete, but canonical CCD alias choice, OID sign conflict, current-vs-historical SR semantics, partial generated manifests, and artifact-level deferrals remain explicit. Full document reading does not imply independent verification or artifact audit.
+
+## 7. Round-2 blocker
+
+Round 2 structure/narrative planning is **not blocked by document coverage**. It is blocked from an unqualified “all current authority is canonicalized and synthesis-ready” label until: (i) the CCD canonical-promotion choice and final writing closeout are resolved, (ii) the OID sign conflict is explicitly quarantined, (iii) the SR active/archive claim matrix is carried forward, and (iv) each manuscript number is paired with its artifact, metric, window, comparator, realization count, and evidence status. Do not reopen CFD, training, raw-artifact audit, archive reading, or literature deep search as a side effect of this coverage pass.
+
+## 8. Audit record
+
+- Working Memory: read first through the Nowledge Mem context bundle; used only for routing and current scope, never as repository evidence.
+- Full-read result: **63 Markdown + 8 selected authority/index JSON = 71 files read-full**.
+- New findings: **A7–A13 (7 addenda)**; two formerly mapped DRL analyses were removed from current denominator.
+- Computation: no CFD, CUDA, training, hash recomputation, or immutable-artifact modification.
+- Output: this file only; original five dossiers were not modified and no commit was created.

docs/JFM_WYQ/ROUND1_BOTTOM_UP/07_CURSOR_PLANS_HISTORY_INDEX.md

@@ -0,0 +1,111 @@
+# JFM Round 1 — Cursor Plans / History Index
+
+> **定位。** 本文件是 Cursor plans 的历史意图、试错和工程演进索引，不是结果报告，也不是 manuscript prose。Plans 记录“当时打算做什么、如何排错、哪些路线被考虑过”；它们不能证明任务执行、数值结果、科学结论或当前状态。当前事实必须回到仓库 authority 文档、可执行代码、测试、manifest 和 immutable artifacts。
+
+## 0. Search and scope record
+
+- **Search root:** `/home/frank14f/.cursor/plans/`；只读检索，没有修改 plans。
+- **Search rule:** 递归检查所有 `*.plan.md` 的文件名和内容，关键词为 `DynamisLab|CelerisLab|Legacy|pinball|DRL|SR|CCD|OID|steady|cloak|illusion|vortex|erase`，大小写不敏感；另以 `Legacy/V5/rewrite/refactor/inference/norm/observation/action/reward/unit-test/rejected/superseded` 做主题复核。
+- **Actual lexical hits:** **167 plan files**（内容或文件名至少命中一项；不能把 167 解释为 167 个独立研究项目）。一部分是 solver/CUDA 低层计划，仅在影响 solver semantics、readback、body/action mapping 或 provenance 时纳入。
+- **Known direct plans:** `dynamislab_karman_re100_rewrite_plan_2d4c4599.plan.md`、`dynamislab_inference_rewrite_plan_903dcdc8.plan.md`、`dynamislab_unitest_migration_79f3bf20.plan.md`、`dynamislab_refactoring_plan_970406f6.plan.md`；另有 `unitest_migration_v3*.plan.md`、`pinball_unitest_migration_plan_d13ccb17.plan.md` 等家族。
+- **Recall:** 已读取 Nowledge Mem context bundle/Working Memory，并作针对 DynamisLab、Legacy/V5、SR、CCD、OID、steady、inference、normalization 的 durable-memory recall。没有匹配的 prior thread；Memory 仅用于导航，当前 repository docs 覆盖 Memory。
+- **Coverage interpretation:** `hit` = lexical discovery；`indexed` = 本文件提炼；`read` = 计划全文已读。167 个命中计划并非全部逐篇全文阅读；工程重复家族和纯 solver 计划采用 mapped/indexed 状态。历史计划的 todo `completed` 仍按 planned-history 处理，绝不自动转成结果。
+
+## 1. Source ledger
+
+### 1.1 Current authority used for comparison
+
+| Current source | Role | Status |
+|---|---|---|
+| `docs/JFM_WYQ/ROUND1_BOTTOM_UP/README.md` | Round-1 non-prose/evidence-index rules | read |
+| `01_DRL_CONTROL_AND_CASES.md` | DRL cases, V5/Legacy contracts and boundaries | read |
+| `02_SR_CONTROL_LAWS.md` | Legacy SR pipeline, formula and Illusion negative boundary | read |
+| `03_FIELD_CCD_OID.md` | corrected CCD authority, OID reset and cross-map | read |
+| `04_STEADY_THEORY_AND_CROSSMAP.md` | steady contract, sign, metric and open NS closure | read |
+| `05_INTRODUCTION_LITERATURE_INDEX.md` | narrative vocabulary and prohibited claims | read |
+| `src/drl_pinball/WRITING_HANDOFF.md` | current DRL claim boundary | read |
+| `src/SR_analysis/HANDOFF.md` | current SR scope and runtime contract | read |
+| `src/CCD_analysis/WRITING_HANDOFF.md` | bounded CCD interpretation | read |
+| `src/OID_analysis/HANDOFF.md` | claim-free active OID core | read |
+| `docs/JFM_WYQ/CELERISLAB_VALIDATION_DOSSIER.md` | solver validation boundary | mapped via dossier |
+
+### 1.2 Plan-family ledger
+
+| Family / representative paths | Historical contribution | Current comparison / status |
+|---|---|---|
+| Legacy/new API: `dynamislab_karman_re100_rewrite*`, `dynamislab_inference_rewrite*`, `legacy-test-and-reproduce*`, `reproduce_inference_plan*`, `new_cfd_reproduction*` | Proposed frozen Legacy reference plus parallel new `Simulation` inference; A/B validation; inlet/readback repair | Current DRL docs confirm separate chains and contract differences, but not universal parity. `read/indexed`; planned parity is `unverified/superseded`. |
+| UniTest/body order: `unitest_migration_v3*`, `pinball_unitest_migration*`, `dynamislab_unitest_migration*`, `action-obs_unified*` | Add-sequence body order, target-vs-experiment separation, scene-specific observation and normalization | Current executable contracts confirm these lessons; `confirmed-by-current-doc` for provenance, not for every planned test. |
+| DRL training/evaluation: `karman_cloak_training_pipeline*`, `karman_cloak_multi-seed-training*`, `karman_cloak_new_training*`, `karman_cloak_retraining*`, `eval-*`, `clean_reproduction_reset*` | Training, retained checkpoints, reproduction, benchmark and cleanup routes | `01_DRL...` artifact ledger wins: retained V5 evidence is named-condition, seed-dependent and not post-fix retraining. Planned outcomes are `historical-only`. |
+| SR Legacy: `sr_legacy_pipeline*`, `sr_analysis_restructure*`, `可信sr全流程重建*`, `joint_sr_article_flow*`, `sr_final_consolidation*`, `sr标准化采集*`, `sr项目收尾整理*` | Freeze candidates; order/trajectory/G audit; Stage-3 closed-loop arbiter; Target/PPO/SR/Zero standardization | `02_SR...` and SR handoff confirm active Kármán only, steady contextual, Illusion negative. Earlier broad/generalization routes are `superseded`. |
+| Illusion/vortex/erase: `illusion-env-and-server-deploy*`, `signature_line_and_illusion_completion*`, `sindy-sr_illusion*`, scene-matrix and vortex plans | Target harmonics, event semantics, erase norms and attempted transfers | Current dossier keeps PPO capability separate; Illusion SR negative; Vortex/Erase excluded from common periodic L2. `historical-only/superseded`. |
+| CCD: `ccd_pipeline_*`, `ccd_analysis_round_*`, `ccd_three-part_reset*`, `ccd_corrected_core*`, `ccd_近体/模态/写作*`, `karman_dynamic_increment*`, `minimal-acquisition-upgrade*` | Successive resets from temporal/phase/correction-field routes to exact-time, mean/dynamic and cycle-template descriptive analysis | `03_FIELD...` confirms current bounded rank-3 and CCD/POD subspace result; causal/mechanism/lag claims remain prohibited. Earlier routes `superseded`. |
+| OID: `oid_*`, `li22b-replication-oid-deep*` | Reset from sensor/force/action “importance” to claim-free Schlegel LR/LE core | Current OID handoff confirms no active scene result or redo runner. Historical positive route `superseded`. |
+| Steady/cloak: `steady-cloak*`, `steady理论*`, `steady结果*`, `accepted-direction_semi-analytic_closure*`, `ns_cloak_mechanism_report*`, `rigorous_steady*`, `稳态pinball解析验证*` | Wrong-sign branch, potential-flow overreach, double subtraction, incomplete CV closure, NS-first/gap-jet hypothesis | `04_STEADY...` confirms accepted `[0,+Omega,-Omega]`, direct `q_in` metric and bounded endpoint; mechanism open. Rejected theory routes `superseded`; plan completion flags are not authority. |
+| Governance/solver: `项目总文档收尾*`, `ccd-project-cleanup*`, `pre-release_audit*`, `train_release_cleanup*`, `paraview*`, `karman-flow-field-plots*`, plus `esopull*`, `runtime_body_*`, `forceregion*`, `body_module_*` | Navigation, hashes, no-clobber, renderer, readback, topology and compile-risk history | Mapped only where affecting provenance or physical semantics; otherwise `historical-only` engineering context. |
+
+### 1.3 Coverage ledger
+
+| Population | Status | Reason |
+|---|---|---|
+| 167 lexical-hit plans | `indexed/mapped`, not all `read` | Broad matching catches generic solver notes and repeated plans; no claim of full-text extraction from each file |
+| Direct `dynamislab_*` plus named migration/reset plans | `read` | Most relevant to Legacy→V5, inference, unit tests, refactor and route history |
+| SR/CCD/OID/steady package plans | `read` or `mapped` | Representative and late reset/closeout plans establish evolution |
+| Pure CUDA/performance/visualization plans | `mapped` | Deep reading deferred unless contract/provenance consequence was visible |
+| External artifacts referenced by plans | `external-inaccessible` / deferred | Plans cannot replace manifests or mounted payloads |
+
+## 2. Timeline and themes
+
+1. **Legacy scene expansion:** original `FlowField`/LegacyCelerisLab plans cover steady, Kármán cloak, Illusion, vortex, Erase and reduced-observation routes. Current dossiers separate periodic, transient, steady and target-illusion contracts.
+2. **Inference before retraining:** rewrite plans prioritize a new inference/evaluation layer, precomputed target/norm metadata, checkpoints, scene registry, fields and old-vs-new comparison. This explains current provenance discipline; it does not prove planned parity or new results.
+3. **Why Legacy→V5 was a rewrite:** the plans identify scientifically consequential differences: cumulative-versus-time-averaged readback, separate body/sensor APIs, add-order IDs, action slots, action smoothing, inlet/BC settings, grid and normalization. Current DRL authority confirms V5 is a deliberate new 2000×600/native-v2/VecNormalize chain, not a numerically identical continuation.
+4. **Unit-test/refactor background:** migration plans aim to port the multi-scenario notebook into reusable helpers/runners and test old/new side by side. Refactor motivation includes cache identity, no-clobber output, exact scene contracts and avoiding duplicated scripts. These are relevant only where they protect observation/action/reward/solver provenance.
+5. **SR closeout:** broad candidate, joint, Illusion and generalization plans narrow to Legacy Kármán; Stage 3 closed-loop CFD becomes the arbiter; standardized Target/PPO/SR/Zero roles preserve failure telemetry. Current SR docs confirm this scope.
+6. **CCD correction:** repeated plans withdraw older temporal/phase and causal-looking routes in favor of exact clocks, common masks, mean-field decomposition and cycle-template residuals. Current CCD supports descriptive rank-3/subspace claims only.
+7. **OID reset:** the sensor/force/action importance route is explicitly invalidated by self-observation, feedback confounding, failed force gates, wrong method naming and incomplete provenance; active OID is now a claim-free CPU core.
+8. **Steady reset:** wrong sign, potential-flow/circulation overreach, double-subtraction and incomplete NS ledgers forced an NS-first boundary. Current accepted branch and metric are bounded; gap/base-bleed is a candidate, not established mechanism.
+9. **Evidence governance:** late plans increasingly emphasize hashes, immutable archives, exact clocks, role manifests and fail-closed gates. This is durable workflow history, not proof that every planned gate ran.
+
+## 3. Current relevance matrix
+
+| Plan-derived lesson | Current label | Current evidence boundary |
+|---|---|---|
+| Body add order controls IDs and policy channels | `confirmed-by-current-doc` | Relevant to observation/action provenance; executable code wins |
+| Legacy fixed `norm.json` must not be replaced by fresh norms; V5 calibration/VecNormalize is different | `confirmed-by-current-doc` | Relevant to normalization and comparison semantics |
+| Legacy readback accumulation/division and new read methods differ | `confirmed-by-current-doc` | Relevant inference/unit-test background; exact implementation remains code authority |
+| V5 is intentionally a different solver/configuration chain | `confirmed-by-current-doc` | Safe bounded explanation of rewrite; no claim of equivalence |
+| Planned bitwise-equivalent new/old rollouts | `unverified` | No universal parity artifact found in current dossier; do not state as result |
+| Legacy and V5 should be pooled in one metric matrix | `superseded` | Prohibited without explicit parity contract |
+| Illusion SR target tracking/generalization/efficacy | `superseded` | Long-window result is historical/negative and near physical zero |
+| OID importance/mechanism/active action result | `superseded` | Active package is claim-free method core |
+| Old CCD temporal lag or causal mechanism | `superseded` | Corrected CCD is descriptive; lag is not physical delay |
+| Potential cancellation as accepted finite-Re mechanism | `superseded` | Sign conflict and incomplete viscous closure remain explicit |
+| Completed todo = completed science | `historical-only` | Plans are not result authority |
+| Missing parity artifacts, external role trees, leaf hashes, post-fix retraining | `potentially-relevant-gap` | Provenance/evidence boundary addenda; not numerical claims |
+
+## 4. Rejected / superseded lessons
+
+- The broad “all scenes / all metrics / all pipelines” narrative was narrowed. Active positive SR is Legacy Kármán/cloaking only; steady is contextual; corrected CCD is descriptive; OID is claim-free; Illusion SR is negative/historical.
+- Potential-flow resemblance, circulation cancellation, attractive plots and incomplete energy/momentum ledgers cannot establish the accepted finite-Re mechanism. Gap-jet/base-bleed remains a hypothesis pending matched intervention, wall-resolved diagnostics and complete NS budgets.
+- Archived OID high-R² or sensor/force comparisons do not establish importance, controllability, causal action effect or mechanism.
+- A generic Legacy/V5 inference pipeline was rejected in favor of chain-specific contracts. Recomputing norms, changing action sign, changing body order, or mixing requested/effective action changes the experiment.
+- DRL reward, SR DTW, CCD singular strength, OID reconstruction and steady `E_inf_vector` are distinct estimands; they cannot be ranked or pooled.
+- “Unit tests pass” and a plan todo marked `completed` are implementation/process evidence at most, never physical validity or mechanism evidence.
+
+## 5. Potential addenda to Round 1
+
+1. Add a short provenance-history paragraph: Legacy remained an immutable reference because API migration changed readback, body/action mapping, normalization and smoothing; V5 was a deliberate new chain.
+2. Add a supplement-level contract table: solver → scene → observation order → action scale/bias/smoothing → norm/calibration → reward/DTW → field comparator, with Legacy and V5 separate.
+3. Add a negative-route paragraph explaining why Illusion SR, OID importance and early CCD/steady mechanism routes were closed; do not turn this into positive evidence.
+4. Add a provenance gap note for exact parity artifacts, external role trees, complete leaf hashes and post-fix retraining status.
+5. Add a physics-gap note: recurring rear-rotation and a smaller dynamic residual are bounded observations; action→near-body→wake→signature remains candidate logic.
+6. Treat plan dates/todo states as decision chronology only; use authority artifacts for experiment dates and completion.
+
+## 6. Explicit non-authority warning
+
+Cursor plans are **historical process records only**. A plan’s `completed`, `in_progress`, or `pending` field is not a result status. Planned thresholds, commands, approval gates, “bitwise-equivalent,” “JFM-ready,” “mechanism completion,” or “final” wording must not be quoted as facts without current authority and artifact provenance. Archive/obsolete plans are navigation/history, not current instructions. Memory aids recall but never outranks current docs, code, manifests or immutable artifacts. Legacy/V5, periodic/transient/steady and target/zero/controlled roles remain separate unless an explicit new parity contract proves comparability.
+
+## 7. Minimal handoff
+
+- **Plans hit:** 167 lexical matches under the stated keyword search.
+- **Narrative/evidence impact:** no new positive scientific claim or scope expansion is authorized. The history does sharpen the paper boundary: explain Legacy→V5 as a provenance- and contract-driven rewrite, keep chains separate, and preserve unresolved parity/hash/retraining gaps.
+- **Mutation scope:** only this file was written; plans and other repository files were not modified; no commit was created.

docs/JFM_WYQ/ROUND1_BOTTOM_UP/08_EXPERIMENT_AND_CASE_METRICS.md

@@ -0,0 +1,243 @@
+# JFM Round 1 — Experimental + Case-Specific Metric Addendum
+
+> **用途**：证据/指标 addendum，不是 manuscript prose。只读审计仓库与 Undermind 现有材料；未启动 CFD、未修改代码或 Undermind workspace。
+>
+> **证据标签**：`[AUTHOR]` 作者陈述/记忆；`[REPO]` 仓库 authority、schema 或代码；`[UM-REPORT]` Undermind 已有报告；`[LIT]` 外部文献背景；`[OPEN]` 未闭合问题。
+>
+> **状态标签**：`implemented`、`existing-data-computable`、`requires-new-offline-code`、`requires-new-acquisition`、`unsupported`。`mapped` 不等于 `read`，`read` 不等于 `independently verified`，均不等于 `artifact-audited`。
+
+## 1. Scope and evidence boundary
+
+本 addendum 的问题不是证明实验已验证 CFD，而是：在作者给定的实验事实和现有 CFD/analysis artifacts 下，哪些 case-specific comparison 可以诚实地计算、怎样命名、需要什么新增数据，以及哪些只能作为未来实验方向。
+
+### 1.1 实验事实：作者来源，与文献背景分开
+
+以下仅作为 `[AUTHOR]`，不由文献建立，也不宣称已有独立 artifact 验证：
+
+- 系统由 Frank 从零搭建；在循环水洞中截取试验段作为静水段。
+- 直线导轨和齿条驱动天车式运动平台；平台搭载力传感器、ADC、树莓派、电机驱动、圆柱结构和电源。
+- Python 软硬件层使 DynamisLab 策略易于部署。
+- 目标速度和水动力很小；背景热对流显著，且有效力反馈难以测量。
+- 实际只完成开环 steady cloak 与 illusion demonstration；可见证据是染色液迹线。steady 液迹线较直；illusion 可见脱涡频率变化。
+- 没有完整同步 PIV/force validation package。因此实验只能定位为 **bounded feasibility / future direction**，不能写成定量验证、闭环实验控制、力学机制确认或 field restoration。
+
+### 1.2 实验系统/状态来源分级
+
+| 来源 | 当前可写成什么 | 不可写成什么 | 状态 |
+|---|---|---|---|
+| 作者陈述：水洞静水段、运动平台、传感器/ADC/Raspberry Pi/电机驱动/Python | apparatus design、deployment feasibility、engineering motivation | 已验证精度、同步误差、传感器不确定度 | `[AUTHOR]`; no visible artifact audit |
+| 作者陈述：低速、小力、热对流、force feedback 困难 | bounded limitation and reason for open-loop scope | 精确背景速度、力噪声、SNR、因果归因 | `[AUTHOR]`; no synchronized package |
+| 染色液迹线（steady/illusion） | qualitative visualization；steady 较直、illusion shedding frequency visibly changed | quantitative velocity/vorticity/force reduction、完整 field match、causal mechanism | `[AUTHOR]`/reader description; raw synchronized artifact not located |
+| 水洞/油洞 literature | facility plausibility、design trade-offs、measurement cautions | present apparatus 的内部结果或 dominant mechanism | `[LIT]`/`[UM-REPORT]` |
+| CFD evaluator/reproduction artifacts | named CFD case 的 signal/field comparison | physical experiment validation or transfer to apparatus | `[REPO]`; Legacy/V5 must remain separate |
+| Confirmation report navigation | historical/current material locator；steady sensor-plane figure candidate | automatically current, independently verified experimental authority | `[UM-REPORT]` navigation only; local source package not found |
+
+**写作边界**：实验段最多承担“平台可部署、定性现象可观察、低速实验仍受热背景与小力限制”的可行性/展望角色；定量主张来自明确的 CFD/analysis artifact，而不是染料图像或作者记忆。
+
+## 2. Sources inspected and coverage ledger
+
+### 2.1 Nowledge Mem
+
+- `read_context_bundle`：已读取 Working Memory。当前相关边界包括：active positive scope 受限于 Kármán/cloaking；Illusion SR 为 historical/negative；steady 为 contextual；Legacy 与 V5 不得合并；DTW 不能解释为 physical delay/causality。
+- `memory_search`：定向检索 `experiment / water tunnel / oil tunnel / towing / DTW / L2 / Confirmation / JFM sensor evaluation`，命中作者实验状态、实验限制和 case-specific metric guidance 的 durable memory。
+- `thread_search`：同主题未返回匹配 thread；因此 Mem 用于路线/边界回忆，不作为外部科学证据。
+- 未保存新的 memory：本文件是当前指定 addendum，不引入新的长期决策；作者事实按 `[AUTHOR]` 保存于本文件而非升级为文献事实。
+
+### 2.2 Undermind 实际覆盖
+
+| workspace/material | 覆盖内容 | 层级/状态 |
+|---|---|---|
+| [Oil-tunnel workspace](https://app.undermind.ai/projects/add6f965-60ca-446e-becd-26e7892263b4) | `Residual motion in nominally still water`；`Low Re pinball experiment design` | 两份 report 全文读取；`[UM-REPORT]`，其引用论文为 report-indexed，未在本 addendum 声称逐篇 full-PDF audit |
+| [DynamisLab workspace](https://app.undermind.ai/projects/4383d145-2c80-4991-8a9c-7ac2dd683438) | root tree、`Confirmation_report/`、`JFM/`、`JFM References/`、相关现有 reports/deep-search 状态 | folders/files mapped；Confirmation/JFM navigation inspected；不把导航当内部结果 |
+| Confirmation | `Conf_1_Intro`–`Conf_5_Conclusion` 的可见导航/文件条目，尤其 Description/Method/Result/Conclusion | mapped/read at report-navigation level；未发现可直接作为 current local authority 的 `.tex/.bib` 或 synchronized experiment bundle |
+| Existing JFM material | `Minimal wake-field L2 evaluation protocol`、`Wake-field L2 evaluation and downstream sensor rationale`、`Wake to force`、CCD/OID/analytical reports | existing Undermind reports indexed/read where relevant；background/evaluation rationale，不覆盖内部 artifact |
+| Existing deep searches | 14 searches in DynamisLab tree | status/list only；没有新 deep search；未修改 workspace |
+
+**Not covered / external-inaccessible**：Oil-tunnel 中未见 raw PIV、force time series、calibration files 或 current experiment report；Confirmation 的本地原始同步数据、图源 schema 和 sensor-plane export 未在可见仓库/Undermind navigation 中定位。它们应标为 `requires-new-acquisition` 或 authority-led deferral，不得由文献或图片推断数值。这里不包括已读完整的 `WAKE_L2_AND_SENSOR_PLACEMENT_RESEARCH_NOTES.md`；该 note 已纳入 read-full coverage。
+
+### 2.3 Repository documents/code mapped or read
+
+| package | inspected authority/navigation | status |
+|---|---|---|
+| `src/drl_pinball/DATA_DICTIONARY.md`, `legacy_test/README.md` | role semantics, phase/event/late schemas, acquisition windows, native/normalized DTW | read |
+| `src/drl_pinball/eval/README.md`, `eval/wake_l2.py`, `eval/tests/test_wake_l2.py` | V5 evaluator bundle, `drl-pinball-wake-l2-v2`, ROI RMS L2, complete-cycle mean, phase8, frequency diagnostics | read; code contract inspected, not rerun |
+| `src/drl_pinball/RESULTS.md`, `CLAIMS.md`, `WRITING_HANDOFF.md` | result families, evidence states, V5/Legacy boundaries and non-claims | read |
+| `src/SR_analysis/README.md`, `HANDOFF.md` | active Kármán SR, steady contextual, Illusion negative/historical, DTW limits | read |
+| `src/CCD_analysis/README.md`, `CLAIMS.md`, `FIGURES_AND_DATA.md` | current descriptive field/mode artifacts and mask/phase semantics | read |
+| `src/steady_pinball_theory/README.md`, `RESULTS.md`, `metrics.py` | steady endpoint/profile/L2 utilities, partial ledger and force/torque limits | read |
+| `docs/JFM_WYQ/ROUND1_BOTTOM_UP/01–05_*.md` | Round-1 DRL/SR/field/steady/introduction dossiers | mapped; 05 read; other dossier claims used only through their current handoff/authority references |
+| `docs/JFM_WYQ/WAKE_L2_AND_SENSOR_PLACEMENT_RESEARCH_NOTES.md` | wake-field L2, registered ROI, phase treatment, downstream sensor geometry, literature rationale and claim boundaries | **read-full** (595 lines); implementation authority cross-checked against `src/drl_pinball/eval/wake_l2.py` and acquisition/schema docs |
+
+## 3. Oil-tunnel literature index and reading levels
+
+### 3.1 Primary index from existing Oil-tunnel report
+
+The report `Low Re pinball experiment design` recommends a covered recirculating channel with mineral oil, dye visualization and time-resolved PIV as a low-Re facility direction. Its relevance is design background, not evidence that DynamisLab has such an oil tunnel. The report explicitly contrasts water/towing limitations with higher-viscosity oil operation. The present author-described facility is instead a recirculating **water** tunnel with an isolated still-water test section; the oil report is therefore a future-design comparator, not a description of the completed apparatus.
+
+The report `Residual motion in nominally still water` argues that nominally still water can retain organized motion from thermal boundaries, free-surface cooling and residual towing. It is directly relevant to the author’s low-speed force/velocity limitation, but it remains a literature synthesis rather than an independently measured DynamisLab background-flow dataset.
+
+| report-indexed cite key | topic used here | reading level in this addendum |
+|---|---|---|
+| `Ana76` | background motion in temperature-controlled towing tank; dye-line measurement and thermal mitigation | full-report synthesis; paper identity/DOI must be resolved before formal manuscript citation |
+| `Kam72`, `Jen21` | post-towing residual current, settling, sloshing and facility uncertainty | full-report synthesis; background only |
+| `Spa61`, `Vol99` | free-surface evaporation/cooling and plume motion without discrete heater | full-report synthesis; mechanism plausibility only |
+| `Wan05`, `Mul04`, `Cau16` | weak differential heating, wall plumes and mm/s-scale structured circulation | full-report synthesis; not present-tank quantification |
+| `Gad83`, `Leo99`, `Zep64`, `Nor04`, `Huu14`, `Son11` | passive wall conduction, mixed materials, stratification, surface effects | report-indexed/secondary support; formal identity audit pending |
+| `Mun07`, `Han04`, `Cha11`, `Ban17`, `Pau13`, `Den18b` | oil/low-Re tunnel design, pinball/rotating-cylinder measurement and wake context | report-indexed design background; not internal experiment evidence |
+
+### 3.2a Read-full wake-field/sensor rationale note
+
+The repository note `docs/JFM_WYQ/WAKE_L2_AND_SENSOR_PLACEMENT_RESEARCH_NOTES.md` was **read in full** and is the implementation/rationale authority for the following addendum points (not copied wholesale here):
+
+- Sensor DTW and full-field L2 answer different questions: six downstream velocity channels are a compact signature/training diagnostic, while spatial velocity-field L2 is the independent field check. Low DTW must not be presented as global field equivalence.
+- The registered ROI is `-6 <= (x-x_s)/D <= 14`, `|(y-y_0)/D| <= 5`, a `20D x 10D` fixed rectangle. It is geometry-guarded, not solver-mask-proven, because exact solver masks were not persisted; zero velocity is never a solid-mask proxy. The older `[-8D,12D]` proposal is superseded.
+- The three-probe layout is one plane at `x/D ~= 10`, transverse positions `y/D = +2, 0, -2`, radius `0.25D`, returning both components (six channels). Centre `u_y` is phase-sensitive; upper/lower pairs provide transverse coverage and symmetric/antisymmetric combinations; `u_x` tracks deficit/recovery. This is physically motivated signature monitoring, not optimal placement or guaranteed observability.
+- Direct pinball precedent at `x/D=10` supports reasonableness, while closer multi-plane sensing studies support the delay/coverage trade-off. The relevant indexed identities are `Rai20`, `Mac20`, `Li22b`, `Mar24b`, `Jin21`, `Gon19`, `Nai20b`, `Ish17`, and `Par20`; exact bibliography resolution remains required before manuscript citation.
+
+### 3.2 Reading hierarchy for eventual manuscript use
+
+1. **Tier A — full-PDF + exact metadata/DOI required before citation**: closest facility papers (`Ana76`, `Kam72`, `Mun07`, `Cha11`) and any paper used for a quantitative apparatus claim.
+2. **Tier B — full-report extraction acceptable for Round-1 index, not yet final bibliography**: thermal/free-surface mechanism papers (`Spa61`, `Vol99`, `Wan05`, `Mul04`, `Cau16`, `Gad83`, `Leo99`).
+3. **Tier C — abstract/snippet or design context only**: broad low-Re facility and application references; use only to motivate measurement choices, not numerical values.
+4. **Internal evidence**: author facts, Confirmation figures, raw experiment files, calibration and synchronization metadata. These must be separately audited and never replaced by literature.
+
+No DOI or exact bibliographic identity is invented here. Before manuscript prose, resolve the report cite keys through Undermind `get_paper_info(show_doi=true)` or the formal bibliography and upgrade only the references actually used.
+
+## 4. Metric contracts and case matrix
+
+### 4.1 Common notation
+
+- `L2*`: normalized spatial velocity error on a declared common fluid ROI/mask, using the same `U0`, coordinates, grid/weights and comparator; state whether it is RMS (`wake_l2.py`) or area-integral (`steady_pinball_theory.metrics.quadrature_l2`).
+- `DTW_native`: existing objective/reward-compatible signal metric; preserve its original normalization and window.
+- `DTW_norm`: current/new normalized sensor-cycle DTW, after explicit channel normalization; similarity only, never physical delay or causality.
+- `force`: only force sensor/force-history data with calibration, units, clock, sign and body ordering; no force claim from reward or dye.
+- `profile`: a declared downstream sensor-plane or station profile, including relative velocity deficit convention, y coordinates, mask and `U_inf`.
+- `phase/event`: periodic phase is a phase coordinate derived from declared crossings; event fields use relative offsets and must not be relabeled as phase slots.
+
+### 4.2 Scene matrix
+
+| scene / comparator | DTW | L2 / profile | force | phase/event semantics | status now |
+|---|---|---|---|---|---|
+| **Periodic Kármán / Illusion**: target vs controlled, plus physical zero where valid | `DTW_native` exists in role contract; `DTW_norm` exists in standardized acquisition path or is existing-data-computable when retained sensors/target channels are present | `E_mean`: complete-cycle mean field; `E_phase8`: eight independently phase-referenced nearest snapshots; compare separately. Use current `drl-pinball-wake-l2-v2`, not one pooled score | CFD force channels may be in timeseries, but force is not automatically an experimental force validation; report only with explicit role/schema and calibration | eight canonical phases `[0, π/4, …, 7π/4]`; crossings independently selected for target/controlled/zero. `phase_fields.npz` is eight fields; no repeated-crossing ensemble. | `implemented` for retained periodic CFD roles where artifacts pass schema; `existing-data-computable` for any complete retained role not yet summarized |
+| **Vortex candidate**: acquired transient downstream ROI | current DTW on retained sensor timeseries; normalized DTW may be computed if target/reference channels and window exist | candidate: per-snapshot downstream-ROI L2 at several acquired event offsets; do not average event snapshots into phase8. Current event schema supports five `event_fields.npz` snapshots with `relative_offsets`; formula/application still needs an offline evaluator wrapper | unavailable unless force history is present and calibrated; no force from velocity alone | `event_fields.npz` uses relative offsets/event labels, not periodic phase. “midpoint” must identify the acquisition event and offset, not imply phase-locking. | `existing-data-computable` for field L2 if event artifacts and target/zero comparator exist; `requires-new-offline-code` for a dedicated summary/DTW bridge; `requires-new-acquisition` if candidate events/target are absent |
+| **Erase candidate**: mean-field target/error against zero or declared baseline | current DTW if role sensor streams exist; use same window/comparator and normalized/native labels | mean-field L2 is the defensible candidate; use complete-cycle mean only if periodic role exists, otherwise a declared late/mean artifact. Do not use Vortex event L2 semantics | no force claim without calibrated force history | no phase metric unless the erase artifact is genuinely periodic and complete-cycle metadata exists; otherwise mean/late-field semantics | `existing-data-computable` for mean-field L2 when `mean_ux/mean_uy` exists; `requires-new-offline-code` for erase-specific comparator/report; `unsupported` for absent exact erase target/role (legacy README says erase unsupported) |
+| **Steady cloak**: controlled/constant vs uniform inflow or declared baseline | DTW is optional and only meaningful if a steady sensor time series has a defined window; do not fabricate periodic DTW | sensor-plane relative velocity-deficit profile is the primary experimental-compatible candidate; CFD spatial L2 is existing-data-computable where common fields/mask exist. `profile_errors` and `profile_station_errors` provide E∞/L2 profile primitives; `quadrature_l2` provides explicit spatial L2 | current steady artifacts establish torque telemetry work sign, not force ownership; force sensor metric requires new calibrated acquisition | steady is nonperiodic: late snapshot/profile, not phase8 or event metric | profile figure exists in Confirmation navigation but source schema/array is not located: `requires-new-acquisition` or authority deferral for experiment; CFD profile/L2 `existing-data-computable` |
+| **Open-loop dye-only experiment** | unsupported from dye images alone; no signal extraction contract | unsupported quantitatively from dye alone; image-derived L2 needs calibrated PIV/segmentation and registration | unsupported without synchronized calibrated force | qualitative streakline/shedding observation only | `unsupported` for quantitative metric; `implemented` only as bounded qualitative figure/observation if raw images are archived |
+
+### 4.3a Registered geometry, normalization and sensor rationale
+
+For current evaluator geometry, `x_s` and conservative downstream solid extents are V5 Kármán `(1200,1036)`, V5 Illusion `(600,440)`, Legacy Kármán `(800,636)`, and Legacy Illusion `(600,440)`. These are implementation coordinates, not universal physical constants. The fixed rectangle is pre-registered and avoids outcome-dependent wake masks.
+
+The current wake metric is `U0`-normalized area-RMS plus baseline-relative `eta = 1 - E_ctl/E_zero`; target-norm normalization can be misleading because a large ROI contains substantial uniform background flow. Report both absolute error and `eta`, with the ROI, comparator, window and mask status. The canonical result note records 23 rows (15 V5 controlled runs across 11 cases, 7 Legacy periodic cases, 1 Legacy steady case); its result-level observation is that mean-field improvement and phase-resolved improvement can disagree, notably for V5 `kar_re400`, seed 43. Treat those as bounded artifact observations and read the canonical summary directly before quoting numbers.
+
+If a project-specific sensor-placement check is later requested, it can be done without retraining or new CFD by an offline virtual-plane sweep at `x/D = 8,9,10,11,12`, checking dominant-band signal/SNR, centre-versus outer-pair phase coherence, phase slips, DTW ranking stability and sensitivity of phase fields/`E_phase8`. Existing-signal checks can compare centre-only, outer-pair-only and all-six-channel DTW, but cannot independently establish streamwise optimality.
+
+### 4.3 Why metrics are not shared
+
+- `E_mean` is a complete-cycle arithmetic mean, not the mean of eight phase snapshots.
+- `E_phase8` is a discrete phase-resolved diagnostic from independently phased nearest snapshots, not a repeated-crossing phase-conditioned estimator.
+- Vortex event snapshots are transient relative-to-event data; they cannot be renamed as phase8.
+- Erase mean-field error and steady profile error answer different estimands.
+- Native DTW and normalized DTW have different scales and purposes. DTW alignment is not a force response time, mechanism, or causality test.
+- L2 must state its common ROI/mask, normalization, comparator and whether it is RMS or quadrature integral. Do not silently compare V5, Legacy, experiment and steady-theory values.
+
+## 5. Existing code and data paths
+
+### 5.1 Current offline CFD paths
+
+- `src/drl_pinball/eval/wake_l2.py`: schema `drl-pinball-wake-l2-v2`; exact `phase_fields.npz` / `late_field.npz` key checks; normalized ROI velocity RMS; complete-cycle `E_mean`; phase8 slots; crossing/frequency diagnostics; no CFD/GPU import. The note’s current data authority is `src/drl_pinball/data/reproduction/wake_l2_summary.json`; verify/read values from that artifact rather than copying prose summaries.
+- `src/drl_pinball/DATA_DICTIONARY.md`: `data/reproduction/{v5,legacy}/<case>/<role>/`; `phase_fields.npz`, `event_fields.npz`, `late_field.npz`, `timeseries.csv`, `metadata.json`.
+- `src/drl_pinball/legacy_test/acquire.py`: standardized role acquisition and phase/event artifact semantics; 480 warm-up + 160 retained boundaries for standard periodic roles; no-clobber/fail-closed publication.
+- `src/drl_pinball/legacy_test/README.md`: target/controlled/zero/SR role meanings; normalized six-sensor cycle DTW; Legacy `erase` explicitly unsupported.
+- `src/steady_pinball_theory/metrics.py`: `profile_errors`, `profile_station_errors`, `quadrature_l2`, common-mask field norms and partial station/ledger utilities. These are not evidence of experimental sensor data.
+- `src/CCD_analysis/data/karman-dynamic/canonical/` and its external authority parent: current descriptive CCD publication; field arrays and mask are useful reader artifacts, not a force/PIV experiment package.
+
+### 5.2 Schema audit summary
+
+| requested quantity | artifact/schema support | required action |
+|---|---|---|
+| current/new normalized DTW | role `timeseries` and standardized acquisition semantics support it | use existing output where present; otherwise small offline adapter, no CFD |
+| periodic complete-cycle mean + phase8 | `phase_fields.npz` exact schema and `wake_l2.py` | implemented for retained compatible roles |
+| Vortex event-midpoint L2 | `event_fields.npz` is represented in dictionary/acquisition semantics, but current `wake_l2.py` does not evaluate it | new offline code only; no CFD if artifacts exist |
+| Erase mean-field L2 | mean fields supported for periodic artifacts; exact legacy erase target is not supported | existing data only for a valid role; otherwise new acquisition/unsupported |
+| steady profile + L2 | code primitives exist; experimental sensor-plane source not located | CFD existing-data-computable; experiment requires source export/new acquisition |
+| experimental force metric | no synchronized calibrated experimental force bundle located | new acquisition; otherwise unsupported |
+| shared phase metric | deliberately not supported as a scientific contract | keep case-specific semantics |
+
+## 6. Gaps, deferrals and negative results
+
+### 6.1 Missing internal experiment package
+
+No visible package was found containing: raw PIV velocity fields; force-sensor calibration and tare; ADC sample clocks; Raspberry Pi/motor encoder clock; synchronization markers; trial IDs and repeat count; geometry/scale calibration; background-flow blank runs; temperature/thermistor channels; or raw dye image manifest. Without these, do not report experimental `L2`, `DTW`, force coefficient, phase error, shedding frequency in physical units, uncertainty, or efficacy percentage.
+
+### 6.2 Confirmation material
+
+Undermind’s Confirmation navigation exposes Intro/Description/Method/Result/Conclusion and a steady sensor-plane figure candidate. The source array/schema, figure-generation script and provenance were not visible in the inspected repository/Undermind navigation. Treat the figure as a candidate reader view or authority-led deferral until its source is linked. Do not digitize a plotted image to create a false metric.
+
+### 6.3 Literature and facility gaps
+
+- Resolve all Oil-tunnel report cite keys to exact identity/DOI before formal manuscript bibliography.
+- If the experiment is retained, document water-tunnel background blank, settling/wait time, thermal monitoring, carriage speed trace, force calibration and synchronization.
+- Oil-tunnel materials are design precedent/future direction, not evidence that the present system used oil or achieved PIV validation.
+- No new deep search is authorized in this addendum; unresolved literature questions remain Round-2 work.
+
+## 7. Candidate uses in manuscript and supplement
+
+| material | main text candidate | supplementary candidate | do not use |
+|---|---|---|---|
+| apparatus sketch + short build description | one bounded paragraph or compact methods inset: platform assembled and policy deployment feasible | component diagram, wiring/data-flow schematic, calibration checklist | precision, closed-loop claim, force-feedback success |
+| dye steady/illusion panels | one qualitative feasibility panel only if raw image provenance is available | full image sequence, timing/condition caption, negative/limitation note | quantitative wake/force/field restoration |
+| water-background explanation | one sentence explaining low-speed experimental limitation, citing resolved literature | facility note with thermal/residual-motion rationale and proposed controls | claim that present motion is proven thermal or that waiting removes it |
+| CFD case metrics | main text only for current bounded Kármán/cloaking claims with exact case/window/comparator | complete case matrix, normalized DTW, phase8/event definitions, per-seed rows | pooled V5/Legacy or DTW-as-delay |
+| steady sensor-plane profile | main text only if source artifact, calibration and baseline are linked | full profile data/metadata and L2 derivation | digitized Confirmation pixels without authority |
+| force metric | main text only after synchronized calibrated acquisition | raw force/ADC/encoder package and calibration uncertainty | reward as force, uncalibrated ADC, or force inferred from dye |
+
+## 8. Figure/data needs
+
+### Minimum to keep experiment in Round-1 supplement
+
+1. Apparatus schematic: test-section isolation, rail/rack, carriage, sensor/ADC/Raspberry Pi, motor drive and cylinder geometry.
+2. Conditions table: fluid, cylinder `D`, spacing, nominal speed, estimated/actual `Re_D`, carriage trajectory, trial ID and rest time.
+3. Raw dye images/video with frame rate, spatial scale, camera view, dye injection timing and blank/background run.
+4. Explicit caption: open-loop qualitative demonstration; no synchronized PIV/force validation.
+
+### Minimum for quantitative experimental metric
+
+- PIV: calibrated `(x,y,t)` velocity arrays, common-fluid mask, `U_inf`, registration and missing-data policy; then profile/L2 can be computed offline.
+- Sensors: channel names/order, units, calibration/tare, sample rate, anti-aliasing, timestamps and synchronization marker; then normalized DTW can be recomputed offline.
+- Force: calibrated body-resolved `F_x,F_y` (and torque only if independently measured), zero/drift records, sign convention, span/normalization and synchronized video/PIV; only then define force L2/DTW or force-vs-baseline metrics.
+- Thermal/background: blank-run velocity or dye displacement, temperature probes near surface/walls/bottom, pump/carriage state and settling interval; use to bound—not automatically subtract—the background.
+- Repeats: independent trials and declared acceptance/failure rules. One dye sequence is descriptive, not uncertainty evidence.
+
+### Figure candidates by status
+
+| figure/data item | status | source/next step |
+|---|---|---|
+| apparatus diagram | requires-new-offline-drawing from author dimensions, or author artifact | no CFD |
+| dye steady and illusion panels | existing qualitative candidate if raw media exists; otherwise requires-new-acquisition | preserve as bounded observation |
+| Confirmation steady sensor-plane profile | existing figure candidate, source not audited | locate source array/manifest before use |
+| CFD mean/phase8 L2 panel | implemented/existing-data-computable for compatible roles | derive reader plot from immutable outputs |
+| Vortex event L2 panel | requires-new-offline-code if event NPZ exists | retain event labels/offsets |
+| Erase mean L2 panel | existing-data-computable only for supported role/target | legacy erase remains unsupported |
+| normalized/current DTW panel | existing-data-computable or small offline adapter | label native vs normalized and window |
+| synchronized force/PIV panel | requires-new-acquisition | do not imply current availability |
+
+## 9. No-new-CFD completion list
+
+The following evaluation work can be completed without new CFD **provided the named artifacts already exist and pass their metadata/schema contracts**:
+
+- Recompute or tabulate current/normalized sensor DTW from retained timeseries and target channels.
+- Recompute periodic Kármán/Illusion complete-cycle mean-field `L2*` and existing `E_phase8` from `phase_fields.npz`, using current `wake_l2.py` semantics.
+- Produce per-event Vortex downstream-ROI L2 from existing `event_fields.npz` after writing a small offline adapter; this is not a new acquisition if target/control/zero event comparators exist.
+- Produce Erase mean-field L2 for a genuinely supported mean-field role; do not revive the explicitly unsupported Legacy erase route.
+- Produce steady CFD sensor-plane/profile errors and spatial L2 from existing resolved arrays and masks using `steady_pinball_theory.metrics`.
+- Replot or audit current CCD field/mask reader artifacts descriptively; this does not produce experimental validation.
+
+The following cannot be completed from current visible evidence without new experimental acquisition or an authoritative source export: experimental PIV L2, experimental normalized DTW if synchronized sensor streams are absent, calibrated force metrics, uncertainty/repeatability, and quantitative dye-image velocity/field error. None of these requires new CFD by itself; they require experimental data and offline processing/validation.
+
+## 10. Round-1 handoff
+
+- **Safe experimental sentence**: the apparatus demonstrates policy deployment and qualitative open-loop wake modification under difficult low-speed conditions; quantitative synchronized force/PIV validation remains future work.
+- **Safe metric sentence**: field and sensor comparisons should be reported with case-specific normalized L2/DTW contracts; periodic phase8, transient event snapshots, mean-field erase, and steady profiles are distinct estimands.
+- **Do not write**: “experimentally validated cloak/illusion”, “force feedback achieved”, “PIV confirms the CFD”, “DTW gives physical delay”, “one metric covers all scenes”, or “thermal convection is proven as the sole cause”.
+- **Next authority needed**: link the local/source Confirmation sensor-plane arrays and use the already read-full `docs/JFM_WYQ/WAKE_L2_AND_SENSOR_PLACEMENT_RESEARCH_NOTES.md` authority together with the source-linked Confirmation sensor-plane arrays; resolve Oil-tunnel cite keys before manuscript prose.

docs/JFM_WYQ/ROUND1_BOTTOM_UP/README.md

@@ -0,0 +1,18 @@
+# ROUND1_BOTTOM_UP
+
+这是 JFM 工作流第一轮的非文章式证据索引，不是 manuscript prose，也不预填科学结论。目标是建立全面、可追溯、便于后续 agent 快速读取的索引与 evidence dossiers。
+
+## 预定文件
+
+- `01_DRL_CONTROL_AND_CASES.md`
+- `02_SR_CONTROL_LAWS.md`
+- `03_FIELD_CCD_OID.md`
+- `04_STEADY_THEORY_AND_CROSSMAP.md`
+- `05_INTRODUCTION_LITERATURE_INDEX.md`
+
+## 证据与协作约束
+
+- 所有数字都必须回到对应的 authority、code、manifest 或 artifact。
+- Legacy 与 V5 必须分链记录，不得静默合并证据链。
+- 每项记录应保留来源、边界、冲突/缺口及必要的 case/seed、metric、comparator 和 evaluation window。
+- 最终由顶层 agent 跨文件综合；本轮文件只负责索引证据，不提前冻结叙事。

docs/JFM_WYQ/ROUND2_TOP_DOWN/01_PHYSICS_NARRATIVE_AND_OUTLINES.md

@@ -0,0 +1,160 @@
+# JFM Round 2 — Physics Narrative and Outline Recommendation
+
+> Top-down structural design, not manuscript prose. Round-1 files `01_DRL_CONTROL_AND_CASES.md`–`08_EXPERIMENT_AND_CASE_METRICS.md` were read in full (current filenames), together with `src/README.md`, DRL/SR/CCD/OID/steady authority documents and the CelerisLab validation dossier. Nowledge Mem Context/Working Memory and targeted recall informed routing and author/adviser direction, not scientific authority. Round-1 audit records 63 current Markdown plus 8 selected authority/index JSON files read-full; raw/external artifacts remain authority-led deferrals.
+
+## 1. Scientific question, central claim and reader takeaway
+
+**Scientific question.** In a moderate-Re fluidic pinball with separation and periodic wake dynamics, can sparse downstream signatures and force measurements support a closed-loop controller whose apparent complexity reduces to a persistent rotation skeleton plus a smaller dynamic correction, and can field/steady analyses bound that interpretation?
+
+**Central claim (recommended, cautious).** For the named Legacy Kármán cloaking contract, DRL produces bounded observer-level and field-level wake modification. Independent SR, corrected CCD and steady/context analyses consistently identify a persistent rear counter-rotation component as the dominant tested backbone, with a smaller phase-coherent, action-associated residual. Together they motivate—but do not establish—a near-constant-rotation skeleton plus dynamic feedback correction as a compact organization of the learned control.
+
+Use “named,” “tested,” “bounded,” “dominant tested element,” “descriptive,” and “motivate.” Do not use mechanism, causality, optimality, stability, universality or Legacy–V5 equivalence language. CCD residual is a reference residual, not a paired counterfactual; steady/theory is context/outer reference; potential circulation cancellation failed by sign.
+
+**Reader takeaway.** A controller that improves a wake signature need not be intrinsically complicated: in this bounded case, most measured improvement is associated with persistent rear rotation, while the remaining correction is smaller and phase-organized. The contribution is evidence triangulation with explicit limits, not a universal law.
+
+## 2. Candidate title families
+
+### A — learned control reduced to a physical skeleton *(recommended)*
+1. *A near-constant rotation skeleton in learned cloaking control of a fluidic pinball*
+2. *From sparse wake signatures to a rotation skeleton for fluidic-pinball cloaking*
+3. *Interpreting learned wake cloaking as persistent rotation with dynamic correction*
+
+Promise: a physical object emerging from control, SR, field and steady evidence. Risk: “skeleton” may overstate definitiveness; retain “near-constant” and bound it to tested cases. Title 1 is safest; title 3 most clearly promises the correction but risks causal expectations.
+
+### B — observer-level cloaking and signature control
+1. *Sparse-signature control of a moderate-Reynolds-number fluidic pinball wake*
+2. *Active wake cloaking with sparse sensing: control, reduction and field evidence*
+3. *Controlling the observer-level wake signature of a fluidic pinball*
+
+Promise: the estimand is the wake signature/field, not unqualified stealth. Risk: under-promises the skeleton interpretation and may resemble a standard DRL paper; define cloaking as a declared signature/reference objective.
+
+### C — triangulated white-box analysis
+1. *A physics-based audit of learned wake cloaking in the fluidic pinball*
+2. *Symbolic and field-level interpretation of sparse-sensor wake control*
+3. *Triangulating learned wake control with symbolic regression and field analysis*
+
+Promise: policy reduction plus field audit. Risk: “audit,” “physics-based,” and “interpretation” invite stronger mechanism expectations than CCD/OID support. Use only if methodological contribution must lead.
+
+**Recommendation:** work with A1; use B1 if “skeleton” cannot survive final evidence audit. Avoid “universal,” “robust,” “mechanism,” and standalone “interpretable” in the title.
+
+## 3. Recommended hierarchy and section bridges
+
+### 1. Introduction — visibility to nonlinear wake signatures
+- **Question:** why nonlinear bluff-body signature cloaking differs from linear-wave cloaking and ordinary drag reduction.
+- **Input:** cloak/illusion, active hydrodynamic, pinball, sparse-observation DRL and interpretable-control literature.
+- **Output:** define observer-level/field objective; motivate pinball; state gap as policy reduction plus field audit.
+- **Bridge:** fix plant, sensors, actions and evidence contracts.
+
+### 2. Problem formulation and evidence contracts
+**2.1 Plant and actuation:** geometry, body order, coordinates, Re convention, wall/action sign, `q_in/q_blk/q_ctl`, and separate Legacy/V5 chains.
+**2.2 Sparse objective:** forces plus six downstream velocity channels; reward terms and DTW; full-field L2 is independent validation; DTW is not delay/causality.
+**2.3 Case-specific metrics:** periodic cycle, transient event and steady profile/`E_inf_vector` semantics remain distinct.
+- **Output:** one plant vocabulary, non-interchangeable estimands.
+- **Bridge:** show capability briefly, then select Re100 for depth.
+
+### 3. Compact capability map and principal case
+**3.1 Visual hook:** aligned Kármán Re100 plus qualified representative vortex/steady/Illusion view, with role/status labels.
+**3.2 Case choice:** Legacy Kármán Re100 is the intersection of periodic signature, field, SR and CCD evidence; V5 Re100 seeds support modern discovery/retention separately.
+**3.3 Breadth:** compact named-condition map; complete matrix and degradation details in supplement; no positive Illusion-SR claim.
+- **Output:** bounded capability, Re100 anchor.
+- **Bridge:** define what success means before interpreting action.
+
+### 4. What was optimized: signature matching under sparse observation
+**4.1** Reward and sensor design as a physical measurement choice. **4.2** Compare reward/DTW/force with independent mean/phase field errors. **4.3** Give only bounded DRL discovery/retention evidence (V5 seed dependence explicit).
+- **Output:** compressed sensor objective requires white-box/field audit.
+- **Bridge:** ask whether the action law has a persistent component.
+
+### 5. Physical question I — is the learned action intrinsically complex?
+**5.1** SR as closed-loop distillation: Stage 1–2–3, causal pairing, symmetry mapping, CFD arbiter. **5.2** Canonical Kármán law: rear constant, rear-lift feedback, tested front term, exact sign/order. **5.3** Closed-loop tests, named transfer and steady correspondence; deletion is parent-relative, not causal necessity.
+- **Output:** rear counter-rotation is the dominant tested SR element; feedback remains nonzero/case-dependent.
+- **Bridge:** what field improvement corresponds to the constant component?
+
+### 6. Physical question II — what changes after removing the constant reference?
+**6.1** Corrected mean field: zero, constant-mean, DRL and exact 93.42/6.58 additive split for the stated Legacy Kármán ROI/mask/window. **6.2** Define separately centered DRL-minus-constant-template residual; show rank-3 stability and phase reconstruction. **6.3** CCD/POD: action-coordinate association, nearly same field subspace; no CCD superiority/unique modes.
+- **Output:** most measured field improvement is associated with constant mean; residual is smaller and descriptively organized.
+- **Bridge:** place recurrence of the constant component in steady context.
+
+### 7. Physical question III — steady context and theory boundary
+**7.1** Canonical free-slip `Re_D=50`, accepted `[0,+Omega,-Omega]`, `s`, direct `E_inf_vector` at `x/D=10`. **7.2** One-cylinder impermeability/circulation and outer potential/dipole reference; show sign failure. **7.3** Gap flux/near-wake/partial NS diagnostics as bounded observations; gap/base-bleed chain remains hypothesis.
+- **Output:** plausible organizing context, not mechanism/stability/causality/no-slip transfer.
+- **Bridge:** state supported claim and decisive future tests.
+
+### 8. Synthesis, limitations and prospective experiment
+**8.1** SR ranking, CCD decomposition and steady endpoint are compatible but non-identical observations. **8.2** Briefly quarantine negative Illusion SR, claim-free OID, old CCD/steady routes and failed theory. **8.3** One compact qualitative open-loop experiment note/panel only; no synchronized PIV/force validation.
+- **Output:** bounded, falsifiable physical organization.
+
+### 9. Methods/reproducibility
+Keep only interpretation-protecting details in the main text; put solver/PPO/SR/CCD implementation, manifests, hashes, normalization, seeds, phase gates and full metric contracts in supplement. OID remains method-only, not a results section.
+
+## 4. Two materially different alternatives (not recommended)
+
+**A — method silos:** Introduction; DRL method/results; SR; CCD; OID; steady theory; experiment; discussion. Easy package ownership, but delays the physics, makes compatible evidence look independent, repeats definitions and gives claim-free OID unjustified weight.
+
+**B — broad capability-first catalogue:** Introduction; full matrix; reward; all Kármán/vortex/erase/Illusion results; SR generalization; fields; theory; experiment. Visually attractive, but forces incomparable metrics into one progression, dilutes Re100, and risks reviving withdrawn Illusion/generalization claims.
+
+A steady/theory-first outline is also not recommended: endpoint evidence is bounded, potential theory failed by sign, and NS closure is incomplete, so it would overpromise the theory.
+
+## 5. Space, equations, tables and figures
+
+### Text (100% of body)
+| Part | % | Function |
+|---|---:|---|
+| Introduction | 10 | vocabulary, gap, question |
+| Problem/objectives/contracts | 10 | plant, sensors, reward, metrics |
+| Capability + Re100 setup | 10 | visual hook, case choice |
+| SR question | 18 | law and closed-loop evidence |
+| Mean/dynamic field + CCD | 20 | decomposition and residual |
+| Steady/theory | 17 | endpoint, reference, sign boundary |
+| Synthesis/limits/experiment | 10 | bounded conclusion and future tests |
+| Methods bridge | 5 | essential reproducibility |
+
+Target **8–10 displayed equations** (5–6 main): objective; reward/DTW; representative field metric; SR law and symmetry; mean identity; CCD residual/operator; steady `s`/`E_inf_vector`; outer reference/clearly incomplete ledger. Target **4 main tables**: contracts with Legacy/V5 separated; Re100 evidence chain; SR term ranking/non-claims; cross-analysis synthesis. Supplement: full matrix, seed/calibration/normalizer, provenance, metrics, sensitivity and artifact ledgers.
+
+Target **7 main figures**:
+1. visual capability hook;
+2. geometry/action sign/sensors/roles;
+3. objective versus independent field validation;
+4. SR law, actions and bounded deletion;
+5. zero–constant–DRL mean fields and 93.42/6.58 split;
+6. corrected CCD residual, rank-3 and POD comparison;
+7. steady endpoint, profiles, theory reference and sign failure.
+
+Experiment belongs as an inset in 1 or 7 only if raw-image provenance exists; otherwise supplement. Target 4–6 supplement figures.
+
+## 6. Main text / supplement boundary
+
+**Main:** scientific question; cloak/signature definition; sparse objective; one visual hook; Legacy Re100; canonical SR law; exact meaning of dominant tested constant; corrected mean/dynamic result; CCD/POD descriptive result; steady endpoint and failed potential route; one experiment limitation; final non-claims.
+
+**Supplement:** full case/seed matrix; all V5 curves; separate Legacy/V5 solver and contract table; reward/DTW implementation; SR topology/deletion/per-case tables; full CCD sensitivities; OID equations/reset history without positive result; MFS convergence; partial NS ledger; steady sweep/reverse details; transient/Erase metrics; experiment apparatus/calibration/dye sequence; paths/hashes/exclusions/negative routes.
+
+## 7. Concrete SR/CCD/theory interweaving
+
+Use **physical-question interweaving**, not three method chapters:
+1. How much is persistent? — SR law and closed-loop term ranking.
+2. What does it account for? — constant-mean versus DRL field errors.
+3. What remains? — corrected CCD residual, rank-3 stability and action coordinates.
+4. Is there steady context? — endpoint, magnitude correspondence, theory as outer reference and sign check.
+5. Synthesis — skeleton-plus-correction as compatible organization, with non-crossable estimands stated.
+
+At each interface state the estimand and chain. Never imply SR law, CCD left action functions, and steady `E_inf_vector` are the same object.
+
+## 8. Honest Legacy/V5 presentation
+
+- **Legacy:** principal analysis chain for SR, corrected CCD, standardized Target/PPO/SR/Zero and field interpretation.
+- **V5:** supporting modern chain for solver qualification context, Re100 five-seed discovery/retention, parameter/seed examples and named-condition demonstrations.
+- Use explicit panel/caption labels `Legacy analysis` and `V5 support`; never pool means/errors.
+- Explain Legacy→V5 as a deliberate rewrite changing grid/API/readback, body/action interfaces, normalization/calibration and smoothing; this is provenance discipline, not numerical continuation.
+- State V5 Re100 late retention is seed-dependent; other points are mostly n=1 and retained evaluations are historical compatibility data, not post-fix retraining.
+- Do not use V5 to rescue Legacy SR, negative Illusion SR, CCD, or steady mechanism claims.
+
+## 9. Structure-freeze decisions (recommendation first; max 5)
+
+1. Adopt working title A1; retain “near-constant,” “tested,” and bounded language. If “skeleton” is rejected, use B1 and keep it as a section-level interpretation.
+2. Freeze Legacy Kármán Re100 as the sole deep case; V5 is separate support; full matrix is supplement.
+3. Freeze physical-question order SR → mean/dynamic field → CCD → steady/theory; no standalone OID results section.
+4. Freeze experiment as qualitative open-loop feasibility/future direction only, compact or supplement.
+5. Freeze strongest mechanism wording at “consistent with/motivates a candidate organization”; no causal, stable, optimal, general or universal wording without matched intervention, closure and independent evidence.
+
+## Evidence dependency gate
+
+Before prose or figure assembly, every number/panel must carry chain, solver/configuration, geometry/body order, signed action, Re definition, case/seed/role, target/reference, metric/ROI/mask, window, normalization/calibration, artifact path/hash and evidence status. Resolve CCD canonical designation while preserving the validated suffixed artifact; quarantine OID sign conflict and historical results; preserve SR semantic current/archive split; verify external role manifests; retain steady sign failure/incomplete NS closure and absent synchronized PIV/force validation. No figure is publication-ready merely because a plot or manifest exists.

docs/JFM_WYQ/ROUND2_TOP_DOWN/02_JFM_STRUCTURE_AND_SPACE_BENCHMARK.md

@@ -0,0 +1,137 @@
+# JFM Round 2 — Structure, Space and Figure Benchmark
+
+> Structural calibration only; not manuscript prose or submission-format advice. Based on six Round-1 dossiers, existing DynamisLab Undermind reports, and ten existing full-PDF reads. No new deep search, CFD, training, experiment, Undermind write, or commit was performed.
+
+## Executive recommendation
+
+Use a physics-led narrative:
+
+`visual capability hook → precise gap → plant/reward/metric contracts → representative capability → SR control skeleton → mean/dynamic field decomposition → steady/theory context → limitations and bounded synthesis`.
+
+Do not make DRL, SR, CCD and theory four isolated method papers. Interleave them around (i) what persistent control component accounts for the tested improvement and (ii) what smaller dynamic correction remains. Kármán Re100 is the deep-analysis case. Other cases are named-condition capability/failure evidence, not universal generalization.
+
+## Sample literature and read level
+
+All entries were selected from the existing DynamisLab workspace and read with Undermind `read_pdfs`; `full-PDF` means paper-level full-text extraction, not independent raw-data or equation-by-equation auditing. Counts below are approximate where extraction was not reliable.
+
+| Key / identity | DOI | Read level | Structural lesson |
+|---|---|---|---|
+| Deng et al., “Low-order model for successive bifurcations of the fluidic pinball”, *JFM* 884 A37 (2020) [Den18b] | `10.1017/jfm.2019.959` | `full-PDF` | Physics-first gap; rigorous model then verification; master phenomenogram; extensive appendices. |
+| Cornejo Maceda et al., “Stabilization of the fluidic pinball with gradient-enriched machine learning control”, *JFM* 917 A42 (2021) [Mac20] | `10.1017/jfm.2021.301` | `full-PDF` | Baseline → simple forcing → asymmetric forcing → feedback; capability paired with mechanism. |
+| Rabault et al., “Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control”, *JFM* 865, 281–302 (2019) [Rab18b] | `10.1017/jfm.2019.62` | `full-PDF` | Very short entry to gap; learning/capability followed by physical wake explanation. |
+| Paris et al., “Robust flow control and optimal sensor placement using deep reinforcement learning”, *JFM* 913 A25 (2021) [Par20] | `10.1017/jfm.2020.1170` | `full-PDF` | Performance → mechanism → robustness → sensor placement; strong claims matched by tests. |
+| Li & Zhang, “Reinforcement-learning-based control of confined cylinder wakes with stability analyses”, *JFM* 932 A44 (2022) [Li21] | `10.1017/jfm.2021.1045` | `full-PDF` | Stability/sensitivity informs reward and probes; it is not an ornamental final section. |
+| Xia et al., “Active flow control for bluff body drag reduction using reinforcement learning with partial measurements”, *JFM* 981 A17 (2024) [Xia23] | `10.1017/jfm.2024.69` | `full-PDF` | Partial-observation failure → history-based fix → robustness and mechanism. |
+| Wang et al., “Dynamic feature-based deep reinforcement learning for flow control of circular cylinder with sparse surface pressure sensing”, *JFM* 988 A4 (2024) [Wan23b] | `10.1017/jfm.2024.333` | `full-PDF` | Information problem → dynamic lifting → sparse capability → harder-Re tests. |
+| Loiseau et al., “Sparse reduced-order modelling: sensor-based dynamics to full-state estimation”, *JFM* 844, 459–490 (2018) [Loi17] | `10.1017/jfm.2018.147` | `full-PDF` | Black-box sensor dynamics precede field reconstruction and grey-box interpretation. |
+| Gautier et al., “Closed-loop separation control using machine learning”, *JFM* 770, 442–457 (2015) [Gau14] | `10.1017/jfm.2015.95` | `full-PDF` | Concise setup → optimization → explicit law → spectral mechanism → benchmark. |
+| Mao et al., “Nonlinear optimal suppression of vortex shedding from a circular cylinder”, *JFM* 775, 241–265 (2015) [Mao15] | `10.1017/jfm.2015.304` | `full-PDF` | Validation precedes interpretation; spatial mechanism and stability follow performance; realizability caveat explicit. |
+
+## Space and opening benchmark
+
+Closest concise control papers reach the gap in about **1–3 pages / 4–7 functional paragraphs**. Long physics/model papers may use about five paragraphs but more pages ([Den18b]). Use five Introduction paragraphs and roughly **2–3 pages**:
+
+1. wake/signature matching as a physical question;
+2. bluff-body/pinball complexity and why scalar force success is insufficient;
+3. DRL and sparse/partial observations as capability, not explanation;
+4. SR and field decomposition as complementary audits with noncausal boundaries;
+5. exact gap and contributions for this named plant.
+
+Do not let linear cloaking history delay the gap; it supplies vocabulary, not nonlinear-Navier–Stokes equivalence.
+
+### Total planning range
+
+Use **8,000–10,000 word-equivalents**, or roughly **22–28 compact LaTeX pages** including figures/equations but excluding references and detailed supplement. This is coarse planning, not a journal limit.
+
+| Function | Share | Approx. words | Approx. pages |
+|---|---:|---:|---:|
+| Introduction | 12–15% | 1,000–1,400 | 2–3.5 |
+| Plant, contracts, metrics, CFD validation | 16–20% | 1,300–1,900 | 4–5.5 |
+| Capability/case map | 10–14% | 800–1,300 | 2.5–4 |
+| SR law and closed-loop validation | 13–17% | 1,100–1,600 | 3–4.5 |
+| CCD descriptive decomposition | 10–14% | 800–1,300 | 2.5–4 |
+| Steady/theory context | 10–14% | 800–1,300 | 2.5–4 |
+| Integrated discussion, limitations, experiment, conclusion | 12–16% | 1,000–1,500 | 3–4 |
+
+A suitable balance is **Methods 25–30% / Results 45–50% / Discussion and limitations 20–25%**. Discussion should be woven into each result subsection, following [Mac20], [Gau14], [Xia23], [Mao15].
+
+## Main figures, tables and equations
+
+Plan **7–9 main figures**, about **20–28 panels**, and **3–4 tables**.
+
+1. Headline Kármán Re100 target/zero/controlled field and signature plate.
+2. Plant/evidence contracts: geometry, action sign, sensors, `q_in/q_blk/q_ctl`, Legacy/V5 boundary.
+3. Compact named-condition capability matrix; separate DTW, field-L2, force roles and n/status.
+4. Re100 time series plus phase-registered field validation and selected-seed provenance.
+5. SR law, symmetry deployment, closed-loop PPO/SR/zero comparison and bounded deletion result.
+6. Mean/dynamic field decomposition, ROI/mask, CCD rank-3 descriptive modes and POD comparison.
+7. Integrated SR–CCD–steady physical map showing compatible observations versus unresolved arrows.
+8. Steady/theory endpoint, reverse/wrong-sign boundary, outer reference and incomplete NS ledger.
+9. Optional bounded open-loop experiment demo; omit or move to supplement if synchronized force/PIV evidence is unavailable.
+
+Tables: (1) evidence/contract ledger; (2) capability summary; (3) SR/CCD/theory estimands and claim boundaries; optional (4) validation/limitations ledger. Full 11-case/15-run matrices, seeds and provenance belong in supplement.
+
+Plan **12–18 numbered equations**: geometry/Re/action (2–3), reward and independent metrics (2–3), SR law/symmetry (2–3), mean/CCD estimands (2–3), steady metric/outer reference (2–4), explicit balance/claim ledger (1–2). Put PPO, DTW dynamic programming, feature libraries, CCD matrices, MFS collocation and long solver details in supplement.
+
+## Capability without a gallery paper
+
+Capability should be **10–14% of main text**, one early hook plus one compact table/summary. Use a three-layer funnel: one Re100 hook; breadth matrix with named degradation/failure; return to Re100 for SR/CCD/theory. Every capability claim needs a same-condition comparator and one explanatory diagnostic. Do not show attractive fields without metric/window/role identity. Illusion SR is historical/negative, not positive capability evidence. Vortex/Erase and complete seed galleries belong in supplement unless they answer a distinct physical question.
+
+## Interweaving and space bounds
+
+Organize results by physical questions:
+
+1. DRL achieves a declared signature/field improvement at named scope.
+2. SR exposes a compact persistent control component and validates it in Legacy closed loop.
+3. CCD separates a dominant mean increment from a smaller action-associated dynamic increment.
+4. CCD modes organize descriptive phase-coherent variation; this is not causality or stability.
+5. Steady/theory contextualizes constant rotation under its own V5 free-slip contract; failed circulation sign and incomplete NS ledger define the mechanism boundary.
+6. Synthesis: persistent rear counter-rotation/base-bleed-or-gap-jet-family **hypothesis** plus smaller dynamic correction, not a proven momentum law.
+
+| Component | Minimum main text | Maximum recommended | Role |
+|---|---:|---:|---|
+| SR | 1.5–2 pages | 3.5–4 | Law, deployment, closed-loop and deletion boundary. |
+| CCD | 1.5 pages | 3–3.5 | Mean/dynamic estimands and descriptive rank-3 result. |
+| OID | 0.25–0.75 page | 1 | Method-only/claim-free; historical positive rows quarantined. |
+| Steady/theory | 1.5–2 pages | 3.5–4 | Contract, endpoint, short outer reference, open gates. |
+| DRL capability | 2–3 pages | 4 | Hook plus Re100 return; no gallery. |
+
+SR is an `observation → action` surrogate under Legacy CFD arbitration; CCD is action-associated field organization; theory is contextual/reference mathematics. Do not claim that any one explains the same physical arrow as another.
+
+## CFD, experiment, negatives, Legacy/V5
+
+**CFD validation:** 1.5–2.5 pages plus one panel/table. Establish active V5 and separate Legacy plant contracts, grid/BC/role checks and qualification limits. Solver validation does not validate DRL mechanism.
+
+**Experiment:** current status is open-loop feasibility (streaklines/dye; no complete synchronized force/PIV package). Use 0.5–1 page at the end or supplement, explicitly as demonstration/future validation.
+
+**Negative results:** keep visible but compact: cross-Re degradation/action saturation and weak Re400 phase; Illusion SR collapse toward physical zero; wrong-sign steady branch and failed potential-circulation explanation; OID force-gate failure/reset. Full failed logs, rejected formulas, incomplete crossings and archive provenance go to supplement.
+
+**Legacy/V5:** state prominently in Methods and at every transition. LegacyCelerisLab supports active SR and corrected CCD; modern V5 supports current DRL/steady retained chains. They are not numerically identical. Use explicit labels such as `Legacy Kármán Re_code=100 (Re_D=50)` and V5 code-Re labels; never pool them silently.
+
+## Supplement boundary
+
+Carry hyperparameters, normalization/seed ledgers, complete cases and metrics, SR feature/split/hash/deletion tables, CCD masks and sensitivity, OID derivations and historical audit, MFS convergence/one-cylinder/Laurent checks, experiment calibration/raw inventory/uncertainty, and all negative/archive provenance. Do not move an essential comparator or essential limitation out of the main text.
+
+## Title, abstract and prohibited strength
+
+Possible bounded title register:
+
+- **Interpreting reinforcement-learning flow control in a fluidic pinball through sparse laws and wake-field decomposition**
+- **A sparse control skeleton for wake-signature matching in a fluidic pinball**
+- **From reinforcement-learning control to a bounded physical decomposition of a fluidic-pinball wake**
+
+Safe terms, with scope qualifiers: `wake-signature matching`, `sparse sensing`, `symbolic-regression surrogate`, `descriptive field decomposition`, `steady reference`, `physics-guided interpretation`. Avoid unqualified `universal`, `optimal`, `robust`, `generalizable`, `causal`, `mechanism`, `stability`, `control authority`, `full-state`, `complete`, `first`, `real-time`, `experimentally validated`, `momentum restoration`, `energy-efficient`, `illusion control` and `target tracking`. Avoid unqualified `hydrodynamic cloaking` if the estimand is only a declared sensor/field space. `Interpretable` should describe the compact SR law, not imply CCD/OID causality.
+
+The abstract should promise only: named-plant DRL discovery; independent sparse-signature and field evaluation; compact SR law under the Legacy contract; descriptive mean/dynamic decomposition; bounded steady/theory context. Include one limitation sentence: named conditions, separate Legacy/V5 chains, no causal/mechanistic closure, limited experiment.
+
+## Warnings for the structure agent
+
+**Over-expand:** cloak history; all cases/seeds/plots; four method silos; OID historical rows or Illusion SR as positive evidence; unvalidated MFS derivation; applications involving seals, marine structures, sediment, defense or low observability.
+
+**Over-contract:** field pair plus reward without comparators; CFD/Legacy–V5/Re conventions; SR without closed-loop arbitration and deletion window; CCD “93%” without ROI/mask/estimand; steady appearance presented as stability/mechanism; negative results and experiment limitations hidden in supplement.
+
+The target is fewer claims, each attached to a comparator, artifact/metric contract, named evidence status and one physical question.
+
+## Provenance
+
+Nowledge Mem `read_context_bundle` was read first and targeted memory recall used. Undermind `get_orientation` preceded `list_workspaces`; existing DynamisLab workspace `4383d145-2c80-4991-8a9c-7ac2dd683438` was used. No new deep search or Undermind mutation occurred. Round-1 `01`–`06` dossiers and Round-2 README were read. All ten samples above were requested through `read_pdfs` and are marked `full-PDF`; uncertain counts are explicitly approximate.

docs/JFM_WYQ/ROUND2_TOP_DOWN/03_EVIDENCE_STRESS_TEST.md

@@ -0,0 +1,183 @@
+# JFM Round 2 — Evidence-to-Structure Stress Test
+
+## Executive verdict
+
+The proposed arc is publishable only as a **bounded Kármán/cloaking case study**: DRL demonstrates named-condition policy discovery, Legacy SR supplies a closed-loop-tested compact surrogate, corrected CCD describes a smaller action-associated residual around a dominant constant component, and steady/theory supply contextual physical constraints. The stronger reading—“DRL discovers a general physical law, CCD proves the dynamic correction, and theory closes the mechanism”—is not supported. **Recommendation: submit the minimal defensible paper; keep the ambitious layer in Discussion/Supplement.**
+
+## 1. Claim–evidence dependency graph
+
+```text
+C0 Visual capability / observer-level objective
+ ├─ DRL: scene, target/zero comparator, observation/reward/action contracts,
+ │       seed/checkpoint/VecNormalize, window
+ ├─ literature: sparse signature motivation only
+ └─ field validation: E_mean and separately E_phase8
+        ↓
+C1 Reward/signature is physically motivated
+ ├─ exact reward, DTW implementation/window/calibration identity
+ ├─ sensor geometry and x/D≈10 precedent
+ ├─ DTW separated from full-field L2
+ └─ no optimality/observability/delay claim
+        ↓
+C2 Kármán policy has a compact control law
+ ├─ frozen Legacy PPO trajectories and causal alignment
+ ├─ SR topology/formula/provenance
+ ├─ offline fit diagnostics (supporting only)
+ └─ Stage-3 Legacy closed-loop gates, long runs, named extensions
+        ↓
+C3 Law contains a near-constant skeleton
+ ├─ fixed-formula deletion/scaling under declared 40-step Kármán window
+ ├─ steady constant-rotation sweep under exact sign/Re/action contract
+ └─ named cross-case observations, not a universal law
+        ↓
+C4 Residual is smaller and dynamically organized
+ ├─ corrected CCD mean-field decomposition, exact ROI/mask/window
+ ├─ cycle-template residual and rank-3 stability
+ ├─ action-coordinate association and phase diagnostics
+ └─ constant and DRL cycles are not paired counterfactuals
+        ↓
+C5 Constant + dynamic correction explains finite-Re mechanism
+ ├─ matched intervention or independent closed-loop designs
+ ├─ wall-resolved vorticity/pressure/shear and complete NS CV budget
+ ├─ action sign/torque/body-order parity
+ └─ held-out endpoints and uncertainty
+        ↓
+C6 Generality / limits
+ ├─ repeated seeds and retained cross-Re/size cases with parity
+ ├─ case-specific metrics and per-case failures
+ ├─ Legacy/V5 separation or explicit parity artifact
+ └─ experiment only qualitative feasibility without synced PIV/force
+```
+
+C0→C1 is green/amber; C1→C2 is amber-green only for Legacy Kármán; C2→C3 is amber; C3→C4 is amber for descriptive organization; C4→C5 is red; C5→C6 is red if phrased as generality, but green for an explicit limits section.
+
+## 2. Link-by-link status and strongest defensible wording
+
+| Link | Status | Strongest surviving wording |
+|---|---|---|
+| Visual capability → task success | **Green/amber** | “Under the declared cases and contracts, PPO policies reduced sensor/field error relative to the case-specific zero comparator.” Separate mean and phase claims and include failures. |
+| Capability → cloaking vocabulary | **Amber** | “Cloaking denotes approach to a declared background/target in measured wake space.” Not invisibility or full-state equivalence. |
+| Capability → reward/signature | **Green** | “The signature objective is computationally economical and physically motivated, with independent registered downstream field evaluation.” |
+| Reward → full-field cloaking | **Amber** | “Sensor optimization was accompanied by, but does not imply, improved mean-field agreement.” Report `E_mean`, `eta_mean`, `E_phase8`, mismatch and regularity separately. |
+| DRL → SR law | **Green/amber** | “A symmetry-mapped SR surrogate reproduced the selected Legacy Kármán policy sufficiently for declared closed-loop gates and named windows.” Not unique or fully explained. |
+| SR → near-constant skeleton | **Amber** | “Within the fixed topology/coefficient/40-step deletion contract, persistent rear constant rotation was the largest tested contribution; steady results provide magnitude context.” Not necessity or sufficiency. |
+| Steady constant → SR skeleton | **Amber** | “Steady and SR observations are consistent with a recurring rear-opposite constant component.” Do not pool their metrics or amplitudes. |
+| SR constant → CCD mean dominance | **Amber** | “Two analyses organize the bounded Kármán evidence around a dominant constant component.” State shared lineage; do not call causal triangulation. |
+| CCD → dynamic correction | **Amber** | “The residual is phase-coherent and action-associated in the declared cycle-template analysis.” Not response time, mechanism or causal correction. |
+| CCD → special physical modes | **Red/amber** | “CCD labels an action-associated organization of a low-dimensional residual; CCD is not shown superior to POD.” |
+| Dynamic correction → mechanism | **Red** | No mechanism sentence is authorized; retain action→near-body→wake→signature as hypothesis/open question. |
+| Theory → finite-Re mechanism | **Red** | “Theory provides an outer reference and sign/contract constraints, not finite-Re mechanism closure.” |
+| Solver qualification → mechanism | **Amber for numerics, red for physics** | “Selected configurations were qualified against bounded benchmark observables.” Kan99 lift sign remains unresolved; no control-mechanism inference. |
+| Legacy/V5 → generality | **Red if pooled; amber if bounded** | “Generality is limited to named, chain-specific conditions.” |
+| CFD → experiment | **Red** | “Experiments demonstrate deployment feasibility and qualitative open-loop wake modification; quantitative validation remains future work.” |
+
+## 3. Duplicate/correlated evidence
+
+1. Reward DTW and evaluation DTW reuse the same signal family, target, lag convention and often trajectory lineage; they are not independent wake evidence.
+2. Offline SR RMSE is measured on PPO-visited states; it is not closed-loop stability or field performance.
+3. SR long runs, deletion and scaling share the frozen formula, Legacy plant and parent lineage. Deletion is sensitivity, not a new realization.
+4. Steady sweep and SR constant term are correlated by control family and context; the sweep is calibration, not an independent intervention.
+5. CCD mean and phase analyses share one Kármán DRL lineage; different estimands add information, not stochastic replication.
+6. CCD rank stability, singular values, phase reconstructions and POD comparison are views of the same operator/19-cycle set.
+7. V5 seeds 41–45 are repeated stochastic searches in one case, not independent physical experiments; selection remains relevant.
+8. Kan99 regularized and inlet-sensitivity runs are numerical closures of one benchmark; their common lift-sign mismatch is a shared risk.
+9. Literature precedent supports plausibility, not local Re, metric, sign or observability validation.
+
+Use “convergent descriptive evidence under different estimands,” with lineage caveats. Do not say “independent triangulation” without independent parent realization, intervention, estimator and artifact.
+
+## 4. Structural risks
+
+- **Legacy/V5:** Legacy SR/CCD and modern V5 DRL differ in grid, API, inlet/BC, readback, calibration, normalization and smoothing. V5 retained `legacy-policy-v1` evaluation is not simply current native-v2. Never pool their numbers.
+- **Re convention:** Legacy `Re_code` uses `2D`; `re100` means `Re_D=50`. The steady contract uses `Re_D=50` directly. Every comparison must print `Re_code`, `Re_D`, `U0`, `D`, `nu`, chain and inlet.
+- **Action sign/body order:** order is `(front, rear_y_plus, rear_y_minus)`; accepted cloak is `[0,+Omega,-Omega]`, with `(Uw,Vw)=(-omega ry,omega rx)`: base-bleed/gap-jet, not boat-tail. Historical reverse labels are withdrawn. OID/CCD sign disagreement remains unresolved and must be quarantined.
+- **Metric/window:** native Legacy DTW, V5 calibrated DTW, target-normalized DTW, complete-cycle `E_mean`, nearest-snapshot `E_phase8`, steady `E_inf_vector`, quadrature L2 and event metrics are non-equivalent. Print comparator, window, crossings, calibration and status every time.
+- **CCD alias:** validated per-cycle sibling is complete; preferred pinball-inclusive publication is derived and canonical promotion remains pending. Name the exact artifact; do not use “canonical CCD” loosely.
+- **Experiment:** dye/open-loop observations support feasibility and qualitative modification only. No synchronized calibrated PIV/force package exists.
+- **Theory:** potential/MFS is an outer impermeability/reference problem, not rotating no-slip finite-Re physics. Circulation cancellation has a sign conflict with the accepted finite-Re fit; NS CV closure is incomplete.
+
+## 5. Structure: method chapters versus physics questions
+
+### Why SR/CCD/theory chapters fail
+They make three methods look like three independent proofs: SR appears to establish the skeleton, CCD to rediscover it, and theory to explain it. In fact they use non-equivalent estimands and related Legacy Kármán lineage. A reader may infer that SR terms are validated by CCD, or that CCD association is causal. OID would add a method chapter with no current positive result.
+
+### Risks of physics-first organization
+Physics-first is better, but each question can silently merge V5 performance, Legacy SR deletion, CCD mean reduction and steady endpoint. “Why does it work?” can promote the gap-jet hypothesis to mechanism; “is it general?” can convert n=1 demonstrations into a law. Every question needs a ledger row naming chain, estimator, window and status.
+
+### Recommended question order
+1. Can sparse observer-level control make the declared signature/field closer to target? DRL + field L2, with failures.
+2. What control structure is recoverable? Legacy SR, closed-loop validation and deletion.
+3. What is mean versus dynamic? Corrected CCD, explicitly descriptive.
+4. What physical context is compatible? Steady endpoint and short theory constraints; mechanism open.
+5. Where does it fail? Cross-Re/size, seeds, phase mismatch, solver/sign and experiment.
+
+Put a one-page `claim | chain | parent artifact | metric | window | comparator | n | status | prohibited inference` ledger before Results.
+
+## 6. Material disposition
+
+### Retain main text
+- One compact capability panel with exact chain labels and target/control/zero.
+- Reward/signature definition plus independent field metric.
+- Legacy Kármán SR law, contracts and closed-loop validation.
+- SR deletion panel for rear constant dominance under the fixed window.
+- CCD mean/dynamic scalar decomposition, with descriptive and CCD/POD caveats.
+- One steady/theory panel with accepted branch, `s` and outer-reference limitation.
+- Explicit limitations/generalization panel.
+
+### Supplement or downgrade
+- Full V5 11-case/15-run matrix and training histories.
+- n=1 cross-Re/size rows and LR-confounded disturbance-size comparisons.
+- Full SR topology/coefficient/scaling history.
+- Full CCD mode/phase/POD diagnostics.
+- OID equations/tests as method appendix only; no result.
+- Solver manifests/hashes, full metric/phase definitions and experiment apparatus/dye detail.
+- Legacy→V5 migration history and negative routes.
+
+### Delete from mainline
+Positive Illusion SR efficacy/target tracking/necessity/generalization; OID importance/causality; DTW-as-delay; potential circulation as accepted mechanism; boat-tail wording; unqualified Re=100; universal/optimal/causal/stability language; unaudited experiment numbers or digitized plots.
+
+## 7. Eight likely reviewer objections and preventive structure
+
+1. **Proxy reward:** Put reward and independent field metric together; state reward success is not field equivalence.
+2. **Cherry-picked seed:** Show all Re100 seeds, selected-seed rule, retention distribution and failures.
+3. **Inconsistent Re:** Add chain-specific `Re_code/Re_D` contract table.
+4. **In-sample SR:** Make closed-loop Legacy validation primary; label RMSE diagnostic; show failed gates.
+5. **False triangulation:** Share lineage and call constant/dynamic results descriptive convergent indications.
+6. **CCD causality:** State non-paired cycles, action association only, rank stability and POD near-equivalence.
+7. **Solver/sign error:** Expose Kan99 partial pass/lift mismatch, action oracle and withdrawn branch.
+8. **Experiment overclaim:** Label dye as qualitative open-loop feasibility; list missing PIV/force/calibration.
+
+## 8. Offline analyses/tables before writing (≤8, ranked)
+
+1. **Mandatory:** claim/evidence ledger with chain, artifact/hash, metric, comparator, window, n and forbidden inference.
+2. **Mandatory:** separate Legacy/V5 contract table (solver, BC, Re, order/sign, norm/calibration, reward/DTW, field metric and time units).
+3. **Mandatory:** all-five-Re100-seed selection/retention table with VecNormalize identity and field metrics.
+4. **Mandatory:** SR closed-loop table by case/window/comparator for PPO/SR/target/zero, including failed gates and deletion variants; separate offline RMSE.
+5. **High:** metric/window concordance table contrasting native/normalized DTW, `E_mean`, `E_phase8` and steady `E_inf_vector`, including Re400 disagreement.
+6. **High:** constant/dynamic provenance panel with CCD values, ROI/mask/window/cycles/rank and non-pairing caveat beside SR deletion and steady context.
+7. **High:** offline virtual sensor-plane/feature robustness (`x/D=8–12`, centre/outer/all-six signals) if artifacts pass contracts; supports reasonable, not optimal.
+8. **Medium:** solver/sign/CCD-alias audit table (Kan99 mismatch, accepted/reverse actions, sibling/derived CCD identities).
+
+Do not spend time on new CFD, theory closure, OID revival or unsupported experiment analysis.
+
+## 9. Paper boundaries
+
+### Minimal defensible paper
+**Claim:** In a named Legacy Kármán/cloaking setting, sparse wake-signature DRL reduces declared downstream field error versus zero; a symmetry-mapped SR surrogate passes declared closed-loop tests and exposes a dominant tested rear constant component with smaller state-dependent terms; corrected CCD descriptively separates a dominant constant-field contribution from a smaller action-associated residual.
+
+Include exact contracts, separate mean/phase metrics, SR closed-loop/deletion evidence, CCD bounded decomposition, steady/theory context and negative limits. Exclude universal generality, optimal placement, causal mechanism, stability, physical delay, Legacy/V5 equivalence, positive Illusion SR, solver-mechanism closure and quantitative experiment validation.
+
+### Ambitious but defensible paper
+Across a separated V5 capability matrix and a Legacy Kármán interpretation chain, recurring control organization is consistent with a near-constant rear-opposite backbone plus smaller dynamic correction, while breakdown is mapped. This demands complete selection accounting, exact chain labels, all failures, no pooled estimands, and a prominent “bounded organizing pattern, not universal law” warning. It remains moderate-to-high risk because readers will mentally pool chains.
+
+### Recommendation
+Choose the **minimal defensible paper** for the main submission. Retain the ambitious material as a controlled second layer in Supplement/Discussion. If the ambitious version is chosen, abstract/title must use “bounded,” “named-condition,” “descriptive” and “chain-specific,” and must not headline “universal,” “mechanism,” “causal,” “optimal” or “validated across Reynolds number.”
+
+## 10. Final editorial verdict
+
+**Green core:** named-condition Kármán DRL capability plus independent field evaluation; real Legacy SR closed-loop pipeline; physically motivated sparse signature.
+
+**Amber bridge:** SR constant interpretation, steady magnitude correspondence, CCD mean/dynamic organization and cross-condition capability. These are useful organizing evidence, not independent proof of a common mechanism.
+
+**Red bridge:** CCD/theory → causal finite-Re mechanism; pooled Legacy/V5 generality; experiment → quantitative validation; solver benchmark → control mechanism; positive Illusion SR.
+
+Write the paper so the red links are visibly absent. A skeptical JFM editor is more likely to accept a precise negative boundary around an interesting interpretation than a broad story whose strongest arrows require evidence the project explicitly does not possess.

docs/JFM_WYQ/ROUND2_TOP_DOWN/04_TOP_LEVEL_SYNTHESIS.md

@@ -0,0 +1,371 @@
+# JFM Round 2 — Top-level synthesis and freeze recommendation
+
+
+> **2026-08-10 author-ratified revision.** The author's final decisions are frozen in `05_REVISED_STRUCTURE_FREEZE.md` and supersede the affected recommendations below without changing their evidence status. In particular, the old title in §3, the capability-only treatment of late cases in §§4–5, the Kármán → steady narrative ordering, the Fig. 1 composition, and the index/card-style Round-3 schema in §12 are superseded. The retained evidence audit, claim ceiling, Legacy/V5 separation, positive-SR boundary, CCD/OID cautions and compact CFD-validation allocation remain in force.
+
+> **文档角色。** 顶层结构综合与作者讨论底稿，不是 manuscript prose。本文基于 Round 2 `01`–`03`、Round 1 `01`–`08`、coverage audit A7–A13、Nowledge Mem Context/Working Memory与定向 closure/manuscript-direction recall，并回读 DRL/SR/CCD/steady handoff与 claims、CelerisLab validation dossier、wake-L2/sensor dossier。未开展新文献检索、CFD、训练或实验分析。仓库 authority、manifest 与 immutable artifact 始终高于本综合。
+
+## 0. 单一推荐
+
+冻结一个**“能力先行、Kármán 深挖、物理问题组织、边界显式”**的结构：
+
+`惊艳但有合同标签的能力钩子 → 任务/测量/数值合同 → reward与独立L2检验 → 持续旋转分量 → mean/dynamic场分解与CCD → steady/theory约束 → 有限综合与失败边界`。
+
+不采用“DRL / SR / CCD / theory”四个平行方法筒仓，也不采用完整 case gallery。主文只深挖 **Legacy Kármán `re_code=100`（`Re_D=50`）**；V5 是独立的发现/保持与能力支持链，绝不与 Legacy 数值池化。全文最高物理结论是：**多种非等价估计量在该命名条件下共同支持一种“near-constant persistent rotation + smaller dynamic correction”的受限组织方式；它不是已闭合机制、普适规律或稳定性结论。**
+
+## 1. Editorial verdict：80% minimal core + 20% controlled ambitious outer layer
+
+### 1.1 精确比例
+
+采用 **80:20** 的论证权重：
+
+- **80% minimal core**：命名的 Legacy Kármán 主链；稀疏 signature/reward 与独立 field-L2；SR 的原 plant 闭环重部署和固定窗口 term ranking；corrected mean/dynamic estimand；CCD/POD 的描述性结果；所有合同、比较器与非主张。
+- **20% controlled ambitious outer layer**：V5 五 seed discovery/retention 支持与少量代表性跨 case 能力；steady `Re_D=50` 低亏损 endpoint 和理论 sign boundary；定性实验可行性；“constant + dynamic”作为讨论层组织假说。
+
+这不是把 20% 做成第二篇论文。外层材料只允许完成三件事：证明问题值得看、显示该组织并非只来自一张图、明确它在哪里失效。它不得改变 core 的 claim ceiling。
+
+### 1.2 为什么不是更激进或更保守
+
+- 若外层超过约 25%，读者会自然池化 Legacy/V5、把 SR/CCD/steady 当作三重独立机制证明，并把 n=1 case 与 qualitative experiment 读成 generality/validation。
+- 若外层低于约 15%，文章会退化为“一个 Legacy policy 的后验解释”，浪费 V5 多 seed、视觉能力、steady endpoint 和失败边界对 JFM 读者的价值。
+- 80:20 允许开头有足够的视觉冲击，同时把约四分之三的结果空间留给 reward、SR、field decomposition 与 Kármán physics。
+
+### 1.3 三路分歧的裁决
+
+1. **minimal vs ambitious**：minimal 是主干，ambitious 只作受控外层；不并列。
+2. **方法节 vs 物理问题节**：一级标题全部用物理问题/论证功能命名；SR、CCD、theory 仅作为二级节中的工具与证据来源。
+3. **“skeleton”是否进标题**：**不进推荐标题**。它可在正文被严格定义为“compact organizing skeleton”并在 Discussion 复现；当前证据只支持 dominant tested element 与 descriptive organization，标题中的 skeleton 会被编辑和审稿人理解成已识别且较稳健的物理结构。
+
+## 2. 一句话 paper identity 与 claim ceiling
+
+- **Question：** 在具有分离和周期脱涡的中等 Reynolds 数 fluidic pinball 中，稀疏下游 signature 驱动的学习控制能否被约化为一个持续旋转分量与较小动态修正，并由闭环 law、场分解和稳态约束在命名条件下共同限定？
+- **Central claim：** 对命名的 Legacy Kármán cloaking 合同，SR 闭环检验把持续的后圆柱反向旋转识别为最大 tested contribution，而 corrected field analysis 显示 constant-mean 分量占该 ROI 中 zero-to-DRL 误差降低的大部分，剩余较小 residual 是 phase-coherent、action-associated 且低秩的；steady 结果仅提供相容的量级和边界语境。
+- **Novelty：** 不是“DRL 首次控制 wake”，而是把**原 plant 闭环可执行的稀疏控制律、独立场误差分解、action-associated residual 与失败的理论解释**放在同一套显式 evidence contracts 下，形成可证伪的 bounded interpretation。
+- **Reader takeaway：** 该命名控制器的主要可测改善与持续后部反向旋转相联系，动态反馈提供较小但非零、相位组织的修正；这是一种紧凑组织，不是普适控制定律。
+- **Claim ceiling：** 最高可写 `consistent with`, `organizes`, `accounts for the measured error reduction under the declared estimand`, `dominant tested element`, `phase-coherent and action-associated`。不得升级为 `causes`, `mechanism`, `restores momentum`, `necessary/sufficient`, `optimal`, `stable`, `robust`, `general`, `universal`, `Legacy–V5 equivalent` 或 `experimentally validated`。
+
+“Cloaking”必须在首次出现时定义为**相对于指定 target/background 的 observer/field-space 接近**，不是 invisibility、全场恒等或跨观察者隐身。
+
+## 3. Working title
+
+### 推荐标题
+
+**Interpreting learned wake-signature cloaking in a fluidic pinball: persistent rotation and dynamic correction**
+
+理由：`wake-signature cloaking` 保留项目辨识度并限定 estimand；`interpreting` 与证据层级匹配；`persistent rotation and dynamic correction` 传达文章核心，但不预先宣判二者为唯一机制或普适 skeleton。
+
+### 有排序的备选
+
+1. **A bounded physical decomposition of learned wake-signature control in a fluidic pinball**  
+   最安全的备选；`bounded` 直接压住 claim，但 `physical decomposition` 可能让读者期待更严格的动力学正交分解。
+2. **A near-constant rotation skeleton in learned cloaking control of a fluidic pinball**  
+   最有记忆点；仅当结构冻结前的 ledger、SR deletion、CCD provenance 与 Legacy/V5 标签全部通过，且作者接受更高审稿风险时使用。
+3. **From sparse wake signatures to persistent-rotation control of a fluidic pinball**  
+   强调 measurement-to-action 路径；但弱化 smaller dynamic correction，容易被读成纯 constant-control paper。
+4. **Sparse-sensor reinforcement learning for wake-signature matching in a fluidic pinball**  
+   最保守；证据完全承受，但过弱，遮蔽 SR/CCD/steady 的真正新意，容易被归类为常规 DRL capability paper。
+
+### 词强度裁决
+
+- **不用过强词**：`mechanism`, `causal`, `universal`, `robust`, `optimal`, `stability`, `momentum restoration`, `experimentally validated` 都越过证据上限。
+- **推荐标题不使用 skeleton**：不是否定这一概念，而是把它留给正文定义、检验和限缩；标题不应把 amber bridge 写成 green result。
+- **不用过弱的纯 DRL 标题作为首选**：它会让编辑期待算法/训练创新，而文章最强贡献实际是受限的物理解释与证据纪律。
+
+## 4. 推荐最终一级/二级目录与空间预算
+
+规划基准：约 **8,500–9,500 word-equivalents，24–27 个紧凑 LaTeX pages（含主图/表/方程，不含 references/supplement）**。下表文字比例总和为 **100%**。页数是含本节图表占位的编辑预算，不是期刊限制。
+
+| 一级/二级目录 | 文字比例 | 约页数 | 物理任务 | 核心证据 | 主图/表/方程 | 主张上限 | 向下一节桥 |
+|---|---:|---:|---|---|---|---|---|
+| **1. Introduction: from visibility vocabulary to nonlinear wake signatures** | **10%** | 2.0–2.5 | 五段完成：定义 signature cloak；说明 nonlinear wake 与 linear cloak 不等价；pinball/DRL gap；解释为何需要 law + field audit；列贡献与限制 | Round1 literature index；JFM structural dossiers | Fig. 1 在文首被引用；不设历史大图 | 只说 linear literature 提供 vocabulary；不说连续物理谱系或 first | “先看控制能做到什么，再说明它在什么合同下成立” |
+| 1.1 Declared observer-space cloaking | 3% |  | 背景/target/zero 的定义 | target/zero/controlled contracts | 1 个简短 objective equation | declared-space approach only | 进入 nonlinear plant |
+| 1.2 Why a fluidic pinball is difficult and useful | 3% |  | separation、periodicity、symmetry/multi-attractor context | pinball literature | — | benchmark motivation, not project dynamics proof | 进入 sparse learning gap |
+| 1.3 Gap and contributions | 4% |  | 从 capability 到可执行 law 与场审计 | 本项目 authority | contribution bullets | bounded named-case interpretation | 引出视觉结果 |
+| **2. Phenomenology first: what the learned control can do** | **12%** | 3.0–3.5 | 先展示惊艳能力，但只用一个 funnel：headline Re100 → 2–3 个 representative named cases → 明确深挖选择 | Legacy/V5 flow roles、DTW、L2、case-specific metrics；失败/退化行 | **Fig. 1**, **Table 1** | named-condition capability；不做 gallery/generalization | “视觉改善是否只是 reward proxy，需先固定合同与独立指标” |
+| 2.1 Headline Kármán cloak | 5% |  | target/zero/controlled signature 与 field 的直接视觉比较 | Legacy principal role；可旁列 V5 support | Fig. 1a–d | observer/field error reduced under named comparator | 进入 breadth |
+| 2.2 Representative breadth and breakdown | 4% |  | 只放 vortex、`075` illusion 或 steady 中最能承担不同任务者；同时给 Re400/大 illusion 退化 | case-specific DTW/L2；V5 n/status | Fig. 1e–h；Table 1 compact rows | demonstration, not law | 解释为何选 Re100 |
+| 2.3 Why Kármán Re100 is the principal case | 3% |  | 说明其为 DRL/SR/CCD 的证据交集 | chain map | Table 1 chain/status | evidence intersection only | 进入合同/方法 |
+| **3. Problem, measurements and evidence contracts** | **15%** | 3.5–4.0 | 把 plant、action sign、sensors、reward、metrics、solver与 Legacy/V5 边界一次固定；Methods 不后置 | executable contracts、validation dossier、L2 dossier | **Fig. 2**, **Table 2**, Eq. 1–4 | methods/qualification only；solver qualification 不证明控制物理 | “在这些合同下检验 reward 是否对应扩展场改善” |
+| 3.1 Geometry, actuation and chain-specific Reynolds conventions | 4% |  | body order、wall law、`Re_code/Re_D`、Legacy/V5 | DRL/steady contracts | Fig. 2a–c；Eq. Re/action | exact contract | 进入 sensing |
+| 3.2 Sparse observations and reward | 4% |  | force + six velocity channels，DTW 的训练角色 | reward code；sensor rationale | Fig. 2d；Eq. reward/DTW summary | physically motivated, not optimal/observable | 进入 independent metrics |
+| 3.3 Independent field metrics and case-specific semantics | 4% |  | `E_mean`, `E_phase8`, L2/DTW分工；event/steady不同语义 | wake-L2 evaluator | Eq. L2/eta；Table 2 | geometry-guarded ROI；phase8 diagnostic | 进入 numerics |
+| 3.4 Numerical qualification and reproducibility boundary | 3% | 0.5–0.75 页纯文字等价 | 只交代所用配置、sensor readback、Sah04 pass、Kan99 partial pass与 Legacy 独立链 | 两份 validation dossier/manifest | Table 2 一角或 supplement pointer | selected-observable qualification only | 转入 reward–field 对照 |
+| **4. From reward success to field success** | **11%** | 2.5–3.0 | 重点解释 reward 的物理选择，并用 L2 反证“低 DTW=全场相同”；展示 seed selection/retention | all-five V5 Re100 seeds；Legacy/V5 separated DTW/L2；mean/phase disagreement | **Fig. 3**, Eq. 5 | reward/signature improvement accompanied by independent field improvement；不说 observability/optimality | “若性能真实，动作律本身是否复杂？” |
+| 4.1 What each reward term asks the flow to do | 3% |  | force balance 与 temporal signature 分工 | reward components | Fig. 3a | design rationale | 进入 metric comparison |
+| 4.2 DTW and L2: agreement and disagreement | 5% |  | L2 为 JFM 主证据，DTW 为 complementary signature metric；突出 Re400 mean/phase mismatch | canonical summary | Fig. 3b–d | association across metrics, not identity | 进入 seed/provenance |
+| 4.3 Discovery, selection and retention | 3% |  | 五 seed、same-selection VecNormalize、late retention | V5 retained artifacts | Fig. 3e；Table 1 | repeated discovery, seed-dependent retention | 进入 law reduction |
+| **5. How much of the learned action is persistent?** | **17%** | 4.0–4.5 | 围绕物理问题组织 SR：从 PPO 行为到原 Legacy plant 的可执行 sparse law，再做 term deletion | Stage1→2→3；canonical Kármán law；PPO/SR/zero；deletion/scaling；steady magnitude correspondence仅作末端 context | **Fig. 4**, **Table 3**, Eq. 6–7 | symmetry-mapped surrogate passes named closed-loop gates；rear constant is largest tested contribution in declared window | “动作中的 constant 是否对应场改善的大部分？” |
+| 5.1 Recovering an executable sparse law | 5% |  | causal pairing、imposed symmetry、CFD arbiter | SR authority | Fig. 4a；Eq. G map | surrogate, not unique/global law | 进入 formula |
+| 5.2 Persistent rear counter-rotation and state-dependent terms | 6% |  | 展示 front/rear law、准确 action order/sign | canonical formula | Fig. 4b–c；Eq. SR law | persistent dominant tested element；feedback secondary but nonzero | 进入 closed-loop |
+| 5.3 Closed-loop arbitration and bounded deletion | 6% |  | PPO/SR/zero、40/200/400 window区别、failed gates | Legacy runs/tables | Fig. 4d–f；Table 3 | parent-relative sensitivity, not necessity/sufficiency | 进入 field decomposition |
+| **6. What does the persistent component account for, and what remains dynamic?** | **18%** | 4.5–5.0 | 把 corrected mean decomposition 与 CCD 放在同一物理问题下，但不混淆 estimand；这是文章最大结果节 | zero/constant/DRL ROI errors；93.42/6.58；19×10 residual；rank-3；CCD/POD principal cosines | **Fig. 5–6**, Eq. 8–10 | constant mean accounts for most measured error reduction in exact ROI/window；residual phase-coherent/action-associated；CCD not causal/superior | “这种 constant recurrence 有何 steady 物理语境？” |
+| 6.1 Constant mean versus DRL mean | 7% |  | 三角色场与 additive scalar ledger | corrected ROI mean artifact | Fig. 5；Eq. mean identity | `accounts for measured reduction`，不是 actuator causal effect | 进入 residual |
+| 6.2 A separately centred dynamic residual | 4% |  | 明确非 paired counterfactual，定义 cycle-template residual | corrected CCD contract | Fig. 6a；Eq. residual | reference residual only | 进入 low-rank organization |
+| 6.3 Action-associated low-rank organization | 7% |  | rank-3 stability、symmetry coordinates、phase reconstruction、POD near-equivalence | CCD/POD publication | Fig. 6b–f；Eq. operator | descriptive action association; only subspace stable | 进入 steady context |
+| **7. What steady physics is compatible with the recurring rotation?** | **11%** | 2.5–3.0 | 用 steady endpoint、sign oracle 和失败 theory 约束解释；不承担 mechanism closure | accepted `[0,+,-]`, `s=3.55`, `E_inf_vector`; reverse mismatch；potential sign failure；partial NS ledger | **Fig. 7**, Eq. 11–12 | bounded free-slip endpoint and outer-reference constraint；gap/base-bleed is hypothesis | “把相容观察与未闭合箭头集中综合” |
+| 7.1 A low-deficit constant-rotation endpoint | 4% |  | direct profile与L2，accepted/reverse标签 | steady v3b | Fig. 7a–c；Eq. `s,E_inf` | settling-qualified endpoint, not optimum/stability | 进入 sign/theory |
+| 7.2 Rotation sign and the gap-facing motion | 3% |  | 显式 cardinal wall velocities，消灭 boat-tail误读 | sign oracle | Fig. 7d | base-bleed/gap-jet family wording only | 进入 theory boundary |
+| 7.3 Outer theory and the failed closure route | 4% |  | one-circle impermeability、rear dipole、`g` sign conflict、partial ledger | theory/NS authority | Fig. 7e–f；短 outer-reference equation | theory constrains; does not explain finite-Re mechanism | 进入 synthesis |
+| **8. Bounded synthesis, failures and outlook** | **6%** | 1.5–2.0 | 将 SR/field/CCD/steady 画成“相容但不等价”的证据图；列 failure、experiment status 与 decisive future tests | claim ledger；negative routes；qualitative experiment status | **Fig. 8**, **Table 4** | candidate organization, not universal law | 结束于可证伪的下一步，而非应用夸张 |
+| 8.1 Near-constant plus dynamic: what is and is not connected | 3% |  | chain/estimand/lineage map | cross-package ledgers | Fig. 8 | bounded organizing pattern | 进入 limits |
+| 8.2 Failure envelope and qualitative experiment | 2% |  | cross-Re/phase failure、Illusion SR negative、experiment feasibility | retained negatives | Table 4；实验仅可作 inset（有 provenance 才用） | no generality/validation | 进入 conclusion |
+| 8.3 Conclusions and decisive tests | 1% |  | 4–5条结论+matched intervention/NS closure需求 | evidence stress test | — | 重述 claim ceiling | — |
+| **Total** | **100%** | **24–27** |  |  |  |  |  |
+
+### Introduction 线性 cloak 历史硬上限
+
+Introduction 仅允许 **1 个功能段、约 180–250 words（正文总字数约 2–3%）** 处理 optical/active linear/water-wave cloak 与 illusion vocabulary。它必须以“not a Navier–Stokes equivalence”结束，随后立即进入 finite-Re wake、active hydrodynamic precedent、pinball 与 sparse-control gap。不得展开 science-fiction、metamaterial 分类史或应用场景目录。
+
+### CFD validation 空间裁决
+
+不接受 1.5–2.5 页独立 validation section。当前文章中 CFD validation 是支撑合同而非科学主角，主文控制在 **0.5–0.75 页 + Table 2 的若干行**；Kan99 mean-lift sign mismatch 必须可见。完整 K2/S2 配置、bands、hashes、smoke/partial matrix 放 supplement。
+
+## 5. 八幅主图 storyboard
+
+| 图 | Storyboard（建议 panels） | 现有数据支持 | 后续离线图/表需求 |
+|---|---|---|---|
+| **Fig. 1 Capability before explanation** | (a–c) Kármán target/zero/controlled 同色标 vorticity/velocity；(d) 三探针 signature；(e) vortex event；(f) `075` Illusion PPO；(g) steady accepted；(h) 一个明确 degradation/failure 小面板 | reproduction plots/manifests；periodic/event/late roles；V5/Legacy retained views | 统一坐标/色标/role/chain/case/seed/phase标签；核对每个 panel 的 raw manifest/hash；Vortex/Illusion若 metric contract 不完整，只作视觉 named demo，不给统一效率条 |
+| **Fig. 2 Plant and evidence contracts** | (a) geometry/body order；(b) wall-sign cardinal arrows；(c) sensors/ROI；(d) `q_target/q_zero/q_constant/q_DRL` roles；(e) Legacy/V5 两条不相交 provenance lanes | geometry/contract code；sensor dossier；role docs | 新绘 schematic；从合同自动导出坐标/Re/action sign；不依赖 CFD |
+| **Fig. 3 What was optimized vs what improved** | (a) reward components；(b) DTW vs `E_mean`；(c) `E_phase8`与frequency/period regularity；(d) Re400 mean/phase disagreement；(e) Re100 five-seed best vs late retention with selected normalizer | retained V5 tables；canonical wake-L2 summary | 必须先完成 five-seed/VecNormalize table 与 metric concordance；不拟合跨 case 通用趋势线 |
+| **Fig. 4 Executable sparse law** | (a) Stage1→2→3/CFD arbiter；(b) canonical formula tree；(c) PPO/SR action traces；(d) PPO/SR/zero signature；(e) constant deletion；(f) feedback/front deletion与failed gate | SR plotting package、telemetry、deletion tables | 统一 window/comparator；offline RMSE 与 closed-loop metric 分栏；exact formula/hash footnote |
+| **Fig. 5 How much the persistent component accounts for** | target/zero/constant/DRL mean fields；zero→constant与constant→DRL difference；ROI/mask inset；`E={0.4694,0.1110,0.0858}` 与 93.42/6.58 additive bar | corrected ROI mean artifact | 从 machine redraw fields 重绘，不 digitize PNG；每个 panel 印 ROI/common-mask/window/Legacy lineage；bar 明确是 error-difference share 非 energy/causal share |
+| **Fig. 6 What remains dynamic** | (a) residual definition；(b) singular strengths/rank checks；(c–d) rear-symmetric/antisymmetric action coordinates and modes；(e) four phase reconstructions；(f) CCD vs raw POD principal cosines | preferred pinball-inclusive CCD publication、plot_data | 解决 manuscript-facing canonical designation；保留 source sibling identity；简化现有四组图为一张主图，其余敏感性进 supplement |
+| **Fig. 7 Steady context and theory boundary** | (a) stationary/accepted/reverse profiles；(b) `E_inf_vector(s)` nearby points；(c) accepted field；(d) wall sign/gap-facing arrows；(e) outer potential reference；(f) `g_opt` vs descriptive `g` sign conflict + incomplete NS ledger badge | steady current display/CSV、theory artifacts、partial ledger | 重绘 accepted/reverse时显式标注 amplitude mismatch；不得画成机制闭环；MFS若 validation 未完成，只展示最短 one-circle/outer-reference结果，不展示“精美但未验证”MFS场 |
+| **Fig. 8 Integrated claim map and failure envelope** | 中心为 `persistent + dynamic`；四周分别 SR estimand、mean-field estimand、CCD association、steady endpoint；实线=定义/同链，虚线=compatible observation，断线=unresolved mechanism；侧栏列 Re400、Illusion SR、OID、experiment limits | Round1/2 ledgers | 新绘 evidence graph；每条边标 chain/metric/status；实验图片只有在 raw provenance 核实后作为小 inset，否则不进主图 |
+
+**实验图裁决：** 不预留第九幅主图。若 raw dye image/video、condition、frame rate/scale 与 trial provenance 在出图前可核实，可把一张小 inset 放 Fig. 8；否则实验全部进 supplement。没有同步 PIV/force 时，任何情况下都不把实验置于 headline Fig. 1。
+
+## 6. Tables 与 equation budget
+
+### 6.1 主文 tables（4个）
+
+1. **Table 1 — Capability and evidence status.** 只列 6–8 个代表行：case、chain、Re convention、seed/n、target/zero、DTW、L2 semantic、status、failure。完整 11-case/15-run 与 Legacy roles 进 supplement。
+2. **Table 2 — Plant/solver/metric contracts.** Legacy 与 V5 并排但绝不合并：grid/BC、`Re_code/Re_D`、body/action sign、observation order、normalization、reward/DTW、field metric、solver qualification状态。CFD validation 数字只保留 Sah04 S2 pass 与 Kan99 K2 partial pass摘要。
+3. **Table 3 — SR law and closed-loop evidence.** formula identity、case/window、PPO/SR/zero、deletion variant、status与 prohibited inference；offline RMSE 单列为 diagnostic。
+4. **Table 4 — Cross-analysis claim ledger.** `claim | chain | parent artifact | estimand | comparator/window/n | status | strongest wording | prohibited inference`，同时承担 final limitations table。
+
+### 6.2 Equation budget
+
+主文 **12 个 numbered equations 上限**，其中只有 8–10 个需要独立展示；其余可行内或合并：
+
+1. `Re_D` 与 Legacy `Re_code` 映射；
+2. wall velocity/action sign 与 `s`；
+3. sparse objective/reward（只保留物理组成，不展开 PPO）；
+4. DTW role 的简式或定义性引用；
+5. registered field RMS 与 `eta`；
+6. SR reflection map；
+7. canonical SR law；
+8. mean-field additive identity与 `E_r`；
+9. separately centered dynamic residual；
+10. weighted CCD operator/mode normalization；
+11. steady `E_inf_vector`；
+12. one-circle/outer dipole或 incomplete NS ledger（二选一主文，另一项 supplement）。
+
+Supplement 放：DTW dynamic programming recursion、PPO objective、完整 SR feature library、CCD/POD代数细节、MFS collocation/images/multipoles、完整 NS CV identity、phase extraction、mask/quadrature实现。Equation 数量服务于论证，不为展示理论深度而扩张。
+
+## 7. Main text vs supplement
+
+### 7.1 Main text 必须保留
+
+- cloak/signature 的精确定义与 named comparator；
+- 一张 capability hook，含至少一个失败/退化信号；
+- Legacy/V5、`Re_code/Re_D`、body order/action sign 的可见合同；
+- reward与DTW的物理角色，以及 L2 为独立主要 field evidence；
+- Re100 五 seed selection/retention 的最小完整交代；
+- canonical Legacy Kármán SR law、imposed symmetry、Stage-3 closed-loop arbiter；
+- rear constant deletion与“dominant tested”边界；
+- zero/constant/DRL mean decomposition及93.42/6.58的精确 estimand；
+- corrected residual、rank-3、action association与POD near-equivalence；
+- steady accepted branch、`s=3.55` endpoint、wrong-sign/reverse与potential sign failure；
+- Kan99 partial pass、Illusion SR negative、OID claim-free、experiment qualitative 的短限制；
+- 一页 cross-analysis claim ledger。
+
+### 7.2 Supplement
+
+- V5 全 11-case/15-run matrix、所有 seeds/curves/checkpoints/normalizers、selection细节；
+- Legacy/V5完整 solver/config/API/readback/migration合同与hash ledger；
+- periodic Kármán/Illusion的完整 native/normalized DTW、`E_mean/E_phase8`、phase/frequency diagnostics；
+- Vortex event L2、Erase mean L2（只有合法 exact roles 时）、steady full profiles/L2；
+- SR 全 topology、feature library、fit/split/provenance、per-case coefficients、200/400/named extension、all deletion/scaling、failed gates；
+- CCD 全 domain/mask/bin/origin/LOCO sensitivity、全部 modes/phase/POD plots、wake-only sibling与pinball-inclusive派生关系；
+- OID 只作为 method/reset/negative audit appendix，不放 positive result；
+- steady sweep/reverse细节、MFS validation、Laurent/one-circle checks、partial NS diagnostics与所有 open gates；
+- CFD K2/S2完整 qualification、bands、hashes、smoke说明；
+- experiment apparatus、raw dye sequence、conditions与 limitations（若 provenance 可核实）；
+- archive/negative-route、exclusion、artifact-path/hash与claim-status ledgers。
+
+### 7.3 OOD/OID裁决
+
+- **OOD（若指 out-of-distribution/generalization）**：不设一级/二级主结果节；只在 capability Table 1 和 Section 8 failure envelope 中占约 0.25–0.5 页。n=1 cross-Re/size 不构成 OOD law。
+- **OID**：不设结果节，不占主图；主文最多 2–4 句说明 active core claim-free、旧路线为何不能承担机制证据。方法方程与 reset history 进 supplement。
+
+## 8. Legacy / V5 的具体呈现语言与布局
+
+### 8.1 固定语言
+
+首次出现时使用：
+
+- **Legacy analysis chain**：`Legacy Kármán re_code=100 (Re_D=50)`；支持 principal SR closed-loop、corrected mean/CCD 与标准化 Target/PPO/SR/Zero roles。
+- **V5 support chain**：`CelerisLab V5 Kármán code-Re100 (Re_D=50 under its stated convention)`；支持现代 solver/context、五 seed discovery/retention 与 named-condition capability。
+
+推荐句式：
+
+> LegacyCelerisLab and CelerisLab V5 are treated as separate evidence chains. Their geometry, boundary conditions, APIs, readback, normalization/calibration and retained-policy contracts differ; no numerical pooling or equivalence is assumed.
+
+> The Legacy chain is used for the principal policy-reduction and field-interpretation analysis, whereas V5 supplies independent support on policy discovery, retention and named-condition capability.
+
+### 8.2 固定布局
+
+- 所有混合 figure 采用两条水平泳道或明显边框：**Legacy analysis** / **V5 support**。
+- 不在同一回归线、均值、error bar 或颜色连续谱中混合两链；若同一 panel 显示，marker shape 与背景带必须同时区分。
+- Table 2 固定双栏；每个数值行打印 solver/config、Re定义、role、seed/n、metric/window。
+- 章节顺序上先用 V5/representative cases 做 capability hook，再明确“从此处起的深分析回到 Legacy Re100”；Section 7 steady 再次标明 V5 free-slip 独立 contract。
+- 不用 `validation across implementations`、`reproduction of Legacy`、`consistent numerical continuation`。可用 `separate supporting chain`, `contract-driven rewrite`, `qualitatively compatible at named scope`。
+
+### 8.3 V5不能做的事
+
+V5不能 rescue Legacy SR、为 CCD 提供 independent replication、把 Illusion SR 负结果改写为正、证明 steady mechanism、或把五 seed policy search 变成物理独立实验。V5 late retention 必须写成 seed-dependent；非 Re100 case 大多 n=1。
+
+## 9. 八项 offline 工作重新排程
+
+这里严格重排 pressure-test 的八项，不新增第九项。
+
+### A. 结构冻结前必要
+
+1. **Claim/evidence ledger**：`claim | chain | artifact/hash | metric | comparator | window | n | status | forbidden inference`。  
+   决定哪些箭头可进入目录、abstract和标题；未完成不得冻结 Section 5–8 的措辞。
+2. **Legacy/V5 contract table**：solver、BC、Re、body/action sign、normalization/calibration、reward/DTW、field metric、time units。  
+   决定 Figure 2/Table 2 与所有跨链桥是否成立。
+3. **All-five Re100 seed selection/retention table**：same-selection `VecNormalize` identity、best/late、field metrics、selected-seed rule。  
+   防止 capability hook 被批评为 cherry-picking。
+4. **SR closed-loop table**：PPO/SR/target/zero、case/window、failed gates、deletion variants，offline RMSE 独立列。  
+   决定 “executable sparse law” 与 “dominant tested” 是否能进入 central claim。
+
+### B. 写正文前完成
+
+5. **Metric/window concordance table**：native/normalized DTW、`E_mean`、`E_phase8`、steady `E_inf_vector`，突出 Re400 disagreement。  
+   用于统一术语，防止正文中把不同指标互换。
+6. **Solver/sign/CCD-alias audit table**：Kan99 sign mismatch、accepted/reverse action、OID/CCD sign conflict、validated sibling与pinball-inclusive publication身份。  
+   用于消除 caption 与 methods 中最危险的 provenance/sign 错误。
+
+### C. 出图前完成
+
+7. **Constant/dynamic provenance panel ledger**：CCD values、ROI/mask/window/cycles/rank/non-pairing caveat，与 SR deletion和steady context并排但不合并。  
+   这是 Fig. 5–8 的 source-of-truth；必须先于视觉合成。
+
+### D. 可选
+
+8. **Offline virtual sensor-plane/feature robustness**：`x/D=8–12`、centre/outer/all-six。  
+   只在作者/审稿人需要“placement reasonable”之外的项目内 robustness 时做；它不能建立 optimality，不应阻塞结构或正文。
+
+## 10. 不纳入文章的内容
+
+| 不纳入主线 | 理由 |
+|---|---|
+| positive Illusion SR tracking/efficacy/generalization/term necessity | standardized long-window结果接近 physical zero，当前为 historical/negative |
+| OID importance、control authority、causality或active scene result | active OID 是 claim-free method core，旧 force gates失败且action受feedback混杂 |
+| old temporal/phase CCD、response lag、CCD superiority | 已被 corrected cycle-template route取代；当前只能描述association，POD子空间近等价 |
+| potential circulation cancellation作为finite-Re机制 | `g` sign冲突；Euler impermeability不提供 rotating no-slip closure |
+| closed momentum/energy restoration、gap jet causes cloak | NS ledger不完整，缺matched intervention与wall-resolved closure |
+| boat-tail表述 | accepted `[0,+,-]` wall motion 对应 gap-facing downstream motion，是base-bleed/gap-jet family，不是已证明boat-tail |
+| universal/continuous Re或size law | 大多数跨条件 n=1，metrics/learning rate/contracts不统一 |
+| full case/seed gallery、所有训练曲线 | 稀释Kármán深分析并制造gallery paper；放supplement |
+| OOD/OID独立一级节 | 证据角色小且易过度承诺；仅限failure/method boundary |
+| 1.5–2.5页 CFD validation chapter | solver在本文角色小；当前qualification partial且不证明control physics |
+| 长线性 cloak/metamaterial史 | 只提供vocabulary，拖延nonlinear gap并暗示错误等价 |
+| seals、marine structures、sediment、defence/low-observability应用 | 已被作者/导师路线否决或证据不足；应用只留一句前瞻 |
+| quantitative experimental efficacy、PIV/force agreement | 无完整同步校准包；dye仅支持qualitative open-loop feasibility |
+| archival transfer analyses与计划中的 parity/post-fix retraining | archive/plans不是当前科学authority |
+| 未完成MFS精美场图或未闭合CV图 | 易把计算 capability 误读为 validated explanation |
+
+负结果不等于删除：Re400 phase failure、Illusion SR collapse、Kan99 lift sign、potential sign failure与实验限制必须在主文短而可见；其完整材料进 supplement。
+
+## 11. 只需作者决定的四个问题
+
+### Q1. 标题是否允许使用 “skeleton”？
+
+- **明确推荐：** 不用于工作标题；保留为 Section 8 的定义性术语。
+- **影响：** 使用会显著提高辨识度，也会把审稿预期从 bounded interpretation 提升到结构唯一性、跨条件稳健性和物理闭合。
+- **默认选项：** 采用推荐标题 *Interpreting learned wake-signature cloaking...*；只有四项结构冻结前 offline ledger 全部通过且作者主动接受更高风险，才切换备选2。
+
+### Q2. 文章的首要名词是 “cloaking” 还是 “wake-signature matching”？
+
+- **明确推荐：** 标题和引言采用 `wake-signature cloaking`，方法定义为 declared target/background matching；表格与指标处优先用 `signature/field error`。
+- **影响：** 纯 `cloaking` 更醒目但容易被要求全场/多观察者等价；纯 `matching` 最安全但失去项目概念新意。
+- **默认选项：** 双层用法：概念层 `wake-signature cloaking`，定量层 `target-relative signature/field matching`。
+
+### Q3. headline capability 图是否保留一个 Illusion PPO panel？
+
+- **明确推荐：** 保留一个明确标注 `PPO capability; no positive SR claim` 的 `075` Illusion panel；不进入后续深分析。
+- **影响：** 保留可兑现“先展示惊艳能力”和 cloak/illusion项目范围，但会增加读者误以为 Illusion SR/跨size是本文主张的风险；删除则文章更聚焦但视觉范围变窄。
+- **默认选项：** 保留单 panel + Table 1 单行，完整 Illusion cases、saturation和negative SR进 supplement/limits。
+
+### Q4. manuscript-facing CCD package 采用哪个正式称呼？
+
+- **明确推荐：** 主文图采用 pinball-inclusive preferred reader publication，但在 caption/ledger 中把 validated wake-only per-cycle sibling列为 primary source lineage；不要把两者合称一个“canonical result”。
+- **影响：** 选择决定 Fig. 6 的可重绘入口、artifact citation与审稿复现路径；错误合并会掩盖 derived-vs-primary关系。
+- **默认选项：** 在用户完成 canonical-promotion choice 前，使用 `preferred pinball-inclusive derived publication, bound to the validated per-cycle source`，不写 `canonical CCD`。
+
+实验是否进主文**不作为作者偏好问题**：由 raw image provenance gate 决定；通过则 Fig. 8 inset，不通过则 supplement。其余 case、metric、sign、claim ceiling 均由证据判断，不推给作者。
+
+## 12. Round 3：先做 section-level evidence briefs，不直接写全文
+
+Round 3 应按本目录生成 **8 个 section-level evidence briefs**，每个 brief 是可审计的“写作接口”，不是段落草稿。建议文件与责任如下：
+
+1. `R3_01_INTRO_AND_GAP_BRIEF.md`
+2. `R3_02_CAPABILITY_BRIEF.md`
+3. `R3_03_CONTRACTS_AND_METHODS_BRIEF.md`
+4. `R3_04_REWARD_TO_FIELD_BRIEF.md`
+5. `R3_05_PERSISTENT_ACTION_BRIEF.md`
+6. `R3_06_MEAN_DYNAMIC_CCD_BRIEF.md`
+7. `R3_07_STEADY_CONTEXT_BRIEF.md`
+8. `R3_08_SYNTHESIS_LIMITS_BRIEF.md`
+
+### 每个 brief 的固定 schema
+
+- **Section question / one-sentence answer**：只写本节要回答的物理问题与当前最强答案。
+- **Claim ladder**：`green / amber / red`，每条 claim 给 strongest wording 与 prohibited wording。
+- **Evidence cards**：每张卡必须含 `chain, case/seed/role, artifact/hash, metric/estimand, comparator, window, n, status, exact number`。
+- **Lineage/correlation note**：指出哪些证据共享 parent realization、不是独立重复。
+- **Required equations**：只列方程ID、定义目的、变量合同和放主文/补充的理由。
+- **Figure/table panel contract**：panel question、source arrays、色标/ROI/mask/phase、caption必须包含的限制、是否需要离线重绘。
+- **Paragraph function map**：每段只列功能（claim → evidence → limitation → bridge），不写成 manuscript prose。
+- **Cross-section imports/exports**：从上一节接收什么定义，向下一节交付什么已建立事实；禁止重复定义或跨链偷渡。
+- **Open gate**：仅列在进入正式写作前真正阻塞该节的 offline work；不新增 CFD/文献调研愿望清单。
+- **Word/figure budget**：不得突破本文件Section 4分配。
+
+### Round 3 的编排顺序
+
+1. 先完成 `03 contracts`、`05 SR`、`06 mean/dynamic/CCD` 三个硬证据 brief；
+2. 再完成 `04 reward-to-field` 与 `07 steady context`；
+3. 用这些结果反向收紧 `02 capability` 与 `01 introduction`；
+4. 最后写 `08 synthesis/limits`，并运行一次 whole-brief claim/terminology/Legacy-V5 consistency audit；
+5. 只有所有 brief 的 artifact/metric/window/claim status 通过，才进入 Round 4 的 rough section drafting。不得在 Round 3 直接并行写全文，因为那会在合同冻结前固化跨链措辞、重复方法并放大三路分歧。
+
+## 13. Freeze checklist
+
+讨论后若以下项目无异议，即可冻结结构：
+
+- [ ] 80:20 core/outer editorial ratio；
+- [ ] 推荐标题register与`skeleton`正文级定位；
+- [ ] 8节物理问题目录及100%空间预算；
+- [ ] Legacy Re100唯一深分析、V5独立support；
+- [ ] 8主图、4表、12方程上限；
+- [ ] L2为主要独立场证据，DTW为signature/reward-compatible补充；
+- [ ] CFD validation 0.5–0.75页，不设独立一级节；
+- [ ] OOD/OID不占显著正文；
+- [ ] experiment由provenance gate决定主文inset或supplement；
+- [ ] central claim不越过 descriptive/bounded organization；
+- [ ] Round 3先做8个evidence briefs，不写全文。
+
+**最终 editorial verdict：** 这篇文章最有竞争力的形态不是“DRL展示很多 cloak/illusion cases”，也不是“SR、CCD和theory各自证明同一机制”，而是：**用早期视觉能力建立问题价值，用reward与L2建立测量可信度，用Legacy闭环SR和corrected field decomposition解释Kármán主例，再以CCD和steady/theory明确动态剩余与物理边界。** 读者应带走一个清晰、有限、可证伪的 `persistent rotation + dynamic correction` 图景，同时明确知道哪些箭头仍未被证明。
+
+
+## 14. Author-ratified revision summary (controls conflicts above)
+
+- **Provisional frozen title:** *Interpreting learned hydrodynamic cloaking and illusion in a fluidic pinball*. There is no colon and no A-opening. Interpretation depth remains cloak/Kármán-centred; illusion and erase are bounded extensions. The title does not authorize positive Illusion SR.
+- **Terminology hierarchy:** concept/goal = hydrodynamic cloaking and illusion; learning/execution = wake-signature matching using forces plus sparse downstream observations; independent quantitative validation = field matching, with L2 primary for JFM and DTW complementary and case-specific. `control skeleton` is interpretation-level language only. CCD remains `CCD`; its manuscript-facing physical outputs are `action-correlated correction-field modes`, descriptive and noncausal.
+- **Corrected progression:** steady → Kármán → vortex → illusion/erase → experiment. Steady rotation was discovered first; later periodic Kármán analysis found a learned law closer to steady than expected, with a dominant steady/near-constant core and smaller dynamic correction. Kármán Re100 remains dominant because it has the richest evidence intersection.
+- **Late extensions:** vortex removes periodicity and supports only a bounded comparable-scale incoming/arbitrary-flow hypothesis; use acquired phase-midpoint event-snapshot downstream-ROI L2 plus current DTW. Illusion/erase extend suppression/background matching toward retargeting/erasure; use periodic Illusion complete-cycle mean/phase L2 plus normalized DTW, and Erase mean-field L2 plus DTW only where exact roles exist. No shared metric contract and no positive Illusion SR.
+- **Figures and experiment:** Fig. 1 remains left-to-right capability/progression, Kármán visually dominant, with steady/vortex/`075` Illusion smaller; Illusion is PPO capability only. Experiment follows the extensions and is qualitative open-loop feasibility: straighter steady dye streaklines and altered illusion shedding frequency, without synchronized calibrated force/PIV validation. Frank will draft that section later.
+- **Structure and workflow:** CFD validation remains fully covered but compact (about 0.5–0.75 page plus contract-table rows and supplement). Kármán deep analysis remains the largest block, while vortex and illusion/erase receive meaningful late main-text space. Round 3 is section/module based and approaches JFM prose: author-readable structured drafts/dossiers with narrative paragraphs or paragraph functions, evidence citations/contracts, figure needs and claim boundaries. A detailed chapter-level JFM style study must precede or guide drafting. The earlier evidence-card/index-first Round-3 recommendation is superseded.

docs/JFM_WYQ/ROUND2_TOP_DOWN/05_REVISED_STRUCTURE_FREEZE.md

@@ -0,0 +1,133 @@
+# Round 2 — Revised structure freeze after author feedback
+
+> **Status.** Author-ratified Round-2 structure freeze and Round-3 handoff. This is not polished manuscript prose and does not alter evidence packages. It controls conflicts with Round-2 `01`–`04`; superseded recommendations remain in those files as provenance. Repository authority, manifests and immutable artifacts still control scientific facts.
+
+## 1. Frozen paper identity
+
+### Provisional working title
+
+**Interpreting learned hydrodynamic cloaking and illusion in a fluidic pinball**
+
+The wording is frozen provisionally: no colon and no opening article “A”. “Hydrodynamic cloaking and illusion” names the concept and project goal, not the evidentiary strength of every case. Interpretation depth is explicitly cloak/Kármán-centred; Illusion and Erase are bounded objective extensions. The title therefore does **not** revive positive Illusion SR, target-tracking, efficacy-over-zero, necessity or generalization claims.
+
+### Terminology hierarchy
+
+1. **Concept/goal:** `hydrodynamic cloaking and illusion`. Cloaking means approach to a declared background/reference; illusion means retargeting a declared hydrodynamic signature. Neither means universal invisibility or full-state equivalence.
+2. **Learning/execution:** `wake-signature matching using forces plus sparse downstream observations`. This names what the policy receives and optimizes; it does not establish observability or optimal sensing.
+3. **Independent quantitative validation:** `field matching`. Spatial L2 is the primary JFM quantitative metric. DTW is complementary, signal-level and case-specific; it is not a physical delay, causal diagnostic or universal cross-case score.
+4. **Interpretation only:** `control skeleton`. It may summarize the dominant steady/near-constant core plus smaller dynamic correction in the Kármán analysis, but is not a title-level mechanism, unique law or cross-case result.
+5. **Field organization:** retain `CCD`; call the manuscript-facing physical outputs **action-correlated correction-field modes**. They are descriptive/noncausal, and CCD is not claimed superior to POD.
+
+## 2. Frozen physical progression
+
+The narrative is not “Kármán to steady.” The historically and physically correct progression is:
+
+`steady discovery → Kármán periodic interpretation → vortex nonperiodic extension → illusion/erase objective extension → qualitative experiment`
+
+### Steady → Kármán
+
+Steady rotation control was discovered first. Later Kármán analysis showed that the periodic learned law differs less from steady control than expected: it retains a dominant steady/near-constant rear-rotation core with a smaller dynamic correction. This ordering supplies the discovery logic, but does not make the separate V5 steady contract evidence for the Legacy SR/CCD numbers. Kármán Re100 remains the deep-analysis case because it has the richest intersection of PPO, executable SR, independent fields, corrected mean/dynamic decomposition and CCD.
+
+### Kármán → vortex
+
+The vortex case removes periodicity. Pulse-like action supports only a bounded hypothesis that comparable-scale combinations of incoming/arbitrary flow and the pinball wake can be handled; do not claim arbitrary-flow capability or universality. Evaluate the acquired **event-snapshot downstream-ROI L2 at the acquisition phase-midpoints**, preserving event labels/offsets rather than relabelling them periodic phase, together with the current DTW contract. If an exact target/control/zero event role or schema is absent, report the gap rather than manufacturing a metric.
+
+### Cloak → illusion/erase
+
+This extends the control objective and the force-centred conceptual argument from suppression/background matching to retargeting/erasure. For periodic Illusion, use complete-cycle mean L2 and phase-resolved L2 separately, plus normalized DTW. For Erase, use mean-field L2 plus DTW only where exact target/control/comparator roles exist. Do not force these estimands into the Kármán or vortex metric contract. Illusion is PPO capability evidence only; Illusion SR remains historical/negative.
+
+### Experiment follows the extensions
+
+The experiment is qualitative open-loop feasibility only: steady operation shows straighter dye streaklines, while illusion operation shows a shedding-frequency change. There is no synchronized calibrated force/PIV validation, no quantitative field match and no closed-loop experimental claim. Frank will write this section later; the structure must reserve its role without inventing details or numbers.
+
+## 3. Frozen section hierarchy and space allocation
+
+Use a physics/progression-led hierarchy, while making Kármán interpretation the dominant analytical block. Percentages are body-text planning shares and sum to 100%; figures/tables may alter page equivalents.
+
+1. **Introduction: hydrodynamic cloaking, illusion and the interpretation gap — 10%**  
+   Define the concept/goal, distinguish nonlinear finite-Re wake control from linear cloaking analogies, introduce force-plus-sparse-observation wake-signature matching, and state the bounded contribution.
+
+2. **Problem, execution and evidence contracts — 13%**  
+   Geometry, action/body/sign conventions, Legacy/V5 separation, observations and reward, L2/DTW hierarchy, case-specific estimands, and compact CFD qualification. CFD validation receives about **0.5–0.75 page**, contract-table coverage, and full supporting supplement treatment.
+
+3. **Capability and progression: steady, Kármán, vortex and illusion — 11%**  
+   Introduce the historical steady discovery, then the Kármán headline and smaller vortex/`075` Illusion demonstrations. Establish why Kármán Re100 is selected for depth. Do not reduce this section to a pooled performance gallery.
+
+4. **Kármán performance: signature objective and independent field matching — 12%**  
+   Put L2 first for JFM validation and DTW in its complementary signal role; include seed/retention and mean/phase distinctions with Legacy/V5 labels.
+
+5. **Kármán action law: dominant near-constant core and smaller correction — 17%**  
+   Present executable Legacy SR, symmetry, closed-loop arbitration and bounded deletion. “Control skeleton” appears only as an interpretation-level organization; positive SR scope remains Kármán/cloak.
+
+6. **Kármán correction fields and action-correlated modes — 17%**  
+   Present the exact constant-mean versus DRL mean estimand, separately centred dynamic residual, and CCD action-correlated correction-field modes. Preserve non-pairing, descriptive/noncausal and CCD/POD boundaries. OID remains claim-free and has no results subsection.
+
+7. **Cross-case extensions beyond periodic cloaking — 10%**  
+   Give meaningful main-text space to vortex nonperiodic event metrics and illusion/erase objective extensions. Keep their metric contracts distinct and show boundaries/failures alongside capability. This is a late scientific extension, not merely an early gallery.
+
+8. **Steady context, physical synthesis and theory limits — 6%**  
+   Return to steady as the historical and physical reference for the Kármán near-constant core; retain accepted sign, separate chain, bounded endpoint, failed potential-sign route and incomplete NS closure. The narrative return does not reverse the discovery order.
+
+9. **Qualitative experiment, limitations and conclusions — 4%**  
+   Reserve a short author-written open-loop feasibility subsection, then state chain/metric/causality limits and decisive future tests.
+
+Kármán-specific Sections 4–6 occupy 46% and remain dominant. Sections 7 and 9 ensure the late extensions have visible main-text roles without diluting the positive Kármán/cloak interpretation scope.
+
+## 4. Frozen figure and table logic
+
+### Figure 1
+
+Keep a left-to-right **capability/progression composition**. Kármán must be visually dominant. Smaller panels show steady, vortex and `075` Illusion. Panel labels and caption must identify case, chain, role, comparator and metric status. The Illusion panel is explicitly `PPO capability; no positive SR evidence`. Do not pool efficiencies or imply one metric contract. The experiment follows the CFD extensions in the manuscript and is not part of the frozen Fig. 1 composition.
+
+### Remaining main figures
+
+- plant, force/sparse-observation, action sign, metric and Legacy/V5 contracts;
+- Kármán DTW versus independent mean/phase L2 and seed/retention evidence;
+- executable Kármán SR law, closed-loop arbitration and bounded deletion;
+- constant/DRL mean-field matching and exact estimand;
+- residual definition and CCD action-correlated correction-field modes with POD comparison;
+- vortex event-midpoint ROI-L2 + DTW and periodic Illusion/Erase case-specific extension panels;
+- steady discovery/context, accepted/reverse sign boundary, outer-theory failure and integrated claim limits.
+
+Use a compact contract table that keeps Legacy/V5 and all metric/window semantics separate. Full solver qualification, case matrices, sensitivities, artifact/hash ledgers and negative routes belong in the supplement, while essential comparators and limitations remain visible in the main text.
+
+## 5. Evidence and claim boundaries that remain frozen
+
+- **Legacy/V5:** separate chains; never pool or imply numerical continuation/equivalence.
+- **Positive SR:** Kármán/cloaking only. Illusion SR is historical/negative; steady is contextual.
+- **CCD:** descriptive action correlation only; no causality, response time, mechanism, stability or superiority claim. Use `action-correlated correction-field modes`.
+- **OID:** active package is claim-free; no positive result, importance or control-authority claim.
+- **Metrics:** periodic complete-cycle mean/phase L2, vortex event snapshots, Erase mean fields, steady profiles/`E_inf_vector`, native/normalized DTW are non-equivalent.
+- **Vortex:** comparable-scale pulse-response hypothesis only, not arbitrary-flow universality.
+- **Experiment:** qualitative open-loop feasibility; no synchronized calibrated force/PIV validation.
+- **Theory:** outer reference and failure boundary only; no finite-Re mechanism closure.
+
+## 6. Round-3 organization approaching JFM prose
+
+Round 3 is **section/module based**, not another bottom-up index and not a set of AI-only evidence cards. A detailed **chapter-level JFM style study** must precede the drafting agents or actively guide them. It should examine paragraph progression, placement of definitions and caveats, figure-to-text cadence, evidence density, transitions, and how comparable JFM papers separate capability from interpretation. Exact bibliography identities/DOIs and read levels remain required.
+
+Each Round-3 output must be an author-readable structured section draft/section dossier containing:
+
+1. section purpose and one-sentence bounded answer;
+2. narrative paragraphs or a paragraph-by-paragraph function map approaching prose;
+3. evidence citations/contracts for every quantitative or physical claim, including chain, case/role, comparator, metric, window and artifact authority;
+4. figure/table needs and caption-level limitations;
+5. explicit claim boundaries, conflicts and non-claims;
+6. definitions imported from previous sections and the bridge exported to the next;
+7. unresolved gates that truly block drafting.
+
+Sentence-level polishing is not required. Agents should not emit raw indexes, card stacks or machine-oriented handoffs as the primary output. Evidence cards may exist underneath as working support, but the delivered module must read as a coherent section plan/draft for Frank. Frank retains authorship of the experiment subsection.
+
+Suggested module order: (i) JFM chapter-level style study; (ii) contracts/metrics; (iii) Kármán performance; (iv) Kármán SR action law; (v) Kármán field/CCD interpretation; (vi) cross-case vortex/illusion/erase extensions; (vii) steady-context synthesis; (viii) introduction and capability progression; (ix) Frank's experiment section; (x) limitations/conclusion and whole-manuscript terminology/claim audit.
+
+## 7. Explicit supersession ledger
+
+The following prior Round-2 recommendations are retained for provenance but no longer control the manuscript:
+
+- `01` A1 and `04` §3 title recommendations → superseded by the provisional frozen title in §1.
+- Kármán → steady as the implied physical progression → superseded by historical steady → Kármán, with a later steady-context return.
+- Vortex/Erase primarily in supplement or only an early gallery → superseded by a meaningful late cross-case extension section using case-specific metrics.
+- `04` §12 evidence-brief/card-first Round 3 and deferral of rough drafting to Round 4 → superseded by author-readable section/module drafts/dossiers approaching JFM prose in Round 3, guided by chapter-level JFM style study.
+- Any generic CCD mode label implying response/causality → superseded by `action-correlated correction-field modes`, with descriptive/noncausal status explicit.
+
+Unchanged recommendations include compact CFD validation, Kármán Re100 analytical dominance, Legacy/V5 separation, L2 primacy with DTW complementarity, case-specific metric semantics, bounded `control skeleton` usage, positive-SR Kármán/cloak scope, negative Illusion SR, descriptive CCD, claim-free OID and qualitative-only experiment.

docs/JFM_WYQ/ROUND2_TOP_DOWN/06_SECTION_LEVEL_JFM_STYLE_GUIDE.md

@@ -0,0 +1,350 @@
+# Round 3 section-level JFM writing-style guide
+
+> **Role.** Author-facing style and section-architecture guide for Round 3 evidence briefs. It is not manuscript prose or a claim authority. Recommendations follow the supplied final intent if it differs from a concurrently revised `04_TOP_LEVEL_SYNTHESIS.md`.
+>
+> **Final intent used here.** Working title *Interpreting learned hydrodynamic cloaking and illusion in a fluidic pinball*; capability Fig. 1; compact contracts/CFD; historical physics logic steady → Kármán, implemented in two passes: a short steady discovery/result seed in the early capability progression, then Kármán as the deep case with SR, mean/dynamic decomposition and CCD, followed by cross-case extensions and a later return to full steady context/theory; qualitative open-loop experiment; conclusion.
+
+## 1. Evidence base and status
+
+`Full-PDF` means paper-level full-text reading through the existing DynamisLab Undermind workspace, not equation-by-equation verification or artifact audit. `Full-report` means a repository literature report/dossier was read. `Snippet` means abstract, metadata or limited text only.
+
+| Exact identity | DOI | Status | Observed section-level lesson |
+|---|---|---|---|
+| Deng, Noack, Morzyński & Pastur, “Low-order model for successive bifurcations of the fluidic pinball”, *JFM* 884, A37 (2020) | `10.1017/jfm.2019.959` | full-PDF | Section order mirrors physical chronology; a visible failure motivates refinement; final phenomenogram synthesizes rather than inventories. |
+| Cornejo Maceda, Li, Lusseyran, Morzyński & Noack, “Stabilization of the fluidic pinball with gradient-enriched machine learning control”, *JFM* 917, A42 (2021) | `10.1017/jfm.2021.301` | full-PDF | Baseline → simple steady forcing → asymmetric forcing → feedback; performance and physical interpretation remain coupled. |
+| Rabault, Kuchta, Jensen, Réglade & Cerardi, “Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control”, *JFM* 865, 281–302 (2019) | `10.1017/jfm.2019.62` | full-PDF | Short entry to the gap; capability precedes wake interpretation; limitations appear beside findings. |
+| Paris, Beneddine & Dandois, “Robust flow control and optimal sensor placement using deep reinforcement learning”, *JFM* 913, A25 (2021) | `10.1017/jfm.2020.1170` | full-PDF | Reference-case efficacy precedes mechanism, robustness and sensor extensions; strong claims follow dedicated tests. |
+| Li & Zhang, “Reinforcement-learning-based control of confined cylinder wakes with stability analyses”, *JFM* 932, A44 (2022) | `10.1017/jfm.2021.1045` | full-PDF | Physics analysis earns early placement because it later changes reward and probe logic. |
+| Xia, Zhang, Kerrigan & Rigas, “Active flow control for bluff body drag reduction using reinforcement learning with partial measurements”, *JFM* 981, A17 (2024) | `10.1017/jfm.2024.69` | full-PDF | Baseline success → partial-observation failure → history-based repair → parametric limits. |
+| Wang, Yan, Hu, Chen, Rabault & Noack, “Dynamic feature-based deep reinforcement learning for flow control of circular cylinder with sparse surface pressure sensing”, *JFM* 988, A4 (2024) | `10.1017/jfm.2024.333` | full-PDF | Information gap → feature method → baseline capability → harder cases → physical interpretation. |
+| Loiseau, Noack & Brunton, “Sparse reduced-order modelling: sensor-based dynamics to full-state estimation”, *JFM* 844, 459–490 (2018) | `10.1017/jfm.2018.147` | full-PDF | Black-box sensor dynamics precede full-state reconstruction and grey-box interpretation; one schematic carries the method logic. |
+| Gautier, Aider, Duriez, Noack, Segond & Abel, “Closed-loop separation control using machine learning”, *JFM* 770, 442–457 (2015) | `10.1017/jfm.2015.95` | full-PDF in this pass; Round-1 card was snippet/preprint-level | Concise plant → objective → explicit law → temporal/spectral interpretation → benchmark/envelope. |
+| Mao, Blackburn & Sherwin, “Nonlinear optimal suppression of vortex shedding from a circular cylinder”, *JFM* 775, 241–265 (2015) | `10.1017/jfm.2015.304` | full-PDF | Numerical integrity precedes interpretation; global performance precedes spatial/stability explanation; realizability caveat remains visible. |
+| Round-1 Introduction, SR, CCD/OID, steady/theory, experiment/metrics dossiers | repository dossiers | full-report | Exact bibliography, evidence/metric boundaries, negative routes and experiment limits. |
+| Round-2 narrative, JFM benchmark, stress test and top-level synthesis | repository reports | full-report | Project-specific order, space/claim ceilings, Legacy/V5 and estimand boundaries. |
+
+**Coverage ledger.** Read: all four current Round-2 top-down reports; the relevant five Round-1 dossiers named above; ten exact JFM papers. Mapped: remaining Round-1 dossiers and existing Undermind `JFM/` and `JFM References/` indexes. Not covered: exhaustive sentence coding of all JFM papers, full facility-literature bibliography and unrelated papers; not required for this bounded task. External-inaccessible: none essential. Existing literature establishes style precedent, never internal results.
+
+## 2. Cross-paper observations and manuscript recommendation
+
+### Observed style
+
+1. A section is an argument: question/baseline → comparison → evidence → physical reading → caveat → next question.
+2. Results normally establish a reference case before extensions. Scalar/time evidence comes before or beside fields, and interpretation follows the declared comparator.
+3. Equations define objects used later; figures carry empirical burden; prose explains how the comparison changes the inference.
+4. Discussion is distributed. Papers interpret a result and state its local limitation before moving on.
+5. Strong words such as robustness, stability or optimality follow tests designed for those claims. Attractive fields alone do not license them.
+6. Bridges transform one result into the next scientific question; they do not merely announce a section number.
+
+### Recommendation
+
+Use the local rhythm above at a lower claim register. Treat **control skeleton** as the interpretation assembled after SR and field evidence, never as an initial premise. Keep terminology fixed:
+
+- concept: **hydrodynamic cloaking / hydrodynamic illusion**;
+- execution: **wake-signature matching**;
+- evaluation: **field matching**, L2 primary and DTW complementary;
+- interpretation: **control skeleton**;
+- dynamic field result: **CCD** and **action-correlated correction-field modes**.
+
+Do not replace the last phrase with causal-response, control-authority, importance or uniquely-CCD language.
+
+## 3. Section-level writing contracts
+
+Paragraph counts below are functional targets, not rigid formatting. Each subsection distinguishes the sample’s observed practice from the recommendation.
+
+### 3.1 Introduction
+
+**Observed.** Closest JFM papers funnel from physical problem to precise difficulty, nearest methods, gap and present contribution. Citations cluster in context/gap paragraphs; equations and detailed local results are absent. Deng uses history only to prepare the model question; Rabault and Paris reach the control gap quickly; Loiseau uses taxonomy because it organizes the contribution.
+
+**Recommended five-paragraph sequence.**
+
+1. **Concept.** Define cloak as approach to a declared background and illusion as approach to a declared non-zero target in observer/field space. Give one compact linear-cloak genealogy and end with non-equivalence to finite-Re Navier–Stokes wakes.
+2. **Physical difficulty.** Separation, shedding and the multi-cylinder pinball make wake-signature matching different from scalar drag reduction. Cite pinball dynamics and active hydrodynamic precedents.
+3. **Capability and gap.** Sparse-sensor DRL can discover nonlinear control, but reward/signature success does not establish field matching or a physical law. Cite nearest DRL/partial-observation work.
+4. **Missing evidence chain.** Motivate executable policy reduction, independent field evaluation, mean/dynamic separation and bounded steady context. Cite SR/low-order methods only for their roles.
+5. **Question, contribution, scope.** Name capability breadth, sole deep Kármán case, qualitative experiment and principal non-claims. Promise capability first, then contracts and interpretation—not a package list.
+
+**Placement, tense, strength.** Present tense for definitions/physics; past or present perfect for literature; present for paper purpose. No project number in paragraphs 1–3. Refer prospectively to Fig. 1 near the end without narrating panels. Use `can be interpreted as`, `under the declared contract`; reserve `demonstrates` for named-condition capability. Do not claim priority without a separate audit.
+
+**Bridge.** Convert the gap into the definitions and estimands needed before the capability progression can be compared.
+
+**Anti-patterns.** Cloak-history catalogue; one paragraph per paper; “AI is revolutionizing science”; contributions as `DRL/SR/CCD/theory/experiment`; citations appended without argumentative grouping; announcing persistent rotation as cause before evidence.
+
+### 3.2 Problem, method and compact CFD contracts
+
+**Observed.** JFM control papers establish geometry/BC, objective, sensing/action and qualification before dependent claims. Definitions use present tense; validation outcomes use past. Equations cluster around reusable contracts.
+
+**Recommended six functional paragraphs.**
+
+1. **Plant:** geometry, body order, coordinates, BCs and Reynolds convention; declare Legacy/V5 separation.
+2. **Actuation:** wall-velocity sign, action order and normalization; use a sign schematic.
+3. **Observation/objective:** forces and six velocity channels; define wake-signature matching and reward roles.
+4. **Evaluation:** field L2 is primary independent evaluation; DTW is complementary alignment. Separate periodic phase, transient event and steady profile semantics.
+5. **Evidence roles:** target/zero/controlled/constant and which chain supports capability versus interpretation.
+6. **Numerical qualification:** selected benchmark checks, unresolved mismatch, configuration scope and supplement boundary.
+
+**Placement, tense, strength.** Put geometry/sensor/role schematic early. Display only equations reused later: Reynolds/action sign, objective, L2 and perhaps compact DTW notation. Cite methods/benchmarks at the defining sentence. CFD qualification supports selected observables, not control physics. Present for contracts, past for checks.
+
+**Bridge.** With contracts fixed, introduce the historical capability progression—steady seed first, then Kármán and bounded extensions—without pooling its metrics.
+
+**Anti-patterns.** Configuration/hash dump; standard-PPO pseudo-code; detached validation chapter; equations never reused; “all parameters are in table”; repeated definitions; silent Re/action shifts.
+
+### 3.3 Capability/results opening and Fig. 1
+
+**Observed.** Rabault, Paris, Xia and Wang establish a reference case and comparator before extending scope. The first results figure combines a quantitative headline and physical view. Variation/failure enters where it changes the headline.
+
+**Recommended four-to-five-paragraph sequence.**
+
+1. **Viewing contract.** State Fig. 1’s question: can learned actuation approach declared targets across distinct tasks? Name comparator and metric roles without repeating Methods.
+2. **Short steady discovery seed.** Introduce the historically first steady-rotation result in one compact comparison. Establish only discovery chronology and visible capability; defer its separate contract, theory and physical synthesis to the later steady return.
+3. **Headline Kármán.** Read target, zero and controlled panels as one comparison; state L2 field matching first and DTW as complementary signature alignment. Explain why Kármán becomes the deep case.
+4. **Kármán → vortex and cloak → illusion.** Mark the changed temporal or target contract and give only bounded, case-specific capability here; full cross-case analysis returns after the Kármán block.
+5. **Scope/selection.** Show one degradation/failure and n/seed status; distinguish the early progression preview from later analytical sections.
+
+**Placement, tense, strength.** Place Fig. 1 at paragraph 1. Every number needs case, chain, comparator and metric semantics. Past for training/evaluation; present for figure reading. Prefer `reduces the declared field error`; breadth is demonstration, not generality.
+
+**Bridge.** With the early progression established under already declared contracts, select Kármán for deep performance and interpretation.
+
+**Anti-patterns.** Caption-like panel narration; gallery; one score across periodic/transient/steady; DTW as field equivalence; hidden failures; Legacy/V5 pooling; illusion image as target-tracking proof.
+
+### 3.4 Kármán performance
+
+**Observed.** Reference-case sections start with baseline and headline metric, move from scalar/time traces to fields and only then interpretation. Paris and Mao pair global performance with spatial evidence; comparator limits appear locally.
+
+**Recommended four paragraphs.**
+
+1. **Exact test:** chain, case/Re convention, roles/comparator, window and seed/selection status.
+2. **Signature result:** reward components and DTW as optimized/reward-adjacent evidence, not independent validation.
+3. **Field result:** L2 mean and phase evidence separately; use agreement/disagreement with DTW as a finding.
+4. **Boundary:** seed dependence, phase limitation or cross-Re degradation; conclude only named-contract improvement.
+
+**Placement, tense, strength.** Lead with field result, not a training diary. Training/selection goes in compact inset/table. Past for computation, present for figure. Use `was accompanied by`, not `therefore produced`; `field matching improved`, not `wake restored`.
+
+**Bridge.** Field matching shows the result is not solely a reward proxy; it does not reveal action-law complexity.
+
+**Anti-patterns.** Chronological training log; all seeds as narrative protagonists; best checkpoint conflated with retained policy; low DTW as full-field identity; no interpretive handoff.
+
+### 3.5 SR interpretation
+
+**Observed.** Gautier isolates an explicit law, then its dynamic behaviour and benchmark. Cornejo Maceda compares simpler/richer spaces and tests the averaged law. Executable comparison—not formula complexity—is primary.
+
+**Recommended five paragraphs.**
+
+1. **Question/estimand:** SR is an observation-to-action surrogate in the Legacy plant, not a field-causal model.
+2. **Recovery contract:** causal pairing, symmetry mapping, formula identity and closed-loop CFD arbitration; library/search detail goes to supplement.
+3. **Law:** display once and immediately translate order, sign and persistent rear counter-rotation.
+4. **Closed loop:** PPO/SR/zero under the same window; offline fit is diagnostic, closed-loop wake performance is arbiter.
+5. **Term tests:** persistent rear term is largest tested; feedback is smaller but non-zero. State fixed-window, parent-relative, non-necessity limits.
+
+**Placement, tense, strength.** Formula before performance figure. Failed gates/weaker deletions sit beside successes. Past for fitting/testing; present for formula meaning. Use `surrogate`, `largest tested contribution`; avoid `mechanism`, `necessary`, `sufficient`, `unique`, `governing law`.
+
+**Bridge.** Term ranking does not show how much field-level improvement the persistent component accounts for.
+
+**Anti-patterns.** Stage/package handoff; coefficient inventory; offline RMSE as control success; hidden imposed symmetry; deletion called causal ablation; positive Illusion SR import.
+
+### 3.6 Mean/dynamic decomposition and CCD
+
+**Observed.** Loiseau moves from predictive success to spatial reconstruction; Deng uses model failure to justify refinement. The estimand and comparison precede modes; prose states the new inference each representation permits.
+
+**Recommended six paragraphs.**
+
+1. **Question/four roles:** target, zero, constant and DRL means on exact ROI/mask/window.
+2. **Mean comparison:** fields and scalar errors before percentages; then additive error-difference ledger.
+3. **Mean limit:** `accounts for measured error reduction` is an estimand statement, not causality, energy share or momentum restoration.
+4. **Residual definition:** separately centred cycle-template residual; constant/DRL cycles are not paired counterfactuals.
+5. **CCD result:** rank/subspace stability, action coordinates and phase reconstructions as **action-correlated correction-field modes**; no response time or causal direction.
+6. **POD comparison/boundary:** near-equivalent subspaces mean CCD adds action-coordinate association, not superior/unique field modes; single lineage remains explicit.
+
+**Placement, tense, strength.** Mean figure precedes residual-mode figure. Put error identity, residual and (only if interpretively needed) CCD operator immediately before their figures. Past for construction, present for figure. Use `separates`, `organizes`, `is action-correlated`, `accounts for measured difference`; avoid `explains`, `drives`, `responds`, `energy captured` for CCD strengths, and `unique CCD modes`.
+
+**Bridge.** Kármán supports a persistent mean contribution plus smaller action-correlated residual; return later to what this bounded organization does and does not connect.
+
+**Anti-patterns.** Linear algebra first; mode gallery; singular strengths as explained variance; 93.42% without errors/ROI/window; SR law equated to CCD left functions; shared parent called independent triangulation.
+
+### 3.7 Cross-case extensions
+
+**Observed.** Xia’s failure-to-fix chronology and Paris/Wang’s harder-case sections change one difficulty at a time, restate the needed comparator and qualify breadth according to dedicated tests.
+
+**Recommended three modules.**
+
+1. **Kármán → vortex:** phase cycle becomes event/relative-offset semantics. State what remains fixed and use one case-specific result; never average event snapshots into phase metrics.
+2. **Cloak → illusion:** define non-zero target matching, show bounded PPO capability and explicitly separate negative Illusion SR.
+3. **Illusion → erase:** change target/reference again; use mean semantics only with exact roles, otherwise mark visual demonstration or omit. End with failure envelope.
+
+**Placement, tense, strength.** One question, comparator, principal result and limit per module. Past for demonstrations; conditional for transfer. Use `extends the objective`, not `generalizes`.
+
+**Bridge.** Execution accommodates changed temporal/target contracts, but the Kármán control-skeleton interpretation does not transfer with them.
+
+**Anti-patterns.** One prose row per case; common efficiency bar for incompatible metrics; n=1 called robustness; positive Illusion SR; Kármán SR/CCD used to interpret vortex/erase.
+
+### 3.8 Steady context, physical synthesis and theory limits (later return)
+
+**Observed.** Cornejo Maceda proceeds from simpler steady actuation to feedback; Deng mirrors physical chronology; Li puts stability first because it changes later choices. Theory earns early placement only when it supplies a baseline or constraint used later.
+
+**Two-pass recommendation.** The early capability section has already introduced steady discovery/result as a short narrative seed, preserving the historical physics logic `steady → Kármán`. This later full section returns only after Kármán performance, SR, mean/dynamic+CCD analysis and cross-case extensions. Its job is retrospective contextualization of the near-constant core, not chronological placement before the Kármán analysis.
+
+**Recommended four-to-five-paragraph sequence.**
+
+1. **Return and scope.** Recall the early steady seed in one sentence, then state why the full analysis returns here: to contextualize the near-constant Kármán core without treating the separate steady chain as proof of periodic-controller mechanism.
+2. **Contract/result.** Separate V5 free-slip `Re_D=50`, accepted sign, spin ratio, comparator and settling-qualified low-deficit endpoint.
+3. **Physical description.** Gap-facing motion/base-bleed or gap-jet-family compatibility as observation/hypothesis; not boat-tail or closed momentum pathway.
+4. **Theory constraint.** Give only the shortest outer-reference equation/sign result. Place circulation-sign conflict and rotating-no-slip limitation beside the attractive comparison.
+5. **Synthesis export.** Relate the returned steady evidence retrospectively to the already established Kármán near-constant core, transferring only recurrence/magnitude context—not metric equality, plant equivalence, stability or mechanism.
+
+**Placement, tense, strength.** Contract/sign schematic precedes field interpretation. Present for mathematics; past for computed endpoint; conditional for physical chain. Use `compatible with`, `provides context`, `constrains`; avoid `explains`, `validates the skeleton`, `optimal`, `stable`, `restores momentum`.
+
+**Bridge.** The later steady return closes the historical loop: it supplies a compatible endpoint and sign-constrained context for the already established Kármán near-constant core, then hands the bounded synthesis to the qualitative experiment and conclusion.
+
+**Anti-patterns.** Detached theory chapter; unvalidated MFS spectacle; failed sign route delayed to Discussion; V5 steady metric transferred into Legacy; universal skeleton pre-announced.
+
+### 3.9 Qualitative open-loop experiment
+
+**Observed.** Experimental JFM papers define facility, observable and actuation before results (Gautier); numerical papers place realizability caveats near constrained findings (Mao). A qualitative demo cannot inherit quantitative CFD wording.
+
+**Recommended two-to-three paragraphs.**
+
+1. **Purpose/apparatus boundary:** establish deployability and open-loop operation; give conditions needed to interpret media.
+2. **Observation:** neutral streakline/shedding description tied to raw-media panel and trial identity.
+3. **Limit/next test:** no synchronized calibrated PIV/force validation; low-speed/background issue; requirements for quantitative validation.
+
+**Placement, tense, strength.** Caption carries provenance, scale/frame/condition. Past tense; use `qualitative`, `open-loop`, `feasibility`. No efficacy percentage, field match, force reduction or experimental validation from dye.
+
+**Bridge.** Experiment establishes qualitative deployability; interpretation and quantitative evidence remain chain-specific computational results.
+
+**Anti-patterns.** Build diary/hardware promotion; image aesthetics as evidence; unregistered CFD overlay; thermal convection declared sole cause; open-loop called policy validation.
+
+### 3.10 Discussion and conclusion
+
+**Observed.** Discussion is integrated; the final section synthesizes rather than replays. Deng uses a phenomenogram, Cornejo Maceda closes across a hierarchy, Li separates physics/control findings. Conclusions add no new result and shift to conditional future work.
+
+**Recommended discussion sequence.**
+
+1. Assemble steady context, SR term ranking and mean/dynamic evidence while naming different chains/estimands: `compatible observations`, not independent proof.
+2. Define dynamic component: smaller, phase-organized, represented by action-correlated correction-field modes; not causal response or unique CCD subspace.
+3. Define envelope: periodicity removal, objective extension, seeds/cross-Re degradation, negative Illusion SR, failed theory and qualitative experiment.
+4. Name only decisive tests that change claim ceiling: matched interventions, independent parents, same-plant field comparison, closed budgets, synchronized experiment.
+
+**Recommended conclusion sequence.**
+
+1. Restate question and named scope, not methods.
+2. Answer in evidence order: capability → L2 field matching → persistent SR/mean contribution → smaller action-correlated residual.
+3. State strongest boundary and one decisive future test; return to cloak/illusion as declared matching, not invisibility.
+
+**Placement, tense, strength.** No new figures, equations, citations or numbers. Past/present perfect for work; present for bounded conclusion; conditional/future for outlook. Use `supports a control-skeleton interpretation under the declared Kármán contract`. Avoid `proves`, `mechanism`, `universal`, `robust`, `optimal`, `experimentally validated` and generic “opens new avenues”.
+
+**Anti-patterns.** Abstract replay; bullet per package; limitations first introduced here; unsupported applications; every caveat converted into future work; three non-equivalent analyses said to validate one mechanism.
+
+## 4. Bridges and integrated discussion
+
+Use bridges as logical transforms:
+
+- Introduction → contracts: convert the gap into definitions and estimands needed for every later comparison.
+- Contracts → capability/progression: ask what control can do under named, non-pooled contracts.
+- Within capability: use a short steady discovery/result seed first, then Kármán, vortex and illusion; this preserves historical `steady → Kármán` logic without placing the full steady analysis early.
+- Capability → Kármán performance: select Kármán as the sole deep case and move from visual result to independent field matching.
+- Performance → SR: success becomes an action-complexity question.
+- SR → mean/dynamic+CCD: term ranking becomes field-accounting and residual-organization questions.
+- Kármán deep block → cross-case extensions: change one task dimension without transferring the Kármán interpretation.
+- Extensions → later steady/theory return: revisit the early steady seed to contextualize the now-established near-constant core and expose theory limits.
+- Steady synthesis → experiment: move from computational interpretation to bounded physical deployability.
+- Experiment → conclusion: separate feasibility from quantitative/causal evidence.
+
+Within each result subsection use:
+
+`observation → comparator → physical reading → alternative/limit → consequence`.
+
+Cite literature at the physical reading/comparison, not decoratively after a number. Put the most dangerous limitation beside the claim it limits. Do not rely on final Discussion to repair upstream overstatement.
+
+## 5. Tense, strength and compression
+
+- **Present:** definitions/equations, what a figure shows, bounded interpretation.
+- **Past:** training, simulations, measurements and obtained outcomes.
+- **Present perfect:** development of literature, sparingly.
+- **Conditional/future:** hypotheses, transfer, experiment and decisive tests.
+
+**Claim ladder**
+
+- Green: `defines`, `shows`, `reduces the declared error`, `passes the stated gate`, `is lower than`.
+- Amber: `is consistent with`, `organizes`, `accounts for measured error difference`, `is action-correlated`, `provides context`, `motivates a control-skeleton interpretation`.
+- Red without new evidence: `causes`, `mechanism`, `necessary/sufficient`, `restores momentum`, `stable`, `optimal`, `robust/general`, `independently validates`, `experimentally validates`.
+
+**Compression/anti-AI rules**
+
+1. One paragraph, one argumentative job; first sentence states it.
+2. One principal figure claim per paragraph; secondary panel details stay in caption.
+3. A number must change the inference or move to table/supplement.
+4. Define each acronym once; do not rename CCD modes for variety.
+5. Use concrete subjects (`registered field error`, `fixed-law deletion`) rather than `the results`.
+6. Delete throat-clearing: `It is worth noting`, `Interestingly`, `As can clearly be seen`, `In order to`.
+7. Avoid repeated three-item lists, identical paragraph templates, `not only…but also`, and conclusions that restate headings.
+8. Delete artifact inventories, package histories and status-ledger prose from narrative; retain them in briefs/tables/supplement.
+
+## 6. Practical Round-3 section-agent prompt
+
+```text
+Prepare the Round-3 evidence brief for [SECTION NAME] of
+“Interpreting learned hydrodynamic cloaking and illusion in a fluidic pinball”.
+Do not draft manuscript prose.
+
+Section question: [ONE PHYSICAL QUESTION]
+Answer ceiling: [ONE GREEN/AMBER SENTENCE]
+Imports from earlier sections: [DEFINITIONS/EVIDENCE]
+Exports to later sections: [ESTABLISHED FACT/NEXT QUESTION]
+Word/figure/equation budget: [BUDGET]
+
+Use repository authority and Round-1/Round-2 dossiers. Keep chain, case/seed/role,
+Reynolds convention, comparator, metric/estimand, ROI/mask/window and status explicit.
+External literature supports narrative/method precedent only, never internal results.
+
+Terminology:
+- concept: hydrodynamic cloaking / hydrodynamic illusion;
+- execution: wake-signature matching;
+- evaluation: field matching, L2 primary and DTW complementary;
+- interpretation: control skeleton;
+- dynamic fields: CCD and “action-correlated correction-field modes”.
+
+Produce:
+1. Section question and bounded one-sentence answer.
+2. Paragraph-function map. For each paragraph give opening move, claim/definition,
+   exact evidence/comparator, figure/equation/citation placement, local limitation,
+   and bridge.
+3. Green / amber / prohibited-red claim ladder.
+4. Figure/table/equation contracts: what each proves and does not prove.
+5. Literature-use cards: exact identity, DOI when available,
+   full-PDF/full-report/snippet status; separate observed style from recommendation.
+6. Cross-section imports/exports and final bridge function (not polished prose).
+7. Anti-pattern audit: index/report/package-handoff language, metric pooling,
+   method silos, caption narration, generic AI prose, delayed limits, overclaiming.
+8. Only genuine drafting blockers; no general wish list.
+
+Required rhetorical sequence:
+[PASTE THE RELEVANT SECTION SEQUENCE FROM THIS GUIDE]
+
+Frozen ordering check:
+- The early capability/progression module introduces steady discovery/result first as a
+  short narrative seed, then Kármán, vortex and illusion capability.
+- The full steady context/physical synthesis/theory-limits module is later: after the
+  Kármán deep-analysis and cross-case extension modules.
+- Preserve the historical physics logic steady → Kármán, but do not describe the full
+  steady/theory analysis as physically or textually placed before Kármán.
+
+Do not invent quotations, page counts, citations, numbers or artifacts. Label sourced
+fact, bounded observation, interpretation, historical/negative result and open question.
+```
+
+## 7. Strongest operational findings
+
+1. Draft section interfaces—question, answer ceiling, imports, exports and bridge—before sentences.
+2. Start results with a comparator, not a method.
+3. Use steady in two passes: an early discovery/result seed preserves historical `steady → Kármán` logic; the full steady/theory analysis returns only after Kármán depth and cross-case extensions.
+4. Keep capability broad but interpretation asymmetric: only Kármán receives SR/mean-dynamic/CCD depth.
+5. Make L2 carry field claims; DTW remains complementary signature evidence, never field equivalence, delay or causality.
+6. Put one visible caveat beside every major result.
+7. Use exactly `action-correlated correction-field modes`; do not imply response, authority or CCD superiority.
+8. Use failed routes (theory sign conflict, mean/phase mismatch, negative Illusion SR, experiment limits) to define the claim boundary and next test.
+9. End sections with scientific questions, not “Section X presents…”.
+10. Run an anti-index pass before drafting: narrative must not read like artifact inventory, report, package handoff or generic AI synthesis.
+
+## 8. Provenance
+
+Undermind orientation preceded workspace access. The ten named papers were read at full-PDF level for rhetorical patterns; relevant Round-1/Round-2 repository dossiers were read as full reports. No new deep search, CFD, training, experiment analysis or Undermind mutation was performed. Existing files and immutable evidence packages were not modified.

docs/JFM_WYQ/ROUND2_TOP_DOWN/README.md

@@ -0,0 +1,36 @@
+# Round 2 — Top-Down JFM Design
+
+## 目标
+
+Round 2 基于 Round 1 的全部 dossiers 与 coverage audit，进行自上而下的 JFM manuscript 设计。目标是明确：
+
+- manuscript title；
+- physics narrative；
+- section hierarchy；
+- approximate text / figure allocation；
+- main-text / supplement boundary；
+- evidence dependencies。
+
+本轮不是写 manuscript prose，而是建立可审计的论文结构、物理叙事和证据依赖框架。
+
+## 预定输出
+
+1. `01_PHYSICS_NARRATIVE_AND_OUTLINES.md`
+2. `02_JFM_STRUCTURE_AND_SPACE_BENCHMARK.md`
+3. `03_EVIDENCE_STRESS_TEST.md`
+4. `04_TOP_LEVEL_SYNTHESIS.md`
+5. `05_REVISED_STRUCTURE_FREEZE.md` — author-ratified revision and Round-3 structure handoff
+6. `06_SECTION_LEVEL_JFM_STYLE_GUIDE.md` — controlling chapter-level JFM style input for Round 3
+
+## 工作约束
+
+- `04_TOP_LEVEL_SYNTHESIS.md` 必须回读 Round 1 的全部 dossiers，以及关键 package authority；不得只依据摘要或二手汇总。
+- 保留并明确区分 Legacy / V5。
+- 保留 case-specific metrics，不将个案指标无依据地泛化为普适结论。
+- 明确 claim boundaries：区分已被证据支持的主张、条件性主张、待验证主张与不应提出的主张。
+- 每一项关键结构或叙事选择都应能追溯到相应的 evidence dependencies。
+
+
+## Author-ratified freeze (2026-08-10)
+
+Round 3 is jointly controlled by `05_REVISED_STRUCTURE_FREEZE.md` for manuscript structure and `06_SECTION_LEVEL_JFM_STYLE_GUIDE.md` for chapter-level JFM style, read together with the evidence dependencies in `04_TOP_LEVEL_SYNTHESIS.md`. It records the provisional frozen title, terminology and metric hierarchy, corrected steady → Kármán → vortex → illusion/erase → experiment progression, revised section/space allocation, and the section/module-based Round-3 handoff. Where it conflicts with recommendations in `01`–`04`, the older recommendation is retained as provenance but marked superseded; evidence status and repository authority are unchanged.

docs/JFM_WYQ/ROUND3_BOTTOM_UP/00_MASTER_AGENT_PROMPT.md

@@ -0,0 +1,29 @@
+# Round 3 Master Agent Prompt
+
+Use this reusable prompt for one assigned module. Replace every bracketed field before dispatch.
+
+## Prompt template
+
+You are the Round-3 section agent for **[MODULE ID — MODULE FILENAME]** in *Interpreting learned hydrodynamic cloaking and illusion in a fluidic pinball*.
+
+**Assigned scope:** [NON-OVERLAPPING SECTION QUESTION]. **Bounded answer ceiling:** [ONE GREEN/AMBER SENTENCE]. **Paragraph target and space:** [TARGET]. **Imports:** [DEFINITIONS, EVIDENCE, BRIDGES]. **Exports:** [DEFINITIONS, RESULT, NEXT QUESTION]. **Relevant figures/equations/tables:** [LIST].
+
+Read before drafting: the Round-2 freeze and style guide; relevant Round-1 dossiers; and relevant current package authority documents/manifests. External literature supplies narrative/method precedent only, never internal results. Do not infer facts from memory. Preserve Legacy/V5, solver, geometry, Re convention, case/seed/role, comparator, metric/estimand, ROI/mask, window and artifact boundaries.
+
+Use fixed terminology: concept `hydrodynamic cloaking/illusion`; execution `wake-signature matching`; evaluation `field matching`, spatial L2 primary and DTW complementary/case-specific; interpretation `control skeleton`; CCD `action-correlated correction-field modes`, descriptive/noncausal. Distinguish pipeline `Level-2/Level-3` and `Stage-3 short/standard gate` from spatial L2; never write `L2 gate` for validation level. Preserve positive SR Kármán/cloak only, negative Illusion SR, claim-free OID, qualitative experiment and compact CFD.
+
+## Deliver exactly
+
+1. Section question and bounded answer.
+2. **Layer 1 — narrative:** prose-like paragraph drafts/maps with a clear argumentative job, short source tags (for example `[R1-SR; authority]`), substantive transitions, and local caveats. Do not insert giant inline `Evidence contract:` blocks, package-path inventories or bold template labels into the narrative.
+3. **Layer 2 — local provenance:** immediately after the narrative, provide compact paragraph notes or a claim-to-evidence ledger. For each quantitative/physical claim record exact source/artifact, chain, solver/geometry, case/seed/role, Re convention, comparator, metric/estimand, ROI/window, status, and whether `artifact-audit deferred`.
+4. Figure, equation and table contracts: role, evidence/artifact, caption-level limitation, and what each does not establish.
+5. Claim ladder: supported/green, bounded/amber, prohibited/red; imports, exports, final bridge, genuine blockers, and concise coverage ledger (`read`, `mapped`, `not-covered`, `external-inaccessible`, with reasons and evidence status).
+6. Cross-module near-collision check: for numerically similar results, compare role acquisition, role set, ROI/domain, mask, weighting/normalization and window; classify them as `same estimand`, `related diagnostic` or `different estimand`. Prohibit arithmetic combination and replication/validation/agreement/triangulation wording unless exact authority licenses it.
+7. Anti-index/anti-AI audit, including metric pooling, citation invention, silent chain transfer, redundancy, source-path dominance, paragraph/sentence overload, and repeated mega-paragraphs.
+
+Authority-led hidden/external artifact deferral is not automatically a blocker. If authority docs/manifests explicitly bind the artifact and exact claim/number without conflict, record `artifact-audit deferred` and continue. Block only when the missing leaf/raw artifact is needed to resolve a conflicting authority, identify an exact role/comparator/window, substantiate a proposed number/panel, or answer the section question. Do not demand universal hash recomputation.
+
+Respect scope interfaces: contextual facts from another module may be one bridge sentence or export, not a duplicate result block. In particular, SR must not absorb full steady results; field/CCD must not reproduce SR derivation; contracts must define contracts, not become results.
+
+Edit **exactly** `[OUTPUT FILE]` and no other file. Do not modify Round-1/2, source code, archives or evidence packages. Report inspected sources and deferrals. This is a first pass: do not label it manuscript-ready; await parent audit and targeted revision.

docs/JFM_WYQ/ROUND3_BOTTOM_UP/00_MODULE_INTERFACE_LEDGER.md

@@ -0,0 +1,25 @@
+# Round 3 Module Interface Ledger
+
+Coordinator-owned dispatch order. Each module writes only its assigned file and imports/exports definitions explicitly. Statuses are `planned`, `first-pass`, `revise`, `second-audit`, `ready`, or `blocked`.
+
+| Order | Module / output filename | Primary question and dependency | Export to | Status | Coordinator audit note |
+|---:|---|---|---|---|---|
+| 1 | `03_CONTRACTS_METRICS.md` | What are the plant, chain, action, role and case-specific evaluation contracts? | all modules | `ready` | Second-audit PASS; chain, role, metric and Level/Stage interfaces are reconciled. |
+| 2 | `04_KARMAN_PERFORMANCE.md` | Under the Legacy Kármán contract, what does independent field matching show? | 05, 06, synthesis | `ready` | Second-audit PASS; DTW orientation and M04/M06 different-estimand boundary are reconciled. |
+| 3 | `05_KARMAN_SR_LAW.md` | What executable SR structure is supported, and what do bounded deletions test? | 06, 08, conclusion | `ready` | Section text ready; standardized PPO/SR/Zero panel remains optional/dependency-gated and must be dropped unless exact role artifacts are bound. |
+| 4 | `06_KARMAN_FIELD_CCD.md` | What do mean/dynamic estimands and CCD action-correlated correction-field modes organize? | 08, 10 | `ready` | Second-audit PASS; corrected four-role estimand and M04 non-equivalence boundary are reconciled. |
+| 5 | `07_CROSS_CASE_EXTENSIONS.md` | What changes across vortex, Illusion and Erase under distinct contracts? | 08, 10 | `ready` | Second-audit PASS; ready for Wave C imports; Vortex event-L2 and Erase quantitative panels remain dependency-gated/blocked, and Illusion field L2 remains conditional. |
+| 6 | `08_STEADY_SYNTHESIS.md` | What bounded steady context can be related retrospectively to Kármán? | 09, 10 | `ready` | Second-audit PASS; text ready for Wave C imports; optional steady/MFS/evidence-map panels are dependency-gated and nonblocking. |
+| 7 | `01_INTRODUCTION.md` | What concept, gap and bounded contribution frame the paper? | 02, all | `ready` | Second-audit PASS; five-paragraph JFM-facing introduction ready, three claim levels and asymmetric cross-case scope reconciled, exact literature identities/read levels retained, and Fig. 1 correctly bridged to the capability section. |
+| 8 | `02_CAPABILITY_PROGRESSION.md` | How does steady discovery → Kármán → vortex → illusion/erase establish capability breadth? | 04, 07 | `ready` | Second-audit PASS; bounded progression text ready, Kármán dominant, steady q_ctl/q_in contract corrected, Vortex/Illusion roles bounded; raw/redrawn Fig1 panels remain artifact-dependency-gated and optional Re400 marker is subordinate/drop-first. |
+| 9 | `09_EXPERIMENT_AUTHOR_INTERFACE.md` | What qualitative open-loop experiment interface is reserved for Frank? | 10 | `ready` | Second-audit PASS as author interface; ready for Frank/M10 imports, but final prose and any experiment figure remain blocked by the author-supply/raw-provenance gate. |
+| 10 | `10_LIMITATIONS_CONCLUSION.md` | What is the integrated claim ceiling and decisive next test? | final audit | `ready` | Second-audit PASS; three claim levels separated (named cross-task capability, Legacy Kármán-only independent field matching, bounded Legacy Kármán control skeleton); M09 imported; final experiment prose/panel remains author-provenance-gated. |
+| 11 | `11_FINAL_CONSISTENCY_AUDIT.md` | Are terminology, chains, metrics, bridges, provenance and redundancy consistent? | manuscript handoff | `ready` | Final consistency audit PASS for author manuscript synthesis; all Modules 01–10 are text-ready under the fixed claim ladder. Remaining REVISE items are assembly-stage deduplication/ownership edits; BLOCK items are optional panel/artifact or author-provenance gates and do not block bounded text. |
+
+## Coordinator-owned near-collision interface note
+
+Although the standardized PPO and corrected CCD mean errors are numerically close and share the same nominal x extent, they arise from different role acquisitions, retained windows, masks and quadrature/normalization contracts; they are separate estimands, not replication or validation, and are not combined arithmetically.
+
+## Interface rules
+
+The coordinator records imported definitions and exported claims after each audit. No export may exceed the source module's claim ceiling or cross Legacy/V5, periodic/transient/steady, or capability/interpretation boundaries without explicit authority. A `second-audit` may be conditionally passed, but `ready` requires imported interfaces to be reconciled. Optional figure dependencies do not block section text unless the panel or claim is retained; missing artifacts, exact roles or unresolved conflicts otherwise become blockers, not inferred content.

docs/JFM_WYQ/ROUND3_BOTTOM_UP/00_QUALITY_GATE.md

@@ -0,0 +1,29 @@
+# Round 3 Quality Gate
+
+The parent coordinator assigns one status per module: **PASS**, **REVISE**, or **BLOCK**. `PASS` means ready for a second audit, never final manuscript approval. Only after the second audit may the coordinator mark a module `ready`.
+
+## First-pass audit protocol
+
+Before revising a module, classify each defect as **systematic prompt defect** (also observed or likely in other modules) or **module-specific defect** (local evidence, scope or reasoning problem). Update the rule/master prompt/quality control first for systematic defects, then issue targeted revisions to affected modules. Do not conceal a control-plane defect as an agent failure.
+
+## PASS checks
+
+- [ ] Two-layer delivery: readable prose-like paragraph drafts/maps with short source tags first; compact local paragraph notes or claim-to-evidence ledger immediately afterward. Full provenance is local and traceable without giant inline contract blocks, package-path dumps, bold template labels or source/status language dominating the argument.
+- [ ] Paragraph cadence is readable: one argumentative job per paragraph, substantive bridges, reasonable sentence/paragraph load, and observation → comparator → physical reading → limit → consequence where applicable. Flag repeated mega-paragraphs or overloaded sentences; do not use rigid word-count automation.
+- [ ] Reads and names the controlling Round-2 freeze/style guide, relevant Round-1 dossiers and current package authorities; ledger distinguishes `read`, `mapped`, `not-covered`, `external-inaccessible` with reasons.
+- [ ] Every number and physical claim has exact provenance: authority/artifact, Legacy/V5 chain, solver/geometry, case/seed/role, Re convention, comparator, metric/estimand, ROI/window and status.
+- [ ] Authority-led hidden/external artifacts explicitly bound by conflict-free authority are marked `artifact-audit deferred`, not automatically blocked. BLOCK applies only when the missing artifact is necessary to resolve authority conflict, identify exact role/comparator/window, substantiate a proposed number/panel, or answer the section question.
+- [ ] Distinguishes pipeline `Level-2/Level-3` or `Stage-3 short/standard gate` from spatial L2 field-error metric; no validation level is called an `L2 gate`. Spatial L2 remains primary for field matching; DTW is complementary/case-specific, with no silent metric pooling or phase relabelling.
+- [ ] Enforces scope interfaces: bridge/context is brief; SR does not duplicate full steady results; field/CCD does not duplicate SR derivation; contracts do not become a results section.
+- [ ] Runs a near-collision check for numerically similar cross-module results: compare role acquisition, role set, ROI/domain, mask, weighting/normalization and window; classify `same estimand`, `related diagnostic` or `different estimand`. Prohibit arithmetic combination and replication/validation/agreement/triangulation wording unless exact authority licenses it.
+- [ ] Defines figure/equation/table roles and caption limits; states imports, exports, section bridge, conflicts, non-claims and genuine blockers.
+- [ ] Uses fixed terminology and claim ceiling: positive SR Kármán/cloak only; negative Illusion SR; descriptive/noncausal CCD action-correlated correction-field modes; claim-free OID; qualitative experiment.
+- [ ] Passes anti-index and anti-AI checks: no inventory/package handoff/card stack, generic prose, caption narration, citation/number/artifact invention, or delayed limitations.
+
+## REVISE
+
+Use when evidence is traceable but narrative cadence, layer separation, style, interfaces, terminology, redundancy, provenance, local caveats or exact metric semantics are incomplete. Return targeted paragraph-level prompts, not a general wish list. Re-audit after revision.
+
+## BLOCK
+
+Use when a controlling source was not read without a justified ledger entry, an authority conflict affects the answer, a required artifact/role is absent, a number lacks provenance, Legacy/V5 are merged, metrics are pooled, or the draft requires unsupported causality/mechanism/generalization. An authority-led artifact deferral alone is not a blocker when its claim and role are explicitly bound without conflict. Preserve the blocker; do not fill it from memory.

docs/JFM_WYQ/ROUND3_BOTTOM_UP/00_WAVE_A_COORDINATOR_AUDIT.md

@@ -0,0 +1,90 @@
+# Wave A independent coordinator audit
+
+## Scope, gate and overall disposition
+
+This audit read in full the revised Round-3 drafting rule, master prompt, quality gate and interface ledger; the Round-2 revised structure freeze and section-level style guide; and current revised Modules 03–06. Current authority was consulted only for two concrete interface questions: `wake_l2.py` and the corrected CCD contract/results for the near-colliding field estimands, and the frozen SR pipeline for July formula-parent versus August standardized-role windows. Evidence status is therefore `read`, not independently recomputed or universally artifact-audited. Working memory was used only as a scope warning and did not establish a result.
+
+**Gate verdicts:** Module 03 — **PASS**; Module 04 — **REVISE**; Module 05 — **PASS, conditionally ready after ledger reconciliation**; Module 06 — **REVISE**. No true scientific blocker remains for the bounded text. Wave B **may begin now** because the core chain, role, metric, SR and CCD interfaces are stable; optional field or standardized PPO/SR/zero panels remain dependency-gated.
+
+| Module | Narrative | Provenance | Scope / claim ceiling | Interface | Gate |
+|---|---|---|---|---|---|
+| 03 Contracts/metrics | Clear six-job cadence; compact qualification | Exact local ledger and honest deferrals | Correctly defines rather than reports results | Canonical exports are usable verbatim; V5 signs are explicitly non-Legacy; Level/Stage is distinct from spatial L2 | **PASS** |
+| 04 Kármán performance | L2-first and readable, but one DTW sentence is internally ambiguous | Strong role/ROI/window ledger; target-hash distinction retained | Legacy principal, V5 secondary; no pooling or causal claim | Imports M03 correctly; export to M06 needs an explicit non-equivalence warning | **REVISE** |
+| 05 Kármán SR law | Correct five-step argument and local limits | Formula, deployment and deletion contracts are traceable; leaf audit honestly deferred | Kármán/cloak only; deletion is not causal ablation; no positive Illusion transfer | Correctly separates July formula-parent DTW/deletion from August same-protocol roles; no same-protocol PPO/SR/zero values invented | **PASS** (conditional `ready`) |
+| 06 Field/CCD | Mean→share→residual→CCD→POD order is strong | Exact corrected authority identities, roles, mask, window and lineage | Descriptive/noncausal; no pairing, energy share, mechanism, OID or superiority claim | SR/CCD separation is correct, but M04 near-collision is not explicitly reconciled and “spatial-L2 context” is too loose | **REVISE** |
+
+## Exact cross-module interface matrix
+
+| Producer | Exact export | Importers and permitted use | Reusable verbatim wording |
+|---|---|---|---|
+| M03 | Legacy and V5 are separate executable chains; chain-specific plant, body/action/Re, calibration, role, comparator, metric and window | M04–06; contracts only, never numerical continuation | “Legacy and V5 define separate executable evidence chains; never pool their numbers or imply numerical continuation/equivalence.” |
+| M03 | V5 `(front, top, bottom)`, `omega=-(12a+b)U0/R`, zero bias, weight-0.1 smoothing; Legacy `(front, upper, lower)`, `alpha=omega/U0`, `Re_D=Re_code/2` | M04–06; Legacy modules import only the Legacy branch | “The V5 map does not apply to Legacy SR.” |
+| M03 | Spatial L2 primary; DTW complementary/signal-level/case-specific; periodic mean, phase8, event, mean-field and steady estimands distinct | M04 owns standardized PPO field evaluation; M05 imports only the distinction; M06 defines its own corrected mean estimand | “Field matching uses spatial L2 as primary evidence and DTW as complementary, signal-level and case-specific evidence.” |
+| M04 | Standardized Legacy PPO target/controlled/zero field roles; ROI `-6..+14D` from `x_s`, `|y-y0|/D<=5`; geometry guard; complete-cycle mean and independent phase8 | M05 as performance context; M06 only as a separate prior field result | “The field result is a performance arbiter for any reduced law; it does not itself identify a necessary, causal or unique action component.” |
+| M05 | Canonical mapped-shared Legacy SR formula, imposed `G`, July 200/400-step formula-parent DTW, and 40-step parent-relative deletion ranking | M06 and synthesis; action-law interpretation only | “These are deletion tests, not causal ablations: they do not establish term necessity, sufficiency, mechanism, uniqueness or universal sensitivity.” |
+| M05 | August Target/PPO/SR/Zero is a separate standardized `480+160` acquisition | Any direct same-protocol panel only | “Exact PPO/SR/zero comparator numbers are not asserted here unless the current role artifacts and proposed panel supply them.” |
+| M06 | Corrected four-role mean ledger; exact common mask; non-paired `19x10` residual; rank-3 action coordinates; POD-equivalent tested subspace | Later synthesis only under corrected estimand | “The percentages are scalar error-difference shares, not explained variance, energy share, causal contribution or term attribution.” |
+
+## Numerical near-collision: mandatory reconciliation
+
+The Module 04 and Module 06 numbers are **different estimands**, despite their close magnitudes. Current authority does not call them replication or validation.
+
+- **M04 standardized PPO field metric:** three independently acquired roles (`target`, frozen-PPO `controlled`, physical `zero`); registered lattice rectangle relative to sensor station, `-6 <= (x-x_s)/D <= 14` with `x_s=800`, equivalent in x extent to `34 <= x/D <= 54`; geometry guard only, with no persisted solver-exact mask; unweighted lattice-cell RMS after division by `U0`; role-local complete-cycle means over accepted crossings, plus a separate eight-nearest-snapshot phase diagnostic. Values: `E_mean_ctl=0.0857710225`, `E_mean_zero=0.4752063239`.
+- **M06 corrected CCD mean ledger:** four roles (`target`, `zero`, `constant mean`, `DRL`) in the corrected CCD acquisition/lineage; inclusive `34 <= x/D <= 54`, `|y/D| <= 5`; **exact intersection of all four solver-fluid masks**; coordinate-weighted RMS; persisted velocities already nondimensional and not divided by `U0` again; retained four-role mean window that also supports the constant comparator. Values: `E_DRL=0.0857787860`, `E_zero=0.4694352273`, `E_constant=0.1110140739`.
+
+The x extents coincide after substituting M04's Legacy `x_s/D=40`, which explains only geometric similarity. Role acquisitions, role set, retained windows, masks and quadrature/normalization differ. Consequently the values must **never be equated, subtracted across modules, averaged, used to compute a cross-module discrepancy, or described as replication, validation, agreement, closure or independent triangulation**. M04's complete-cycle and phase8 results do not validate M06's corrected four-role ledger; M06's `93.42/6.58` decomposition exists only inside its own ledger.
+
+Recommended shared sentence, reusable verbatim in the interface ledger and later synthesis: **“Although the standardized PPO and corrected CCD mean errors are numerically close and share the same nominal x extent, they arise from different role acquisitions, retained windows, masks and quadrature/normalization contracts; they are separate estimands, not replication or validation, and are not combined arithmetically.”**
+
+## SR interface finding
+
+Module 05 passes this interface. Its July formula-parent evidence uses native Legacy DTW over Stage-3/deletion windows (40-step deletions; 200-step standard deployment; finite 400-step extension). The August standardized collector instead acquires Target/PPO/SR/Zero after 480 warm-up intervals plus 160 retained boundaries. Module 04's PPO field values come from standardized target/controlled/zero field roles and spatial estimands, not from the July formula-parent deletion contract. Module 05 explicitly avoids claiming a same-protocol PPO/SR/zero numerical comparison and makes the panel conditional on exact standardized artifacts. No formula-parent DTW or deletion delta is relabelled as M04 spatial L2.
+
+## Targeted revision prompts
+
+### Module 04 — REVISE
+
+1. **Layer 1 paragraph 2, first quantitative sentence:** replace the wording that says controlled normalized DTW is `0.932899` “compared with `0.625127` for zero” and then says “its native ... mean is `0.952132`.” Higher similarity appears to be favorable, so remove any ambiguity introduced by calling these “DTW” without the authority's similarity orientation. State explicitly that these are normalized/native **similarity scores** (if that is the authority label), that controlled is higher than zero, and preserve the exact formula/window. Do not call a higher value an error or distance. Mirror the same nomenclature in ledger row 31 and the DTW figure contract.
+2. **Layer 1 paragraph 5 or M06 export:** append the mandatory reconciliation sentence above after the M06 handoff. This is an interface caveat, not a new result. It may instead be recorded in the coordinator ledger/audit if later synthesis is required to import it verbatim; M04 prose need not duplicate M06's values.
+
+`ready` condition: correct the DTW metric orientation/naming in narrative, ledger and figure contract; retain the target-hash clarification; and bind the non-equivalence sentence in the coordinator interface ledger or M04 export. No raw-field access is required for textual readiness. A redrawn field panel remains optional and artifact-dependent.
+
+### Module 06 — REVISE
+
+1. **Layer 1 paragraph 1, after the estimand definition:** insert the mandatory reconciliation sentence above, shortened only by retaining all four distinctions: role acquisition, retained window, exact mask versus geometry guard, and coordinate-weighted/already-nondimensional versus unweighted `U0`-normalized metric. Do not say the M04 and M06 errors agree.
+2. **Layer 2 SR-bridge row and Interfaces paragraph:** replace generic “spatial field-error context” with “M06's corrected four-role coordinate-weighted mean-field estimand.” Add that M04's standardized PPO `E_mean`/phase8 values are separate and cannot validate or numerically close the M06 ledger.
+
+`ready` condition: make this estimand boundary explicit in either M06 prose plus ledger, or in M06 ledger plus the coordinator interface ledger with mandatory downstream import. All corrected authority-led array/hash deferrals remain nonblocking. Optional panel redraw becomes a blocker only if the panel is retained and cannot be sourced under its documented contract.
+
+## Non-revision findings
+
+- **M03 definitions:** no silent V5-to-Legacy sign or normalization transfer occurs. The V5 negative action map and observation-only `VecNormalize` are explicitly chain-local; Legacy uses native `alpha`, order and compatibility scaling. Pipeline `Level-2/Level-3` and `Stage-3 short/standard` are explicitly distinct from spatial L2. This module can become `ready` after this second-audit PASS is recorded; no prose edit is required.
+- **M05:** narrative, provenance, scope and interface checks pass. It can become `ready` once the coordinator records reconciled M03/M04 imports and either drops the optional standardized PPO/SR/zero panel or binds exact same-protocol role artifacts before retaining it. This reconciliation can be performed in the ledger/audit without editing M05 prose.
+- **Deferrals:** absent raw fields, solver-exact mask for M04, leaf-hash recomputation, model/normalizer payloads, and dense CCD arrays are correctly nonblocking where current conflict-free authority binds the exact bounded text. They become blocking only for a retained new panel, a reproducibility claim, or resolution of a concrete authority conflict. The Kan99b lift-sign issue is quarantined to affected lift claims and does not block Modules 03–06. No current true blocker remains.
+
+## Systematic prompt audit and Wave B decision
+
+One repeated defect remains across **Modules 04 and 06**: the prompt/gate requires exact local provenance but does not require an explicit **near-collision check across modules when similar numbers occupy nearby narrative roles**. Add one coordinator-level check to the master/quality/interface controls for future waves: “For numerically similar cross-module results, compare role acquisition, role set, ROI, mask, weighting/normalization and window; explicitly classify same estimand, related diagnostic or different estimand, and prohibit arithmetic or validation language unless exact authority licenses it.” This recommendation meets the two-module threshold. No broader prompt change is justified for DTW orientation, which is local to Module 04.
+
+**Wave A readiness:** core interfaces are stable enough for Wave B. Record M03 and M05 as second-audit PASS/`ready` under the conditions above; retain M04 and M06 as `revise` until the two targeted wording fixes are re-audited. Wave B agents may proceed using the canonical exports and mandatory non-equivalence sentence, while treating new field figures and a standardized PPO/SR/zero panel as optional dependencies rather than section-text blockers.
+
+## Coverage ledger
+
+- `read`: revised Round-3 rule/master/quality/interface ledger; Round-2 freeze/style; revised Modules 03–06; targeted current `wake_l2.py`, corrected CCD math/results, and SR pipeline authority.
+- `mapped`: module-linked manifests and external artifact identities, only to understand documented bindings.
+- `not-covered`: broad Round-1/package re-exploration, raw NPZ/dense arrays, universal hash recomputation, new CFD/training and external literature; unnecessary for the identified inconsistencies.
+- `external-inaccessible`: authority-bound external/symlink payloads; artifact audit deferred and nonblocking for this bounded audit.
+
+
+## Second-audit closure — targeted Modules 04 and 06
+
+This closure re-audits only the targeted revisions against the ready conditions above and the updated cross-module near-collision gate; it does not reopen the first audit.
+
+- **Module 04 — PASS; `ready`.** The narrative, provenance row, figure contract and claim ladder now identify the normalized and native DTW quantities as similarity scores, state that higher is better, provide controlled and zero values under both contracts, and prohibit error/distance wording. The target-hash clarification remains intact. Its Module 06 export now binds the mandatory non-equivalence rule and names different acquisitions/role sets, windows, masks and weighting/normalization; it prohibits arithmetic combination and replication/validation/agreement/triangulation language. All M04 ready conditions are met.
+- **Module 06 — PASS; `ready`.** Layer 1 now distinguishes the standardized PPO and corrected CCD ledgers by role acquisition/set, retained window, geometry guard versus exact four-role common mask, and unweighted `U0` normalization versus coordinate weighting on already nondimensional velocity. The provenance bridge and interface paragraph explicitly identify M06's corrected four-role coordinate-weighted mean-field estimand and prohibit using M04 `E_mean`/phase8 to validate or numerically close it. All M06 ready conditions are met.
+- **Module 03 — PASS; `ready`, unchanged.** No targeted change alters its chain-local action/normalization definitions, Legacy/V5 separation, or Level/Stage versus spatial-L2 distinction.
+- **Module 05 — PASS; conditional-ready state unchanged.** M03/M04 imports are now coordinator-reconciled, but a direct standardized PPO/SR/zero panel remains optional and may be retained only after exact same-protocol role artifacts, comparator values, metric and window are bound. If that panel is omitted, no prose or scientific blocker remains.
+
+**Remaining optional dependencies, exactly scoped:** (1) a newly redrawn M04 Legacy field panel requires raw role payload access plus manifest/hash audit; (2) a reproduced/redrawn V5 selected-policy panel requires the same-selection `best_model.zip` and matching VecNormalize artifact; (3) an M05 direct PPO/SR/zero panel requires complete standardized same-protocol Target/PPO/SR/Zero role artifacts and exact comparator values; and (4) a newly redrawn M06 figure must remain sourceable under the documented publication/data contract. These dependencies condition only retained optional panels or reproducibility/artifact-audit claims, not the bounded section text. Canonical CCD alias promotion, universal leaf-hash recomputation, M04 solver-exact-mask reconstruction and the quarantined Kan99b lift-sign issue remain nonblocking at the present claim ceiling.
+
+**Final Wave A disposition:** the core Modules 03, 04 and 06 are `ready`; Module 05 is conditionally ready solely on the optional direct-comparator panel decision. No true core blocker remains, the near-collision interface is closed, and **Wave A core modules are ready for Wave B**.

docs/JFM_WYQ/ROUND3_BOTTOM_UP/01_INTRODUCTION.md

@@ -0,0 +1,69 @@
+# Round 3 revised candidate — Section 1: Introduction
+
+**Working title:** *Interpreting learned hydrodynamic cloaking and illusion in a fluidic pinball*
+
+**Status:** ready for coordinator second-audit closure; not manuscript approval.
+
+## Section question and bounded answer
+
+**Question.** What physical problem, literature gap and bounded contribution justify this paper without overpromising applications, mechanism or priority?
+
+**Bounded answer.** The paper separates three evidence levels: (a) named wake-signature or visual PPO capability across the declared Kármán, Vortex and Illusion tasks under distinct contracts; (b) independent spatial field matching only for named Legacy Kármán Re100; and (c) a bounded Legacy Kármán control-skeleton interpretation. The latter combines executable action-law evidence with a corrected four-role constant-mean comparator and a separately centred, non-paired residual organized descriptively by action-correlated correction-field modes. This does not establish linear-cloak/Navier–Stokes equivalence, a causal mechanism, universal transfer, positive Illusion symbolic regression, priority or quantitative experimental validation.
+
+## Layer 1 — five-paragraph near-prose draft
+
+### Paragraph 1 — concept and compact genealogy
+
+Hydrodynamic cloaking and illusion are observer-centred matching objectives. Cloaking asks a controlled body to approach a declared background or incident signature in the field or signal space seen by an observer; illusion instead asks it to approach a declared non-zero signature associated with another target. A compact genealogy runs from transformation-based and conformal optical cloaks to active exterior cloaking and non-zero-target illusion in linear field equations, with later wave analogues and active hydrodynamic proposals showing how cancellation or retargeting can be formulated under restricted assumptions (Pendry, Schurig & Smith 2006; Leonhardt 2006; Vasquez, Milton & Onofrei 2009; Ma et al. 2013; Urzhumov & Smith 2012). This history supplies vocabulary for declared references, active sources and target signatures; it does not make a linear cloak equivalent to a finite-Reynolds-number Navier–Stokes wake, nor does “cloaking” mean invisibility to every possible observer. [R1-Introduction literature]
+
+### Paragraph 2 — finite-Reynolds-number pinball difficulty
+
+The fluidic pinball makes the distinction concrete. Three interacting circular bodies generate separated shear layers, vortex shedding, symmetry breaking and multiple temporal organisations; independent rotation must modify a spatially extended wake rather than a single scalar force. The pinball therefore supplies a compact but nonlinear benchmark in which mean deficit, phase, frequency and downstream spatial organisation need not change together (Deng et al. 2020, 2021, 2022). Earlier rotating-cylinder and machine-learning control studies establish that actuation can alter separation, wake structure and force, but their objectives and observables do not by themselves answer whether a specified downstream signature has been matched (Cornejo Maceda et al. 2021; Gautier et al. 2015). The physical problem here is finite-Re wake-signature matching with limited actuation and sensing, not scalar drag optimisation relabelled as cloaking.
+
+### Paragraph 3 — capability and explanation gap
+
+Deep reinforcement learning offers a way to search this nonlinear control problem from forces and sparse downstream velocity observations. It can discover effective feedback in settings where a tractable controlled-flow model is unavailable, as shown for cylinder control and later studies of robustness, stability and partial observation (Rabault et al. 2019; Paris, Beneddine & Dandois 2021; Li & Zhang 2022; Xia et al. 2024; Wang et al. 2024). That capability leaves a scientific gap. A reward or sparse-signal similarity score establishes performance only in its declared observer space; it does not independently establish field matching, reveal which action structure survives closed-loop deployment, or explain how a persistent wake correction is organised. Adjacent active hydrodynamic stealth work and sparse-model approaches sharpen the motivation, but do not supply this paper’s rotating-pinball evidence chain or a basis for a priority claim (Ren, Wang & Tang 2021; Loiseau, Noack & Brunton 2018). The question is how to interpret learned capability without promoting policy correlation into a physical law.
+
+### Paragraph 4 — evidence chain needed for a bounded interpretation
+
+Answering that question requires linked but non-equivalent tests. The execution problem is wake-signature matching, whereas independent spatial field matching is reserved for the named Legacy Kármán Re100 comparison, with L2 primary and DTW complementary, signal-level and case-specific. The learned action must be tested as an executable closed-loop law under the same Legacy plant and action convention, rather than accepted from offline action fit alone. The corrected four-role analysis then compares a constant-role mean field with zero and DRL target errors, while a separately centred, non-paired dynamic residual is organized descriptively by action-correlated correction-field modes. These tests address distinct performance, action-law and field-role questions; they do not independently confirm one another or establish a mechanism.
+
+### Paragraph 5 — question, contribution, scope and bridge
+
+This paper asks whether named force-plus-sparse-signature control can produce hydrodynamic cloaking and illusion in the fluidic pinball, and what can be learned about the Kármán cloak without confusing capability with mechanism. Its contribution has three levels: named wake-signature or visual PPO capability across declared Kármán, Vortex and Illusion tasks; independent spatial field matching only for Legacy Kármán Re100; and a bounded Legacy Kármán control-skeleton interpretation. The interpretation is Kármán/cloak-only: Vortex and Illusion remain bounded capability demonstrations, Erase is unsupported quantitatively, positive Illusion SR is historical/negative, and any experiment is author-described qualitative, open-loop and provenance-gated. Applications are prospective only. The next section fixes the plant, action, chain, role and metric contracts; the following capability section presents the Kármán-dominant Fig. 1 progression, with no experiment. [Round-2 freeze; M02; M03–M10]
+
+## Layer 2 — provenance, literature identities and style recommendation
+
+| Paragraph | Exact identity / read level | Boundary |
+|---|---|---|
+| P1 | Pendry, Schurig & Smith, “Controlling Electromagnetic Fields,” *Science* 312, 1780–1782 (2006), DOI `10.1126/science.1125907`; Leonhardt, “Optical Conformal Mapping,” *Science* 312, 1777–1780 (2006), DOI `10.1126/science.1126493`; Vasquez, Milton & Onofrei, “Active exterior cloaking for the 2D Laplace and Helmholtz equations,” *PRL* 103, 073901 (2009), DOI `10.1103/PhysRevLett.103.073901`; Ma et al., “Experiments on active cloaking and illusion for Laplace equation,” *PRL* 111, 173901 (2013), DOI `10.1103/PHYSREVLETT.111.173901`; Urzhumov & Smith, “Flow stabilization with active hydrodynamic cloaks,” *PRE* 86, 056313 (2012), DOI `10.1103/PhysRevE.86.056313`. Full-PDF per R1 index. | Linear genealogy is vocabulary/method precedent only; no Navier–Stokes equivalence. |
+| P2 | Deng et al., “Low-order model for successive bifurcations of the fluidic pinball,” *JFM* 884, A37 (2020), DOI `10.1017/jfm.2019.959`; Deng et al., *JFM* 918, A4 (2021), DOI `10.1017/jfm.2021.299`; Deng et al., *JFM* 934, A24 (2022), DOI `10.1017/jfm.2021.1105`; Cornejo Maceda et al., *JFM* 917, A42 (2021), DOI `10.1017/jfm.2021.301`; Gautier et al., *JFM* 770, 442–457 (2015), DOI `10.1017/jfm.2015.95`. Full-PDF per style guide. | External precedent only; no project threshold, solver, Re or result imported. |
+| P3 | Rabault et al., *JFM* 865, 281–302 (2019), DOI `10.1017/jfm.2019.62`; Paris et al., *JFM* 913, A25 (2021), DOI `10.1017/jfm.2020.1170`; Li & Zhang, *JFM* 932, A44 (2022), DOI `10.1017/jfm.2021.1045`; Xia et al., *JFM* 981, A17 (2024), DOI `10.1017/jfm.2024.69`; Wang et al., *JFM* 988, A4 (2024), DOI `10.1017/jfm.2024.333`; Ren et al., *Physics of Fluids* 33, 093602 (2021), DOI `10.1063/5.0060690`; Loiseau et al., *JFM* 844, 459–490 (2018), DOI `10.1017/jfm.2018.147`. Full-PDF per R1/style records. | Capability and interpretation motivation only; unresolved `Ren19b` excluded; no priority claim. |
+| P4–P5 | Round-2 freeze/style; R1 Introduction, DRL, SR, field/CCD/OID, steady/theory and metrics dossiers; current Modules 03–08, ready M09 and M10, and current authorities. Full-report/read for repositories; not independently recomputed or universally artifact-audited. | Three levels remain separate: named wake-signature/visual PPO capability across declared tasks; Legacy Kármán Re100 spatial field matching only; bounded Legacy Kármán control-skeleton interpretation. Vortex has no event L2; Illusion field L2 is conditional and Illusion SR historical/negative; Erase has no supported quantitative route; experiment is qualitative/open-loop and provenance-gated. |
+
+**Observed literature style.** The supplied JFM papers generally move from physical problem and comparator to evidence, interpretation and a local limitation. Deng uses physical chronology and transition to motivate refinement; Rabault and Paris reach the control gap quickly; Gautier places an explicit law beside closed-loop behaviour; Loiseau moves from sparse observations toward fuller representation. Strong words follow dedicated tests, and bridges introduce the next scientific question.
+
+**Manuscript recommendation.** Use that rhythm at a lower claim register: concept and gap → finite-Re difficulty → named capability and explanation gap → independent evidence chain → bounded question and bridge. Keep genealogy compact, cite around physical transitions, state dangerous limits here, reference Figure 1 prospectively without panel narration, and organize the contribution as capability, field matching and Kármán interpretation rather than a package list.
+
+## Figure, claim and interface contracts
+
+- **Figure 1:** presented in the following capability section as a Kármán-dominant progression across named contracts; it does not establish a common score, independent field matching beyond Legacy Kármán Re100, positive Illusion SR, interpretation transfer or pooled Legacy/V5 evidence. It contains no experiment.
+- **Definitions next:** cloak/illusion as declared reference matching; wake-signature matching as execution; spatial L2 primary and DTW complementary. No project equations or numbers belong here.
+- **Green:** named wake-signature or visual PPO capability across the declared Kármán, Vortex and Illusion tasks; independent spatial field matching only for named Legacy Kármán Re100; and bounded Legacy Kármán control-skeleton interpretation.
+- **Amber:** a constant-role mean-field comparator accounts for most of the measured zero-to-DRL target-error reduction, while a smaller separately centred, non-paired residual is organized descriptively by action-correlated correction-field modes under the Legacy Kármán contract; steady is contextual.
+- **Red:** linear-cloak/Navier–Stokes equivalence, all-observer invisibility, universal/causal/mechanistic/necessary/unique/optimal/robust/priority claims, positive Illusion SR, pooled chains, DTW-as-delay, causal CCD, application catalogues or quantitative experiment validation.
+
+## Imports, exports, bridge and coverage ledger
+
+**Imports:** frozen title and vocabulary; R1 literature and motivation; current Round3 controls and ready Modules 02–10, including M09’s author-described experiment/provenance gate and M10’s three-level manuscript-wide claim ladder. **Exports:** cloak/background and illusion/non-zero-target definitions; wake-signature matching; L2-primary/DTW-complementary evaluation only for the named Legacy Kármán Re100 field comparison; Kármán/cloak-only control skeleton; broad capability versus asymmetric interpretation depth. **Bridge:** declare plant, chain, action, role, Re, comparator, metric and window before comparing cases, then present Fig. 1 in the capability section.
+
+| Status | Coverage and reason | Boundary |
+|---|---|---|
+| `read` | Round3 controls/interface/Wave-A audit; Round2 freeze/style; R1 Introduction and current-document audit; ready Modules 02–10, including M09 and M10; current Confirmation Introduction; cited JFM/Confirmation material | Read directly; not independently verified or artifact-audited |
+| `mapped` | Broader manifests, archive navigation and authority-linked external payload identities | Mapped only; not promoted as read evidence |
+| `not-covered` | New literature search; unresolved Ren19b; raw fields/NPZ/models; new CFD/training; hashes; Figure 1 generation; quantitative experiment | Unnecessary for bounded introduction; no claims inferred |
+| `external-inaccessible` | External/symlink-backed fields and hidden raw artifacts | Authority-led `artifact-audit deferred`; no raw-artifact claim made |
+
+## Final targeted self-audit
+
+The narrative retains five argumentative paragraphs and explicitly separates the three manuscript-wide levels: named wake-signature/visual PPO capability; Legacy Kármán Re100 spatial field matching; and bounded Legacy Kármán control-skeleton interpretation. The constant-role CCD comparator, non-paired residual, cross-case gaps and provenance-gated qualitative experiment are bounded without causal or action-magnitude language. Literature identities and DOI/read levels are preserved; no new result or citation was added. Figure 1 is correctly bridged to the following capability section. This final targeted revision passes the paragraph-boundary, claim-ladder, provenance and interface checks; the module is ready for coordinator second-audit closure, not manuscript approval.

docs/JFM_WYQ/ROUND3_BOTTOM_UP/02_CAPABILITY_PROGRESSION.md

@@ -0,0 +1,131 @@
+# Round 3 first-pass module — Section 3: capability and progression
+
+**Status:** revised candidate for second audit; not self-ready and not final manuscript prose. The bounded narrative is text-ready for parent review, while raw and redrawn Fig. 1 panels remain artifact-dependency-gated. This module does not add experiment or reopen the analytical modules.
+
+## Section question and bounded answer
+
+**Question.** How should Fig. 1 and its accompanying text establish the progression from steady discovery through periodic Kármán cloaking, nonperiodic vortex response and the `075` Illusion objective extension while reserving analytical depth for Kármán?
+
+**Bounded answer.** Named visual and quantitative cases establish a progression in temporal dynamics and objective, while only Legacy Kármán Re100 has the full evidence intersection for deep interpretation; the figure therefore demonstrates breadth of execution under distinct contracts, not a pooled score, arbitrary-flow/generalization result or positive Illusion SR.
+
+## Layer 1 — readable paragraph draft/map
+
+### Paragraph 1 — viewing contract and figure composition
+
+Fig. 1 should answer a capability question before the paper turns to interpretation: can the same broad force-plus-sparse-observation execution be used for named targets whose temporal and objective contracts differ? The answer is presented as a left-to-right progression, not as a common efficacy ranking. A small steady panel supplies the historical discovery seed; the visually dominant Kármán block establishes the principal field-matching comparison; smaller Vortex and `075` Illusion panels show, respectively, an event-relative transient extension and a non-zero-target objective extension. Legacy and V5 lanes must be printed in the panel labels and caption. Each panel names its case, chain, role, comparator and metric status, and the composition contains no experiment. [R2 freeze/style; M03; M04; M07; M08]
+
+**Figure placement:** place Fig. 1 beside this opening paragraph. Allocate approximately 50–55% of the visual area to Kármán, with the steady, Vortex and `075` Illusion panels subordinate. Use visual juxtaposition to show the progression; do not place a common efficacy bar, pooled score, or gallery inventory beside it.
+
+### Paragraph 2 — short steady discovery seed
+
+The first capability seed is steady control: near-uniform or constant rear rotation was explored before the periodic Kármán analysis. In this opening section it should be stated only as historical discovery and visible low-deficit capability under its own steady contract, not as a temporal performance score or as evidence for the Legacy SR/CCD chain. The full V5 steady/free-slip result, accepted sign, profile metric and theory limits belong to the later steady return. This ordering preserves the physical chronology—steady discovery followed by Kármán interpretation—without allowing the early seed to pre-empt the Kármán analysis. [R1 DRL/cases; R1 steady dossier; M08]
+
+**Seed panel contract:** V5 steady/free-slip, accepted steady q_ctl against uniform/no-body q_in under E_inf_vector/profile status; q_blk is nonranking, and stationary/reverse views are contextual only. The panel establishes discovery chronology and a bounded visual endpoint only. It does not establish periodic phase matching, Legacy numerical continuation, mechanism, stability, optimum or pooled efficacy.
+
+### Paragraph 3 — Kármán headline and reason for depth
+
+The headline comparison is the Legacy `karman_re100` target, physical-zero and frozen PPO roles under one Legacy solver and geometry contract. Here `Re100` is the Legacy code convention `U_0(2D)/\nu=100`, equivalent to single-cylinder `Re_D=50`; it is not a V5 numerical continuation. In the registered downstream rectangle, the controlled field is closer to the target than physical zero in both principal complete-cycle mean-field L2 and the separately defined eight-slot phase diagnostic: `E_mean=0.085771` versus `0.475206` for zero, and `E_phase8=0.112680` versus `0.630878`. The corresponding zero-relative values are `\eta_mean=0.819508` and `\eta_phase8=0.821392`. The exact principal field claim is spatial L2 under the Legacy `drl-pinball-wake-l2-v2` contract; one concise complementary annotation is the six-channel target-normalized DTW similarity, `0.932899` controlled versus `0.625127` zero under the 30-step cycle contract (higher is better), not a field norm or physical delay. [M03; M04]
+
+Kármán is selected for depth because it has the evidence intersection needed for interpretation: a named Legacy target/zero/controlled field comparison, executable Legacy SR evidence, corrected mean/dynamic field roles and CCD description, with V5 Re100 seed/retention information as secondary context. The selection is not a claim that steady, Vortex or Illusion are less important, nor an inference that they failed. It is the case in which capability, independent field matching and bounded action/field interpretation coexist under traceable contracts. [M04; M05; M06; M08]
+
+### Paragraph 4 — Kármán to Vortex, then cloak to Illusion
+
+The smaller Vortex lane changes the temporal coordinate: periodic phase is removed and named Lamb/Taylor disturbances are read at event-relative offsets. The visual should retain the event labels and offsets—rather than relabelling them phase—and may show pulse-like action and downstream fields; target/control/zero roles are conditional and are not treated as audited together. A quantitative event-relative ROI L2 is not currently available in the authority chain, so no event L2 number should appear in Fig. 1; current Vortex DTW remains a case-specific signal comparison and is not a physical delay. The bounded reading is a named comparable-scale transient capability hypothesis, not arbitrary-flow control or transfer of the Kármán control skeleton. [M07; R1 metrics] The `075` Illusion lane changes the objective instead: the policy is asked to match a declared non-zero target signature rather than approach the physical background. The geometry is the target **radius ratio 0.75** (physical radius 15 at `L0=20`), even where historical labels use diameter-like names. Use the retained V5 `ill_075L`, seed 43, as PPO capability evidence with target and controlled visual roles; physical zero and any field metric are conditional on exact role artifacts and must not be implied by the visual alone. This is a V5 historical capability lane, not a Legacy numerical continuation and not positive Illusion SR. The negative Legacy Illusion SR result may be stated locally as the reason SR is not promoted, without allowing that historical caveat to overwhelm the opening capability progression. [M07; R1 DRL/cases]
+
+### Paragraph 5 — boundary, degradation and bridge to Kármán depth
+
+The breadth preview must carry its boundary beside its claim. If space permits after the steady, Vortex and Illusion progression is readable, one evidence-backed degradation indicator may be retained as a small optional secondary mark; otherwise it is dropped: in the V5 `kar_re400`, seed 43, the mean field improved relative to zero (`E_mean=0.177469` versus `0.221541`) while the phase diagnostic worsened (`E_phase8=0.490471` versus `0.450626`, `\eta_phase8=-0.08842`). This is a named-condition warning that mean and phase behavior can diverge, not a cross-Re law or a failure explanation. The figure must not combine this value with the Legacy Re100 L2 values, the corrected CCD mean ledger, Vortex event observations, steady profile quantities or Illusion PPO scores. With the progression and its limits visible, the text bridges to the Kármán sections: Kármán is the sole deep case because its evidence intersection permits performance, executable SR and descriptive mean/dynamic organization to be examined together; the later cross-case section returns to Vortex and Illusion under their own contracts. [M04; M06; M07]
+
+## Layer 2 — panel, provenance and claim ledger
+
+### Fig. 1 panel contracts
+
+| Panel / visual role | Chain, case, roles and comparator | Metric/status shown | Caption-level limit |
+|---|---|---|---|
+| **A. steady discovery seed** | V5 steady/free-slip; accepted steady q_ctl against uniform/no-body q_in under E_inf_vector/profile status; q_blk is nonranking; stationary/reverse views are contextual only; late/profile view | Visual steady endpoint or profile; nonperiodic late/profile status; no periodic score | Contextual historical seed only; not Legacy SR/CCD evidence, phase evidence, mechanism, stability, optimum or pooled efficacy |
+| **B. Kármán headline (dominant, 50–55%)** | Legacy `karman_re100`; target, physical zero and frozen PPO controlled roles; each role is under the Legacy solver/API and geometry; target is the comparator | Principal spatial L2: complete-cycle `E_mean`; secondary separately reported `E_phase8`; exact values `0.085771` vs `0.475206` and `0.112680` vs `0.630878`; `\eta` values optional in text/caption; DTW similarity only as complementary signal annotation (`0.932899` vs `0.625127`, target-normalized) | Geometry-guarded ROI; solver-exact common mask not persisted. L2 is not a validation Level/Stage, DTW is not field equivalence/delay/causality. No V5 pooling |
+| **C. Vortex extension** | Legacy `vortex_lamb`/`vortex_taylor` named visual roles; target/controlled/zero roles are conditional on exact artifact binding and are not presumed jointly audited; event offsets remain literal | Event-relative visual and action trace; no quantitative event L2 currently. DTW, if annotated, is native Vortex signal comparison with its own window | Offsets are event-relative, not phase. No invented event L2; no arbitrary-flow, universal or Kármán-skeleton transfer claim. Quantitative event panel remains dependency-gated |
+| **D. `075` Illusion objective extension** | V5 `ill_075L`, seed 43; separately constructed non-zero target and controlled PPO; physical zero only where exact role is bound; target radius ratio 0.75 | PPO capability visual, with reward/force/DTW status only under exact V5 role and calibration identity; physical zero and field metrics are conditional on exact role artifacts | Historical V5 capability only; target matching is not full-state equivalence, common efficacy or positive Illusion SR. Negative Legacy Illusion SR is separate historical evidence |
+| **E. one degradation indicator (small, optional)** | V5 `kar_re400`, seed 43; controlled versus zero under the V5 periodic field contract | `E_mean` improves while `E_phase8` worsens: `0.177469` vs `0.221541`; `0.490471` vs `0.450626`; `\eta_phase8=-0.08842` | Named diagnostic only, n=1; demonstrates estimand divergence, not cross-Re generalization, failure cause, or a pooled comparison |
+
+**Fixed composition and density hierarchy:** A–D are left-to-right; B is visually dominant. The steady, Vortex and Illusion progression has priority over E. E is optional, may be dropped if it compromises that progression, and if retained is only a small marker/inset attached to B or its caption, never a fifth gallery panel. Explicitly show Legacy/V5 lanes and role labels. Do not include experiment.
+
+### Quantitative provenance rows
+
+| Claim | Exact authority/artifact | Chain, solver/geometry, case/seed/role, Re | Comparator, metric, ROI/window | Evidence state |
+|---|---|---|---|---|
+| Steady discovery chronology | `ROUND1_BOTTOM_UP/01_DRL_CONTROL_AND_CASES.md` chronology; `ROUND1_BOTTOM_UP/04_STEADY_THEORY_AND_CROSSMAP.md`; current steady README/RESULTS and M08 | V5 steady/free-slip, three-cylinder pinball, `Re_D=50`; separate from Legacy Kármán | Accepted steady q_ctl versus uniform/no-body q_in under E_inf_vector/profile status; q_blk nonranking; stationary/reverse contextual only; late/profile, nonperiodic | `read`; authority-bound; not independently recomputed or artifact-audited; full theory deferred to M08 |
+| Legacy Kármán mean/phase headline | M04 §2 ledger; `src/drl_pinball/eval/wake_l2.py`, schema `drl-pinball-wake-l2-v2`; bound Legacy role metadata | Legacy API/solver, `1280×512`, `karman_re100`, code Re100 = `Re_D=50`; target/zero/controlled, frozen Legacy PPO | Registered `-6≤(x-x_s)/D≤14`, `|(y-y_0)/D|≤5`, `x_s=800`; complete-cycle mean `E_mean`; eight independently phased nearest-snapshot `E_phase8`; accepted-crossing window | `read`; current authority values; raw payload and solver-exact mask `artifact-audit deferred` |
+| Legacy Kármán DTW | M04 `dtw_summary.json` rows and `legacy_test/core/dtw_metrics.py` contract | Legacy `karman_re100`; target/controlled/zero; target-normalized six-channel cycle | 30-step window, lag channel 3, target-channel max-abs normalization; similarity, higher is better; native score separate | `read`; role authority; artifact audit deferred; complementary signal metric only |
+| V5 seed/retention context supporting selection | M04 §2; V5 `latest_results_summary.json`, `latest_eval_summary.csv`, Re100 metadata | V5/CelerisLab `2000×600`, `kar_re100`, seeds 41–45, selected seed 45; code Re100 = `Re_D=50` | Best-checkpoint discovery and retained evaluation; no Legacy numerical continuation | `read`; retained historical authority; same-selection model/normalizer audit deferred for a redraw |
+| Vortex visual/event boundary | M07 §§2 and ledger; `DATA_DICTIONARY.md`; reproduction plot manifest; `legacy_test/core/dtw_metrics.py` | Legacy Vortex named Lamb/Taylor cases; transient `SI=800`, `MAX_STEPS=150`; target/control/zero roles conditional | `event_fields.npz`, `relative_offsets=[-10,-5,0,5,10]`; current evaluator lacks event-field L2; native Vortex DTW has no lag and uses final-window roll | `read`; visual navigation and contract authority; raw events not independently verified; event-L2 artifact/evaluator `not-covered`, external roots `external-inaccessible` where applicable |
+| V5 `ill_075L` PPO capability | M07 §3 and ledger; `env_illusion.py`, V5 training/result authorities | V5/CelerisLab `ill_075L`, seed 43; target cylinder radius ratio 0.75, physical radius 15; target and controlled roles | PPO reward/force/DTW capability view under V5 target/calibration; no field L2 unless exact roles pass | `read`; retained historical capability; model/normalizer/target leaf audit deferred |
+| Illusion negative SR boundary | M07 ledger; `src/SR_analysis/CLAIMS.md`, `HANDOFF.md`, standardized Illusion role authority | Legacy `illusion_1L`; SR versus physical zero; separate historical standardized acquisition | Long `480+160` window, native/normalized DTW and mean/phase L2 near equal or slightly worse for SR; historical/negative | `read`; authority-bound negative evidence; not a comparator to V5 PPO and not positive SR |
+| Optional V5 degradation marker | M04 §2 ledger and current field summary | V5/CelerisLab `kar_re400`, seed 43; controlled versus zero; code Re400 | Same periodic field contract; `E_mean=0.177469` vs `0.221541`; `E_phase8=0.490471` vs `0.450626`; `\eta_phase8=-0.08842` | `read`; bounded diagnostic; artifact audit deferred; n=1 and not generalization evidence |
+
+### Figure caption requirements
+
+The caption must say that Fig. 1 is a capability/progression composition, not a common metric comparison. It must identify for every panel: chain (Legacy or V5), solver/geometry where relevant, case and seed, target/reference and role, temporal coordinate (steady late/profile, periodic cycle/phase, or event-relative offset), metric status and comparator. It must explicitly state that the Kármán L2 values are Legacy Re100 field estimands, that DTW is complementary signal similarity, that Vortex offsets are event-relative and have no current quantitative event L2, and that `ill_075L` is V5 PPO capability with target radius ratio 0.75. The caption must prohibit pooled efficacy, arbitrary-flow generalization, positive Illusion SR and experiment-as-evidence. Any retained E marker must be labelled a named V5 diagnostic showing mean/phase divergence, not a failure cause or cross-Re law.
+
+## Claim ladder
+
+**Green — supported at named scope**
+
+- The historical progression is steady discovery → periodic Kármán cloaking → event-relative Vortex demonstration → non-zero-target Illusion PPO capability.
+- Legacy Kármán Re100 target/controlled/zero is closer to its target under both the declared complete-cycle mean-field and eight-slot phase L2 rows, with the exact values and ROI above.
+- Legacy Kármán DTW similarity is higher for controlled than zero under its own target-normalized cycle contract; it complements, rather than replaces, spatial L2.
+- Named Vortex visual roles and V5 `ill_075L` PPO provide bounded extension evidence under separate contracts.
+- Kármán is the selected deep case because it has the full evidence intersection for performance and interpretation.
+
+**Amber — bounded interpretation**
+
+- The cases establish breadth in temporal dynamics and objective: steady is nonperiodic, Kármán periodic, Vortex event-relative and Illusion non-zero-target periodic.
+- Pulse-like Vortex actuation supports a comparable-scale event-response hypothesis in the named cases.
+- The V5 degradation marker shows that mean and phase estimands can diverge under a named cross-Re case.
+- A force-centred sparse-observation execution can be retargeted under named cases, while the positive control-skeleton interpretation remains Kármán/cloak-specific.
+
+**Red — prohibited in this section**
+
+- A pooled efficacy score, common bar, arithmetic combination, replication, validation, agreement or triangulation across these panels or with M06 CCD values.
+- Arbitrary-flow control, continuous/general Re or offset/scale generalization, robustness or universality.
+- Positive Illusion SR, target tracking by SR, or V5/Legacy numerical continuation/equivalence.
+- Vortex event offsets called phase, Vortex event L2 invented, DTW called a physical delay or field equivalence.
+- Illusion visual similarity called full-state equivalence or efficacy over physical zero without exact role artifacts.
+- Steady discovery promoted to Legacy SR/CCD evidence, mechanism, stability or optimum.
+- Experiment included in Fig. 1 or used as quantitative validation.
+
+## Imports, exports and bridge
+
+**Imports.** M03 supplies the chain, role, action, ROI and metric vocabulary. M04 supplies the exact Legacy Re100 L2/DTW headline and V5 seed/retention boundary. M05 supplies positive SR scope limited to Legacy Kármán/cloak. M06 supplies descriptive Kármán mean/dynamic and CCD boundaries, but no numbers are imported into Fig. 1. M07 supplies Vortex event semantics, Illusion PPO capability and negative Illusion SR. M08 supplies steady chronology and the later full steady/theory return.
+
+**Exports to M04/M07.** This section selects Legacy Kármán Re100 as the principal depth case while preserving M04’s field estimands and M07’s event/Illusion boundaries. It exports no cross-case score and no Kármán interpretation to Vortex or Illusion.
+
+**Final bridge.** Fig. 1 establishes capability breadth under named, non-pooled contracts. The next section can therefore begin with the Legacy Kármán field comparison and ask what independent spatial matching shows beyond the wake-signature objective; later modules return to SR/field interpretation and full cross-case metrics without treating the progression figure as a results gallery.
+
+## Genuine blockers, deferrals and coverage ledger
+
+**Genuine blockers for retained figure content:** a quantitative Vortex event-L2 panel is blocked until exact target/control/zero event roles, target identity, common ROI/mask, event window and evaluator output are bound. A quantitative Erase panel is not part of Fig. 1 and remains unsupported under the current historical route. An independently redrawn V5 selected-policy panel requires the matching model, VecNormalize and target artifacts. These blockers do not block the bounded text or the Vortex visual/Illusion capability panels.
+
+**Authority-led deferrals:** raw Legacy Kármán fields, external/symlink-backed Vortex/Illusion payloads, exact solver masks and model/normalizer leaf audits are `artifact-audit deferred` where current authorities bind the role and summary without conflict. No publication-ready figure is claimed to exist.
+
+| Status | Sources inspected and reason | Evidence state |
+|---|---|---|
+| `read` | Round3 rule, master prompt, quality gate, interface ledger and Wave-A audit; Round2 freeze and section style guide; ready Modules 03–08 (especially M04/M07/M08); Round1 DRL/cases, current-document coverage, field/CCD/OID, steady and case-metrics dossiers; current DRL, Legacy acquisition, SR, wake-L2 and steady authorities/manifests | Read directly for this first pass; not independently verified or artifact-audited |
+| `mapped` | Existing reproduction-plot navigation and external role links used to identify candidate visual roles and panel dependencies | Indexed/mapped only; not treated as read or numerical authority |
+| `not-covered` | New CFD/training, new offline Vortex event evaluator, raw NPZ/field arrays, exact event-L2 results, publication panel generation, experiment package, new literature search | Outside scope; no claim or number inferred from these omissions |
+| `external-inaccessible` | Symlink/external raw Vortex/Illusion/Kármán payloads and selected model bundles not exposed in ordinary checkout inspection | Authority-led `artifact-audit deferred`; nonblocking for bounded text, blocking for any new artifact-audit/reproducibility claim |
+
+## Cross-module near-collision check
+
+The Legacy M04 principal PPO field errors (`0.085771`, `0.475206`) must not be compared arithmetically with M06 corrected CCD errors (`0.0857787860`, `0.4694352273`) despite similar magnitudes and nominal x extent. M04 uses three independently acquired target/controlled/physical-zero roles, a geometry guard, unweighted lattice-cell RMS after `U0` normalization, and a complete-cycle/phase8 contract; M06 uses four roles, a retained four-role window, an exact common solver-fluid mask and coordinate-weighted RMS on already nondimensional velocities. They are **different estimands**, not replication, validation, agreement or triangulation. The same prohibition applies to steady profiles, Vortex event observations, Illusion PPO reward/DTW and any proposed common figure bar.
+
+## Self-audit under the current gate
+
+- **Two layers:** Layer 1 has five argumentative paragraphs with transitions and local limitations; Layer 2 carries panel contracts, exact provenance, roles, chains, Re conventions, comparators, estimands, ROI/windows and evidence status. This is a revised candidate for second audit, not a self-readiness declaration; the narrative may be text-ready while retained panels remain artifact-gated.
+- **Coverage:** the controlling Round2 documents, Round1 dossiers and all ready Modules 03–08 were read. The ledger distinguishes read, mapped, not-covered and external-inaccessible; read is not independent verification or artifact audit.
+- **Progression fidelity:** steady appears only as a short historical seed; Kármán is visually dominant and quantitatively exact; Vortex uses event-relative semantics with no invented event L2; Illusion is V5 PPO capability at target radius ratio 0.75; experiment is excluded.
+- **Claim boundaries:** no pooled score, arbitrary-flow/generalization, positive Illusion SR, Kármán interpretation transfer, CCD number import or experiment claim. The negative Illusion SR caveat is local and does not replace the PPO capability statement.
+- **Figure integrity:** every proposed panel has chain, role, comparator and metric status. The Kármán and optional degradation values are not merged with M06 or other estimands. No gallery inventory or common efficacy bar is proposed.
+- **Anti-index / anti-AI check:** the narrative is progression-led rather than a package inventory; numbers appear only when they establish the headline or a visible boundary; figure contracts follow the narrative rather than replacing it; no citation, number, artifact, mechanism or generalization was invented.
+
+**Disposition:** revised candidate for second audit. The bounded text is text-ready for parent review; raw/redrawn panels remain artifact-dependency-gated. Do not label this module self-ready or manuscript-ready.

docs/JFM_WYQ/ROUND3_BOTTOM_UP/03_CONTRACTS_METRICS.md

@@ -0,0 +1,109 @@
+# Round 3 Wave-A revised candidate — Section 2 contracts and metrics
+
+## Section question and bounded answer
+
+**Question.** What must be fixed about the plant, executable chain, actuation, observations, roles and estimands before later comparisons answer the same physical question rather than merely share a label?
+
+**Bounded answer.** Valid comparisons use separate Legacy and V5 executable chains, each with its own plant, action convention, calibration and role artifacts. The policy performs wake-signature matching from forces and sparse downstream velocities; independent field matching uses spatial L2 as the primary evidence and DTW only as complementary, case-specific signal evidence, with periodic, event, mean-field and steady windows kept distinct. This is a revised candidate for second audit, not a ready module.
+
+## Layer 1 — six-paragraph near-prose section draft
+
+### Paragraph 1 — plant and chain
+
+The comparison begins with a declared plant, not with a policy score. The fluidic pinball comprises three ordered circular bodies and a downstream three-probe signature plane, while Kármán and Illusion introduce different target constructions. In the active V5 Kármán chain, CelerisLab uses a `2000×600` D2Q9 FP32 MRT configuration with double-buffer streaming, uniform regularized inflow, `neq_extrap` outlet and free-slip lateral walls. The V5 body order is front, top, bottom; its common lattice convention is `L0=20`, `D=20`, `U0=0.01`, and `Re_code=U0(2L0)/nu`, so `re100` corresponds to `Re_D=50`. V5 Illusion translates the pinball and obtains the non-zero target from a separate target-cylinder acquisition; its historical `target_diam` parameter is passed as a radius ratio. Legacy uses the separate LegacyCelerisLab plant and historical geometry/API, so a shared case label never licenses numerical pooling. [V5 train/env; Legacy role authority]
+
+### Paragraph 2 — actuation, sign and readback
+
+A normalized action becomes a physical rotation only after body order, sign and persistence have been fixed. In V5, `a=(a_front,a_top,a_bottom)` lies in `[-1,1]^3` and maps to
+
+\[
+\omega_i=-\frac{(12a_i+b_i)U_0}{R},\qquad b_i=0,
+\]
+
+with `R=L0/2` and an explicit weight-`0.1` exponential smoother. The negative sign and front/top/bottom order belong in the plant schematic and every downstream action-law comparison. They must not be transferred to Legacy: Legacy SR and acquisition use the native `(front, upper, lower)` ordering, `\alpha=\omega/U_0`, and `Re_D=Re_code/2` for its Kármán notation, with its own action scale/bias and `FlowField.run` persistence. CelerisLab uploads a host-side `set_body` value at the next run; force readback is an `[f_x,f_y]` vector whose time normalization is explicit, while sensor readback is footprint-area averaged and optionally time averaged. [V5 environments; Legacy SR/action authority; CelerisLab API]
+
+### Paragraph 3 — observations, calibration and objective
+
+The execution contract is wake-signature matching rather than an assertion of full-field observability. V5 Kármán supplies six controlled-body force components followed by six velocity channels, ordered as the two components of each of the three probes; V5 Illusion appends reconstructed target drag and lift. Solver readback, calibrated force and sensor scales, and online observation whitening are separate operations: active V5 calibration stores native units, while SB3 `VecNormalize` whitens observations only and must travel with the selected policy. The historical Legacy compatibility path may use `SENSOR_CC=78` before its old-policy scaling; that factor is not active V5 calibration. Roles are likewise explicit: `target` is a reference trajectory, `zero` is physical-zero/uncontrolled, `controlled` is the policy trajectory, and `constant` is a contextual comparator where acquired. Force terms and six-channel sensor similarity form the reward; aggregation and target comparison are case-specific, with V5 Kármán averaging controlled-body forces against zero and V5 Illusion summing them against one target body. [V5 training pipeline; Legacy acquisition authority]
+
+### Paragraph 4 — field matching and signal complement
+
+The independent field comparison is registered before performance is inspected. The current downstream ROI is the fixed rectangle `-6≤(x-x_s)/D≤14` and `|(y-y_0)/D|≤5`, a `20D×10D` box guarded against known solids but not proven by persisted solver masks. For periodic roles, the primary estimand is the `U0`-normalised area-RMS difference between complete-cycle mean velocity fields, reported both absolutely and relative to the same-case zero comparator through `η=1-E_controlled/E_zero`. The eight-slot phase diagnostic uses independently phased centre-probe transverse-velocity crossings and nearest saved snapshots; it is not a repeated-crossing ensemble. DTW remains a complementary sparse-signal measure, not a spatial L2 surrogate, physical delay or causal test. This hierarchy asks separately whether the persistent field and the evolving signature approach the target. [wake-L2 authority]
+
+### Paragraph 5 — case-specific evaluation windows
+
+The estimand changes with the dynamics and must be named in every result row. Periodic Kármán and periodic Illusion can use complete-cycle mean-field L2 and a separately reported eight-slot phase diagnostic, with native or normalized DTW labelled by its own window and formula. Vortex comparisons, if exact roles exist, use event-relative downstream snapshots at declared offsets and retain event labels rather than calling them phase. Erase uses a mean-field comparison only when target, controlled and zero roles are exact; the historical unsupported route is not repaired by relabelling. Steady comparisons use a late snapshot or profile, not a temporal periodic mean. Pipeline `Level-2/Level-3` or `Stage-3 short/standard` terminology must also remain separate from the spatial L2 metric. [case-metric authority; wake-L2 authority]
+
+### Paragraph 6 — compact CFD qualification and scope
+
+The numerical qualification is a compact contract check, not a second results section. The retained CelerisLab K2 rotating-cylinder package uses D2Q9 MRT, FP32, double-buffer streaming, Bouzidi moving boundaries, no LES and `neq_extrap` outlet under the Kan99b open-flow convention; it is a partial pass because the stored mean-lift sign disagrees with the reference even though the accepted frequency, drag and fluctuation observables fall within their declared bands. The retained Sah04 S2 confined-cylinder package provides a named Strouhal check under its parabolic/channel-stabilised, no-slip and realised-Re convention. These anchors qualify selected solver observables only; they do not qualify a DRL policy, SR law, CCD interpretation or cloaking mechanism. Full matrices, hashes, sensitivity rows and smoke diagnostics belong in the supplement. [CelerisLab validation dossier; Kan99b/Sah04 authorities]
+
+## Layer 2 — paragraph and claim-to-evidence ledger
+
+| Paragraph / claim | Exact provenance and contract | Comparator, metric, ROI/window and status |
+|---|---|---|
+| P1: V5 plant | `src/drl_pinball/train/TRAIN_PIPELINE.md` §§1–2; `train/env_karman.py`; `DATA_DICTIONARY.md`; active V5 Kármán, CelerisLab `2000×600`, D2Q9 FP32 MRT, double buffer, regularized uniform inlet, `neq_extrap`, free-slip; body order front/top/bottom; `L0=20`, `D=20`, `U0=.01`; `Re_code=U0(2L0)/nu` | Same matched config/calibration/target for target, zero and controlled; case/seed/role-specific; field window is the registered downstream ROI below; supported executable contract, not independently rerun |
+| P1: V5 Illusion geometry | `train/env_illusion.py` lines 50–60, 157–184; active V5 Illusion target acquisition and translated controlled plant; target cylinder exists only during target construction; `target_diam*L0` is passed as radius | Target versus controlled only under matched target/calibration identity; periodic or case-specific metric later; supported executable contract |
+| P1: Legacy separation | `ROUND1_BOTTOM_UP/01_DRL_CONTROL_AND_CASES.md` §§4–5; `src/drl_pinball/legacy_test/README.md`; Legacy Kármán/Illusion role artifacts and LegacyCelerisLab API | No cross-chain score pooling; Legacy evidence is retained historical unless current artifact says otherwise; read authority, raw payload artifact-audit deferred |
+| P2: V5 action | `train/env_karman.py` `_action_to_omega`; `env_illusion.py` `_action_to_omega`; active V5, action `(front,top,bottom)`, `[-1,1]`, `b=0`, `\omega=-(12a+b)U0/R`, `R=L0/2`, EMA weight `.1` | Record command, applied omega, body ID/order and sign per trajectory; supported source contract |
+| P2: Legacy action/SR convention | `ROUND1_BOTTOM_UP/01_DRL_CONTROL_AND_CASES.md` §§6–7; `src/SR_analysis` current Kármán authority; Legacy Kármán SR, native `(front,upper,lower)`, `\alpha=\omega/U0`, `Re_D=Re_code/2`, distinct scale/bias and `FlowField.run` EMA | Legacy SR comparisons remain Legacy-only; formula/action artifacts and closed-loop trajectory are the comparator; active Kármán scope, read authority |
+| P2: readback | `CelerisLab/README.md` §§API/telemetry; `CELERISLAB_VALIDATION_DOSSIER.md` §§3–4; CelerisLab current solver/API, `set_body` host action uploaded at next run; force `[f_x,f_y]`, `normalize=True` per-step average, sensor area normalization always applied and time normalization separate | Record sampling interval and normalization; no force-tensor transpose; array/API orientation safeguard is stated once here; read authority, implementation not rerun |
+| P3: V5 observations | `train/env_karman.py` module contract and `_normalize_obs`; `env_illusion.py`; active V5, Kármán 12-vector = six forces then six sensor components; Illusion 14-vector adds target `C_D,C_L` | Reward uses the corresponding target role; observation order is not field metric; supported source contract |
+| P3: normalization | `TRAIN_PIPELINE.md` §§4–5; `DESIGN_DECISIONS.md`; active V5 native calibration-v2 scales plus observation-only `VecNormalize`; Legacy compatibility `SENSOR_CC=78` path | Selected policy requires matching normalizer; inference freezes `training=False`; native V5 and Legacy scaling not comparable; supported contract, retained-policy compatibility artifact-audit deferred |
+| P3: roles/reward | `legacy_test/README.md` §§standardized acquisition; `TRAIN_PIPELINE.md` §§3–5; `DESIGN_DECISIONS.md`; target, zero, controlled, constant; Kármán controlled-force mean versus zero, Illusion controlled-force sum versus one target; sensor similarity is reward-adjacent | Target has no policy reward; native DTW and normalized offline DTW retain separate formulas/windows; reward is not field error; supported role contract, exact selected artifacts artifact-audit deferred |
+| P4: ROI | `WAKE_L2_AND_SENSOR_PLACEMENT_RESEARCH_NOTES.md` §§3–5; `src/drl_pinball/eval/wake_l2.py`, schema `drl-pinball-wake-l2-v2`; registered `-6≤(x-x_s)/D≤14`, `|(y-y0)/D|≤5`, `D=20`; geometry guard because solver masks not persisted | Controlled and zero versus target on same case/role window; `U0`-normalised area-RMS; exact solver-mask audit deferred, not a blocker when authority binds ROI |
+| P4: periodic L2 hierarchy | Wake-L2 note §§5–6; `wake_l2.py`; periodic target/controlled/zero roles; `E_mean` is complete-cycle arithmetic mean-field error; `E_phase8` is eight independently phased nearest snapshots; `η=1-E_ctl/E_zero` | L2 primary field matching; mean and phase windows distinct; phase8 not ensemble and DTW not L2; implemented where role schema passes, raw external fields artifact-audit deferred |
+| P5: case-specific semantics | `ROUND1_BOTTOM_UP/08_EXPERIMENT_AND_CASE_METRICS.md` §§4–5; `ROUND2_TOP_DOWN/05_REVISED_STRUCTURE_FREEZE.md` §§2,5; wake-L2 note §§5–7; periodic, vortex event-offset, Illusion, Erase mean-field and steady late/profile contracts | No pooled efficacy/phase score; exact Vortex/Erase roles are needed only for future quantitative panels; current authority defines semantics, event/role payloads artifact-audit deferred |
+| P5: pipeline terminology | `.cursor/rules/jfm-round3-section-drafting.mdc` line 11; relevant package stage/level documentation | `Level-2/Level-3` and `Stage-3 short/standard` are pipeline labels; spatial `L2` is field error; no “L2 gate” wording; current drafting rule |
+| P6: CelerisLab qualification | `docs/JFM_WYQ/CELERISLAB_VALIDATION_DOSSIER.md` §§3,6–10; Kan99b K2 and Sah04 S2 manifest-backed full artifacts; current solver revision recorded by dossier | K2: Kan99b `Re=U∞D/nu`, partial pass due mean-lift sign; S2: Sah04 realised-Re Strouhal check; named windows and artifacts in manifest; qualification evidence, selected artifacts not rerun here |
+| P6: qualification boundary | Same dossier §§5,10–12; `CelerisLab/README.md` FP16S/EsoPull limitations | Selected numerical observables only; no DRL/SR/CCD/mechanism validation; full matrix and hashes supplement; K2 sign issue remains a genuine blocker only for claims using affected lift convention |
+
+## Figure, equation and table contracts
+
+1. **Plant/action/sensor/ROI schematic.** Shows the two lanes, body order, V5 negative action sign, Legacy action distinction, probe plane and registered ROI. It does not establish optimal placement, observability, chain equivalence or a solver-exact mask.
+2. **Legacy/V5 contract table.** Defines solver, geometry, body/action/Re convention, normalization and role names. It does not authorize pooled statistics or transfer of the V5 action map to Legacy SR.
+3. **Definitions block.** Reuses chain-specific Re/action equations, observation/reward role definitions, `E_mean`, `E_phase8` and `η`. It does not turn reward into field error or DTW into delay.
+4. **Compact CFD table.** Gives the selected K2/S2 qualification status and essential configuration; full matrices and artifact identities belong in the supplement. It does not validate policy, SR, CCD or cloaking mechanism.
+
+## Claim ladder
+
+**Green.** Legacy and V5 are separate executable chains; each later comparison must carry its plant, body/action/Re convention, calibration, role, comparator, metric and window. V5 action order/sign/smoothing, readback semantics, observation normalization, role definitions, wake-signature reward, and L2-primary/DTW-complementary hierarchy are executable contracts. K2 is a partial Kan99b qualification and S2 is a named-condition Sah04 solver anchor.
+
+**Amber.** Sparse probes are physically motivated signature monitors, not optimal or fully observable. Reward/DTW can define wake-signature matching, while L2 independently tests registered field matching. Mean, phase, event and steady estimands may support different bounded comparisons under their own roles.
+
+**Red.** Do not pool Legacy/V5 values or transfer V5 action signs/maps to Legacy SR; call DTW a physical delay, causal test or field equivalence; relabel event snapshots as phase; call steady late snapshots temporal means; use a validation Level/Stage as an “L2 gate”; claim solver qualification proves DRL/SR/CCD/cloaking mechanism; or call all Kan99b/Sah04 matrix cases completed.
+
+## Imports, exports, bridge and genuine blockers
+
+**Imports.** Round-2 frozen terminology and hierarchy; Round-1 DRL/action/role and metric dossiers; current V5 environments, training/evaluation and data-dictionary authorities; Legacy acquisition/SR authority; wake-L2 implementation/rationale; and the CelerisLab validation dossier and benchmark specifications.
+
+**Audited exports — exact canonical wording for Modules 04–06.**
+
+- **Legacy/V5:** Legacy and V5 define separate executable evidence chains; never pool their numbers or imply numerical continuation/equivalence.
+- **Action:** V5 uses body order `(front, top, bottom)` and `\omega=-(12a+b)U0/R` with zero bias and weight-0.1 smoothing. Legacy SR uses native `(front, upper, lower)`, `\alpha=\omega/U0` and `Re_D=Re_code/2`; the V5 map does not apply to Legacy SR.
+- **Observation/reward:** Wake-signature matching uses forces plus six sparse downstream velocity observations. Force aggregation and target comparison are case-specific; calibration scales, Legacy `SENSOR_CC=78` compatibility, and V5 observation-only `VecNormalize` are distinct layers.
+- **Evaluation:** Field matching uses spatial L2 as primary evidence and DTW as complementary, signal-level and case-specific evidence. Periodic complete-cycle mean L2, eight-slot phase diagnostics, transient event-offset L2, Illusion/Erase mean-field comparisons and steady late/profile errors are distinct estimands.
+- **Qualification:** CelerisLab qualification is compact and configuration-specific; Kan99b K2 is a partial pass because the mean-lift sign/convention remains unresolved, while qualification does not establish a control or mechanism claim.
+
+**Final bridge.** With these definitions fixed, Module 04 can lead with independent Kármán field matching, while Modules 05–06 can interpret Legacy SR and field structure without silently changing plant, action sign, Reynolds convention, role or metric.
+
+**Genuine blockers.** The Kan99b force/lift sign remains a blocker only for claims that use that affected lift comparison. Exact Vortex/Erase target-control-zero roles remain blockers only for proposed future quantitative panels. A selected-policy reproducibility claim is blocked if its exact model/normalizer/target identity is not bound; authority-led external fields, masks and hashes are `artifact-audit deferred`, not blockers.
+
+## Concise source coverage ledger
+
+| Status | Sources and reason | Evidence state / boundary |
+|---|---|---|
+| `read` | Round-3 revised rule, master and quality gate; Round-2 freeze/style guide; Round-1 DRL, coverage and metrics dossiers; current V5 environment/training/evaluation/data-dictionary authorities; Legacy acquisition/SR authorities; wake-L2 note/code; CelerisLab dossier, README and Kan99b/Sah04 specifications | Read in full for this revision; not independently rerun or universally artifact-audited |
+| `mapped` | Broader package indexes and manifests referenced by the read authorities | Mapped only; not claimed read-full or artifact-audited |
+| `not-covered` | Raw selected model/normalizer/target bundles, external reproduction payloads, solver-exact masks and unretained benchmark matrix cases | Authority-led `artifact-audit deferred`; no values inferred |
+| `external-inaccessible` | External/symlink-backed field roots and source exports not exposed in the visible checkout | Authority-led deferral; no essential internal contract is established from inaccessible material |
+
+## Anti-index / anti-AI audit
+
+- Layer 1 now contains six concise near-prose paragraphs with short source tags; repeated template labels and inline evidence blocks have been removed.
+- Layer 2 carries exact provenance, chain, geometry, role, Re, comparator, metric, ROI/window and status without making the narrative a path inventory.
+- Contracts no longer narrate Kármán performance, SR, CCD or cross-case success; detailed outcomes are retained only where needed to define qualification boundaries.
+- Pipeline Level/Stage terminology is separated from spatial L2, and no validation item is called an “L2 gate”.
+- Authority-led artifact gaps are marked `artifact-audit deferred`; only affected lift-sign claims, exact future Vortex/Erase roles and selected-policy identity can block downstream claims.
+- The action-sign/body-order warning is preserved; the force-array orientation safeguard appears once in the ledger rather than as a repeated `tyx/txy` theme.
+- This is a revised candidate for second audit, not a readiness declaration.

docs/JFM_WYQ/ROUND3_BOTTOM_UP/04_KARMAN_PERFORMANCE.md

@@ -0,0 +1,108 @@
+# Round-3 revised candidate — Kármán performance
+
+**Status:** targeted M04 revision completed; second-audit candidate, not final manuscript approval and not self-declared ready. This revision separates near-prose from provenance, makes Legacy Kármán Re100 the principal result, and retains V5 Re100 only as secondary support.
+
+## Section question and bounded answer
+
+**Question.** Under the named Kármán contracts, what does learned wake-signature matching achieve, and what independent field-matching evidence remains after the signal objective is separated from spatial evaluation?
+
+**Bounded answer.** Under the Legacy `karman_re100` target/controlled/zero contract and its registered downstream ROI, the frozen PPO trajectory is closer to the target than physical zero in both complete-cycle mean-field and eight-slot phase-resolved L2; its normalized six-channel DTW is complementary evidence, not the field claim. V5 Re100 supplies secondary evidence that high-scoring checkpoint discovery repeats across five seeds while late retention varies, so the result remains named-condition field matching rather than a pooled or robust cross-chain conclusion.
+
+## Layer 1 — concise near-prose draft
+
+The principal comparison is the Legacy Kármán Re100 target, physical zero and frozen PPO trajectory, evaluated within one Legacy solver and geometry contract. Here `Re100` is the code convention `U_0(2D)/\nu=100` (`U_0=0.01`, `D=20`, `\nu=0.004`), equivalent to single-cylinder `Re_D=50`; it is not the V5 Re100 case under another name. In the registered rectangle `-6\le (x-x_s)/D\le14`, `|(y-y_0)/D|\le5`, with Legacy `x_s=800`, the controlled field is closer to the Legacy target than the corresponding physical-zero field. [Legacy field authority; Module 03]
+
+The signal result is consistent with that field comparison but does not define it. Under the same 30-step window, lag channel and target-channel normalization, the Legacy controlled role has a six-channel target-normalized DTW similarity score of `0.932899`, higher than the zero role’s `0.625127`; the corresponding native reward-compatible similarity scores are `0.952132` for controlled and `0.688244` for zero. Higher scores indicate greater similarity. These are role-local Legacy quantities, not V5 scores, and they are never interpreted as errors or distances. DTW records signal alignment and similarity; it is not a spatial norm, physical delay or causal response. [Legacy Re100 `dtw_summary.json`; Module 03]
+
+The independent field result is carried by spatial L2. For complete accepted cycles, the Legacy controlled mean-field error is `E_mean=0.085771`, versus `0.475206` for zero, giving `\eta_mean=0.819508`. The corresponding eight-slot phase diagnostic is `E_phase8=0.112680`, versus `0.630878` for zero, giving `\eta_phase8=0.821392`. Here `E_mean` is the `U_0`-normalized area-RMS difference between complete-cycle arithmetic mean fields, whereas `E_phase8` is the RMS over eight independently phased nearest saved boundary snapshots; neither uses a solver-exact persisted common-fluid mask. Thus, for this named Legacy comparator and ROI, the field result agrees across persistent mean and phase-resolved diagnostics, without implying that all downstream flow states are identical. [Legacy `wake_l2.py` row; Module 03]
+
+The agreement has a defined dynamic boundary. Legacy target and controlled roles each contain eight complete-cycle fields and nine accepted crossings, while zero contains four complete cycles and five crossings; the controlled relative frequency mismatch is `0.004831`, compared with `0.376794` for zero. This supports the bounded reading that the controlled trajectory is closer to the target under both declared field estimands, but the phase metric remains a one-snapshot-per-slot diagnostic rather than a repeated-crossing ensemble. The companion V5 `kar_re100` package is secondary: five seeds repeatedly found high-reward checkpoints, and the retained deterministic evaluation selected seed 45 with reward `0.931258` and DTW `0.918458`; late retention ratios ranged from `0.162632` to `0.858143`. This supports a discovery/retention distinction, not Legacy/V5 pooling or a general convergence claim. [Legacy metadata; V5 retained evaluation authority]
+
+The result therefore hands two bounded questions forward. To Module 05, the Legacy Re100 field comparison establishes the performance target for testing whether a reduced action law reproduces the controlled trajectory under the same Legacy plant and window; it does not identify action-law complexity. To Module 06, the exact Legacy mean/phase L2 roles and their separate windows provide the performance baseline for mean/dynamic field organization, while the field metric remains geometry-guarded and descriptive rather than causal. [Module 03; current field authorities]
+
+## Layer 2 — compact provenance ledger
+
+| Claim / quantity | Exact source and artifact | Chain; solver/geometry; case/seed/role; Re | Comparator and metric/estimand | ROI/window and status |
+|---|---|---|---|---|
+| Principal target/zero/controlled roles | `src/drl_pinball/data/reproduction/legacy/karman_re100/{target,zero,controlled}/metadata.json`; schema `drl-pinball-legacy-acquisition-v2` | Legacy solver/API; `1280×512`; `karman_re100`; controlled frozen model `d1a3o12_re100.zip` with frozen `norm.json`; Re code 100 = `U0(2D)/nu=100`, `Re_D=50` | target live builder trajectory; zero physical-zero/uncontrolled counter-bias trajectory; controlled frozen-policy trajectory | `SI=800`, 480 warm-up, 160 post-step boundaries; 8 phase fields; read, authority-bound, artifact-audit deferred for raw payloads |
+| Registered field ROI | `src/drl_pinball/eval/wake_l2.py`, schema `drl-pinball-wake-l2-v2`; Module 03 §4 | Legacy; `D=20`, `x_s=800`, conservative solid extent 636 | fixed inclusive ROI; no outcome-dependent mask | lattice bounds `x=680..1080`, `y=156..355`; geometry-guarded ROI, solver-exact mask not persisted; read and agreed with Module 03 |
+| Legacy complete-cycle mean L2 | Exact `_evaluate` row from current Legacy roles, using `wake_l2.py`; row values reproduced from bound role artifacts | Legacy; Re code 100; controlled vs target and zero vs target; no seed field in Legacy role | `E_mean_ctl=0.08577102245825166`; `E_mean_zero=0.47520632385441813`; `eta_mean=0.8195078260689812`; `U0`-normalized ROI RMS difference of complete-cycle mean fields | accepted rising centre-probe `u_y` crossings; target 9, controlled 9, zero 5; 8, 8 and 4 complete cycles; current authority value, raw artifact audit deferred |
+| Legacy phase L2 | Same exact `_evaluate` row and `phase_fields.npz` role schema | Legacy; Re code 100; controlled/zero versus target | `E_phase8_ctl=0.11267999504986256`; `E_phase8_zero=0.6308775162147815`; `eta_phase8=0.8213916455195072`; eight-slot nearest-boundary RMS | independent role phasing, canonical slots; one snapshot per slot, not repeated-crossing ensemble; current authority value, artifact-audit deferred |
+| Legacy current/normalized DTW similarity | `controlled/dtw_summary.json`, `zero/dtw_summary.json`, `target/dtw_summary.json`; native implementation `legacy_test/core/dtw_metrics.py` and offline metrics contract | Legacy; `karman_re100`; controlled and zero role histories versus target; no V5 transfer | target-normalized six-channel cycle DTW similarity score: controlled `0.9328993565511687`, zero `0.6251268882807846`; native similarity score: controlled `0.9521319744438127`, zero `0.6882437288676819`; higher is better and controlled is higher under both contracts; these are not errors or distances | historical window/`CONV_LEN=30`, lag channel 3, target-channel max-abs normalization; read, current role authority, artifact-audit deferred |
+| Dynamic boundary | Legacy role metadata plus the exact field-evaluation row | Legacy Re100; target/controlled/zero | controlled relative frequency mismatch `0.004831181571494927`; zero `0.3767940098934097`; period CV controlled `0.00020952970987246327`, zero `0.06474211289191471` | physical time is Legacy-relative; relative same-case diagnostic only; bounded observation, not cross-chain frequency evidence |
+| V5 secondary five-seed support | `src/drl_pinball/train/results/latest/tables/latest_results_summary.json`; `latest_eval_summary.csv`; `train/output/kar_re100_seed41–45/meta.json` | V5/CelerisLab; `2000×600`; `kar_re100`; seeds 41–45; code Re100 = `Re_D=50` | best rewards `0.9266385, 0.9301960, 0.9160065, 0.9221958, 0.9411612`; selected retained evaluation seed 45 reward `0.931258`, DTW `0.918458` | 500 iterations; retained evaluation 360 steps, last-180 tail; historical `legacy-policy-v1` compatibility, not current-contract retraining; read authority, artifact-audit deferred |
+| V5 retention support | Same V5 `latest_results_summary.json` retention records | V5; same five named seeds and case | validated last-50 retention ratios `0.6703388, 0.2004286, 0.1626321, 0.8581434, 0.6373129` | training-retention diagnostic; best-versus-retained distinction explicit; normalizer pairing required only for reproducibility/selected-policy panel, not for this textual seed-retention claim |
+| V5 Re400 boundary, if retained in figure | Current summary authority and current wake-L2 summary authority | V5/CelerisLab; `kar_re400`, seed 43; code Re400; support chain only | `E_mean_ctl=0.177469` vs zero `0.221541`, but `E_phase8_ctl=0.490471` vs zero `0.450626`; `eta_mean≈0.1989`, `eta_phase8=-0.08842` | same periodic field contract; current authority diagnostic, artifact-audit deferred; do not use as Legacy principal evidence |
+
+### Authority/interface reconciliation with Module 03
+
+- **Agreement:** Module 03 and `wake_l2.py` agree on `drl-pinball-wake-l2-v2`, registered `[-6D,+14D]×[-5D,+5D]` ROI, `D=20`, `E_mean` as complete-cycle mean-field RMS, `E_phase8` as eight independently phased nearest-snapshot diagnostic, `eta=1-E_role/E_zero`, and geometry-guarded/no-persisted-exact-mask status.
+- **Agreement:** Module 03’s Legacy Re100 role semantics agree with the current metadata: Legacy target is live builder-generated, zero is physical-zero/uncontrolled, controlled is frozen Legacy PPO; Legacy roles use their own chain and are not V5 continuations.
+- **Conflict/clarification:** Module 03 says every role must carry target identity/hash when available; Legacy metadata binds run-generated target-array hashes separately for target, controlled and zero. These hashes differ because the trajectories are independently generated; they must not be treated as one shared target identity without an explicit target-role reconciliation. This is a provenance clarification, not a numerical conflict.
+- **Deferral:** Module 03’s normalizer/VecNormalize contract applies to V5 retained policy reproducibility. It does not condition the Legacy field text above, whose frozen Legacy `norm.json` is directly bound in the controlled and zero metadata. Same-selection V5 model/normalizer verification is therefore a selected-policy reproducibility/panel dependency, not a blocker to the Legacy principal result.
+
+## Figure, equation and table contracts
+
+1. **Principal target/zero/controlled Legacy plate.** Show Legacy Re100 target, zero and controlled fields with chain, role, phase slot/window and fixed vorticity limits. It supports visual comparator context and can accompany the exact L2 rows; it does not prove mask exactness, causality or V5 equivalence.
+2. **L2-first result panel.** Report Legacy Re100 absolute `E_mean`, `E_phase8` and zero-relative `eta` before the DTW companion. Caption must state ROI, `U0`, accepted-crossing window, eight-slot nearest-snapshot semantics and geometry-guarded mask status. It does not establish a universal cloak score or repeated-crossing uncertainty.
+3. **DTW companion.** Show Legacy controlled/zero target-normalized DTW similarity scores and label native versus target-normalized similarity pipelines. State explicitly that higher is better (`0.932899` controlled versus `0.625127` zero normalized; `0.952132` controlled versus `0.688244` zero native) and never label these values as errors or distances. The panel supports signal-level matching under the declared window; it does not become field L2, a delay or a causal test.
+4. **V5 seed/retention inset.** Show best discovery and late-retention spread only as secondary support, with selected seed 45, historical compatibility path and best/retained labels. A same-selection model/normalizer pair is required for a reproducibility or selected-policy panel, not for the textual statement that retention varies.
+5. **Equations.** Use Module 03 canonical `E_mean`, `E_phase8` and `eta` definitions; do not add a second DTW equation here unless Module 03 requires it. Distinguish spatial L2 from any pipeline Level-2/Level-3 or Stage-3 validation gate.
+
+## Claim ladder
+
+**Green — supported at named scope**
+
+- Legacy `karman_re100` controlled is closer to its target than its physical-zero comparator in both the complete-cycle mean-field and eight-slot phase-resolved registered spatial L2 rows above.
+- Legacy controlled target-normalized DTW similarity is higher than zero under the declared role-local normalized cycle contract; native similarity is reported separately, and higher is better for both.
+- V5 Re100 has five retained seeds with repeated best-checkpoint discovery and seed-dependent late retention; this is secondary support, not pooled field evidence.
+
+**Amber — bounded interpretation**
+
+- The Legacy Re100 result is consistent with learned wake-signature matching producing a controlled trajectory whose declared downstream field is closer to target over the registered ROI.
+- Agreement of mean and phase L2 under this Legacy role supports a stronger named-case reading than DTW alone, while phase-slot and mask limitations remain attached.
+- The V5 seed pattern motivates separating discovery from retention and does not transfer a V5 result into the Legacy action-law interpretation.
+
+**Red — prohibited without new evidence**
+
+- A global statement that field matching always survives beyond reward, pooled Legacy/V5 values, or a single cross-chain Kármán score.
+- DTW as physical delay, causal response, field equivalence or mechanism.
+- Robust/general convergence, universal Re transfer, stability, optimal sensor placement, exact solver-mask L2, or statistical uncertainty from the current phase slots.
+- V5 selected-policy reproducibility claims without the matching model/normalizer pair; this restriction does not block the Legacy principal text.
+
+## Imports, exports and bridge
+
+**Imports from Module 03:** canonical role/chain vocabulary; Legacy/V5 solver and Re boundaries; registered ROI; `E_mean`, `E_phase8`, `eta`, phase and mask semantics; spatial-L2 versus pipeline validation-level distinction. This revision reconciles those definitions above and records the one target-hash clarification.
+
+**Exact exports to Module 05 — Kármán action law:** use Legacy `karman_re100` target/zero/controlled as the principal closed-loop performance comparator; preserve its Legacy solver/API, `Re_code=100`/`Re_D=50`, frozen policy and `norm.json`, `SI=800`, 480 warm-up plus 160 retained boundaries, complete-cycle window and role-local target/zero semantics. The field result is a performance arbiter for any reduced law; it does not itself identify a necessary, causal or unique action component.
+
+**Exact exports to Module 06 — Kármán field/CCD:** use Legacy Re100 `E_mean_ctl=0.08577102245825166` versus zero `0.47520632385441813`, `eta_mean=0.8195078260689812`, and `E_phase8_ctl=0.11267999504986256` versus zero `0.6308775162147815`, `eta_phase8=0.8213916455195072`, with the registered Legacy ROI and geometry-guarded mask status. Treat these as separate mean and phase estimands; do not infer action correlation or causality from them. V5 seed/retention and Re400 disagreement remain support-chain context only. Although the standardized PPO and corrected CCD mean errors are numerically close and share the same nominal x extent, they arise from different role acquisitions and role sets, retained windows, masks, and weighting/normalization contracts; they are different estimands, are not combined arithmetically, and must not be described as replication, validation, agreement or triangulation.
+
+**Final bridge.** The named Legacy comparator establishes what the learned policy achieved in the registered Kármán wake; Module 05 can now test action-law compression against that same result, and Module 06 can organize mean and phase field differences without turning the field metric into a causal explanation.
+
+## Genuine blockers and scope-local deferrals
+
+- **No blocker to this section text:** external/symlink-backed raw fields and absent solver-exact masks are `artifact-audit deferred` because current metadata, evaluator code and authority-bound rows identify the roles, comparator, window and exact summary values without conflict.
+- **Panel/raw-audit dependency:** a newly redrawn field panel or independent artifact audit still requires access to the raw role payloads and exact manifest/hash check. Do not claim that this pass performed that audit.
+- **V5 reproducibility dependency only:** same-selection `best_model.zip` plus matching VecNormalize is required to reproduce or redraw a selected V5 policy panel; it does not condition the Legacy principal field/DTW text or the textual seed-retention observation.
+- **No unresolved Module 03 metric conflict:** canonical definitions agree; the target-array hash distinction is recorded as a clarification.
+
+## Coverage ledger
+
+| Status | Sources inspected | Evidence status / reason |
+|---|---|---|
+| `read` | Revised `.cursor/rules/jfm-round3-section-drafting.mdc`; Round3 master and quality; current Module 03; Round1 DRL/control, coverage audit and case-metrics dossiers; `WAKE_L2_AND_SENSOR_PLACEMENT_RESEARCH_NOTES.md`; `RESULTS.md`, `CLAIMS.md`, `WRITING_HANDOFF.md`, `eval/README.md`; `wake_l2.py`; Legacy Re100 metadata and DTW summaries; V5 result tables and Re100 metadata/calibration | Full read; current authority values extracted where stated; raw NPZ/model binaries not independently audited |
+| `mapped` | `data/reproduction_plots/manifest.json`; current plot navigation and external role paths | Role/phase-slot mapping read; derived views, not numerical authority |
+| `not-covered` | New CFD, retraining, external literature, independent mask reconstruction, new panel generation | Outside this revision; no claim depends on them |
+| `external-inaccessible` | External Optane root is symlink-backed and raw field payloads are permission-restricted in ordinary inspection | Authority-led `artifact-audit deferred`; exact role/summary rows were still obtained through the bound evaluator and metadata |
+
+## Anti-index / anti-AI audit
+
+- Layer 1 is five concise prose-like paragraphs with source tags and scientific cadence; it does not contain repeated bold template labels, giant inline evidence blocks or training-path inventories.
+- Layer 2 carries the complete provenance ledger, including chain, solver/geometry, case/role/seed, Re, comparator, metric, ROI/window, status and deferral state.
+- The principal result is Legacy Re100; V5 is visibly secondary, and the 23-row aggregate is not used as the principal claim.
+- L2 remains primary and DTW complementary; native and normalized DTW, mean and phase8, Legacy and V5 are not pooled.
+- Best checkpoint, retained evaluation and late retention are separated; VecNormalize is scoped to V5 reproducibility/panel dependency rather than treated as a universal blocker.
+- No global beyond-reward claim, causal language, generalization claim, invented number or silent target-hash merge is used.
+
+**Disposition:** revised candidate submitted for second audit; not self-declared ready.

docs/JFM_WYQ/ROUND3_BOTTOM_UP/05_KARMAN_SR_LAW.md

@@ -0,0 +1,153 @@
+# Section 5 — Kármán action law: dominant near-constant core and smaller correction
+
+> **Revised candidate — second coordinator audit required.** This module owns the executable Legacy Kármán SR law, its deployment symmetry and bounded term tests. It does not self-declare ready; full steady results belong to Module 08 and field/CCD accounting to Module 06.
+
+## Section question and bounded answer
+
+**Question.** What executable symbolic-regression structure is supported for Legacy Kármán cloaking, and what do its closed-loop arbitration and fixed-window term tests establish?
+
+**Bounded answer.** A symmetry-mapped SR surrogate is documented and passes the named Legacy Kármán formula-deployment validations; its parent-relative deletion tests identify persistent rear counter-rotation as the largest tested contribution, with smaller but non-zero rear-lift feedback. This supports a bounded control-skeleton interpretation under the declared contract, not necessity, uniqueness, mechanism or Illusion transfer.
+
+## Layer 1 — five-paragraph near-prose draft/map
+
+### 1. Question and estimand
+
+Symbolic regression asks whether the frozen PPO action can be represented by a compact, executable observation-to-action surrogate for the named Legacy Kármán cloaking cases. The estimand is controller-level: finite closed-loop wake-signature matching under an unchanged Legacy plant, rather than a model of the flow field or its cause. The imported interface is provisional pending coordinator audit: native action order `(front, upper, lower)`, `alpha=omega/U0`, `Re_D=re_code/2`, force-plus-sparse-sensor observations, and Module 04’s separation of spatial L2 field matching from complementary, case-specific DTW. Offline action error remains a fit diagnostic on PPO-visited states; deployment behavior is the arbiter. [R1-SR; SR authority; provisional Modules 03–04 interface]
+
+This framing makes the next distinction essential: recovering PPO actions is not the same test as executing the recovered law.
+
+### 2. Recovery contract and imposed symmetry
+
+The executable route is Stage 1 trajectory capture, Stage 2 topology discovery and coefficient refit, followed by Stage 3 static checks, a **Stage-3 short gate (40 control steps)** and **Stage-3 standard validation (200 control steps)**. Causal fitting pairs post-state `i` with next action `i+1`, while contiguous trajectory-local splits prevent boundary leakage. The deployed map is
+
+\[
+G(\alpha_F,\alpha_U,\alpha_L)=(-\alpha_F,-\alpha_L,-\alpha_U),
+\qquad
+\alpha_F(x)=\frac{h_F(x)-h_F(Gx)}{2},\quad
+\alpha_U(x)=h_R(x),\quad
+\alpha_L(x)=-h_R(Gx).
+\]
+
+The map imposes reflection symmetry at deployment; it does not show PPO equivariance. Stage-3 CFD, finiteness, action/order/schema compliance and provenance decide whether the formula is executable. [SR pipeline; Legacy interface]
+
+The symmetry map fixes the action space; the canonical expression gives the resulting structure.
+
+### 3. Canonical law and physical translation
+
+The mapped-shared topology A is
+
+\[
+\boxed{\alpha_F=\operatorname{odd}\!\left(-0.381391\,C_{d,\mathrm{rear},a}\right)},
+\qquad
+\boxed{\alpha_U=1.307782\,C_{l,\mathrm{rear},s}-3.431209},
+\qquad
+\boxed{\alpha_L=-\alpha_U(Gx)}.
+\]
+
+Here `Cd_rear,a=(Cd_upper-Cd_lower)/2` and `Cl_rear,s=(Cl_upper+Cl_lower)/2`. The rear law therefore combines a persistent negative upper-action constant with rear-mean-lift feedback, while the front action is an odd-projected rear drag-asymmetry correction. Because `alpha=omega/U0`, the constant corresponds to opposite-sign rear rotations of approximately `3.431209 U0`; the wording describes executable action structure, not a unique physical mechanism. The separate steady analysis supplies only a brief contextual bridge: its disturbance-free deployments are consistent with the order of this rear-rotation magnitude and are owned in detail by Module 08. [SR claims; canonical formula authority; steady context]
+
+The formula’s acceptance must still be stated with the correct comparator semantics.
+
+### 4. Closed-loop arbitration and comparator boundary
+
+The formula-parent evidence is a Legacy SR deployment against the declared Kármán target using the native Legacy DTW contract; it should not be described as though every validation were a same-window PPO/SR/zero acquisition. The reported canonical SR similarities for the 200-control-step named cases are `0.9543, 0.9427, 0.8560, 0.7827` at re-codes `50,100,200,400`; the frozen law also has finite 400-control-step observations `0.952178, 0.944339, 0.863850, 0.833365`. These are formula-deployment results at named cases and windows. Separately, the standardized collector defines comparable `Target`, `PPO`, `SR` and `Zero` roles using 480 warm-up intervals and 160 retained boundaries; exact PPO/SR/zero comparator numbers are not asserted here unless the current role artifacts and proposed panel supply them. The native DTW lag is alignment metadata, not a physical delay, and neither DTW result alone is field matching. [SR claims; Stage-3 runs; standardized Legacy acquisition]
+
+Thus the law is supported as a finite executable surrogate, while the direct role comparison remains a distinct visualization/data dependency.
+
+### 5. Fixed-window term tests and inference ceiling
+
+Parent-relative term deletion keeps the canonical topology, coefficients, Legacy solver/geometry, four named Kármán cases and native DTW window fixed. Removing the rear constant gives the largest listed degradation across re-codes 50/100/200/400 (`ΔDTW≈−0.1065,−0.0939,−0.1785,−0.1911`); removing rear-lift feedback gives smaller but non-zero changes (`≈−0.0288,−0.0417,−0.0614,−0.0489`); the tested front correction has small mixed changes (`≈−0.0065,+0.0006,−0.0052,−0.0252`). The bounded interpretation is therefore a largest-tested persistent rear contribution, secondary rear feedback and weak tested front correction under one fixed short window. These are deletion tests, not causal ablations: they do not establish term necessity, sufficiency, mechanism, uniqueness or universal sensitivity. Illusion SR remains historical/negative and supplies no positive transfer. [SR deletion table; claim ledger]
+
+This action-space ranking exports a narrower field question to Module 06: how do the persistent and smaller dynamic components relate to independently defined field estimands?
+
+## Layer 2 — paragraph notes and claim–evidence ledger
+
+### Paragraph notes
+
+| Paragraph | Claim/role | Exact evidence identity | Chain; solver/geometry | Case/seed/role; Re | Comparator; metric/estimand; window | Status and audit state |
+|---|---|---|---|---|---|---|
+| 1 | SR is an observation-to-action surrogate; offline fit is not acceptance | `src/SR_analysis/README.md`, `CLAIMS.md`, `PIPELINE.md`, `results/README.md`; `tables/offline_action_rmse.md` | LegacyCelerisLab; Legacy Kármán plant; action `front,upper,lower` | `karman_re50/100/200/400`; re-code uses `2D`, so `Re_D=re_code/2`; 197 causally aligned PPO rows/case | PPO-visited next-action RMSE: `1.4435,1.0895,2.2860,2.1699`; diagnostic only; no closed-loop window | Current authority; displayed RMSE is derived view; artifact-audit deferred, non-blocking |
+| 2 | Recovery and deployment contract | `stage_1_infer.py`, `stage_2_fit.py`, `stage_3_validate.py`; `utils/data_contracts.py`, `feature_builder.py`, `g_operator.py`, `formula_schema.py`, `metrics.py`, `provenance.py`; `src/drl_pinball/legacy_test/README.md` | LegacyCelerisLab; Legacy solver/geometry; native action order | Four named Kármán cases; Stage 1 post-state `i`→action `i+1`; formula deployment under `G` | Stage-3 short gate `40` control steps; Stage-3 standard validation `200` control steps; standardized collector `480` warm-up + `160` retained boundaries | Current contract documentation; exact leaf recomputation artifact-audit deferred, non-blocking |
+| 3 | Canonical law and action translation | `article-refit-karman-topology-a-20260718`; `results/catalog.json`; `src/SR_analysis/CLAIMS.md` | LegacyCelerisLab; Legacy Kármán plant | re-code `50/100/200/400`; canonical mapped-shared topology A; `alpha=omega/U0` | Formula is evaluated in closed loop against declared Kármán target; native Legacy DTW where quoted; steady only contextual | Current documented formula identity; formula leaf hash audit deferred, non-blocking; steady details owned by Module 08 |
+| 4 | Formula-parent deployment is distinct from standardized role acquisition | `article-L2-karman-20260718`, `article-L3-karman-20260718`, `article2-long-karman-20260720`; `src/drl_pinball/legacy_test/acquire.py`; `src/SR_analysis/CLAIMS.md` | LegacyCelerisLab; Legacy solver/geometry | Named Kármán cases; formula-parent SR role for July runs; standardized collector roles Target/PPO/SR/Zero | July native `legacy_dtw_v1_abs_n_unclipped` / `legacy_reference_cycle_vs_last_recorded_cycle`; 200 control steps: `0.9543,0.9427,0.8560,0.7827`; 400: `0.952178,0.944339,0.863850,0.833365`; standardized window `480+160` | Current finite/pointwise formula evidence; direct PPO/SR/zero panel numbers are a future dependency if claimed; artifact-audit deferred |
+| 5 | Rear constant ranks largest in fixed-window deletion tests | `article-ablation-formulas-v2-20260718`; `article-ablation-L2-k_front0-20260718`; `article-ablation-L2-k_rear0-20260718`; `article-ablation-L2-k_rear1-20260718`; `article-scaling-L1-v2-20260718`; derived `article2-plotting-package-20260721/tables/term_deletion.md` | LegacyCelerisLab; Legacy solver/geometry | `karman_re50/100/200/400`; canonical parent vs named deletion variants | Same parent-relative native DTW; 40 control steps; rear constant deltas `−0.1065,−0.0939,−0.1785,−0.1911`; rear feedback `−0.0288,−0.0417,−0.0614,−0.0489`; front `−0.0065,+0.0006,−0.0052,−0.0252` | Current bounded ranking; deletion is not causal ablation/necessity; parent/deletion artifact audit deferred |
+
+### Exact canonical formula and symmetry identity
+
+- Formula run: `article-refit-karman-topology-a-20260718` (documented current authority; leaf hash recomputation `artifact-audit deferred`).
+- Native action order: `(front, upper, lower)`.
+- Reflection/action map:
+
+\[
+G(\alpha_F,\alpha_U,\alpha_L)=(-\alpha_F,-\alpha_L,-\alpha_U),
+\quad
+\alpha_F(x)=\frac{h_F(x)-h_F(Gx)}{2},
+\quad
+\alpha_U(x)=h_R(x),
+\quad
+\alpha_L(x)=-h_R(Gx).
+\]
+
+- Canonical deployed law:
+
+\[
+\alpha_F=\operatorname{odd}(-0.381391 C_{d,\mathrm{rear},a}),
+\quad
+\alpha_U=1.307782 C_{l,\mathrm{rear},s}-3.431209,
+\quad
+\alpha_L=-\alpha_U(Gx).
+\]
+
+- `alpha=omega/U0`; rear constant translates to opposite-sign rear physical angular velocities of magnitude approximately `3.431209 U0`. Exact coefficient/formula/source/deployment hashes are bound by current authority/manifests but were not recomputed in this documentation pass: `artifact-audit deferred`, not a blocker.
+
+### Comparator and metric ledger
+
+- **Formula-parent July evidence:** canonical formula deployment against the declared Legacy Kármán target; native Legacy DTW `legacy_dtw_v1_abs_n_unclipped` / `legacy_reference_cycle_vs_last_recorded_cycle`; Stage-3 short gate `40` control steps, Stage-3 standard validation `200`, finite extension `400`; named re-codes `50/100/200/400`; finite/pointwise current evidence.
+- **Standardized August role acquisition:** separately reruns `Target`, `PPO`, `SR` and `Zero` under one Legacy contract, with `480` warm-up intervals and `160` retained post-step boundaries, independent role phase and complete-cycle outputs. It is the appropriate source for a same-protocol PPO/SR/zero comparison. Exact comparator numbers are not claimed in Layer 1 because the current SR authority read here does not expose a complete PPO/SR/zero table for the proposed visual.
+- **Spatial field matching:** Module 04 owns field matching, with spatial L2 primary and DTW complementary/case-specific. The SR native DTW numbers above must not be relabelled as spatial L2 or as field equivalence.
+- **Deletion:** parent-relative formula variants, same Legacy Kármán cases and same 40-control-step native-DTW window. It ranks tested terms; it does not establish causal effect or necessity.
+
+### Claim ladder
+
+**Green — documented/current:** the exact mapped-shared formula and imposed deployment symmetry are documented by current SR authority; the formula-parent Kármán deployments have the named finite DTW observations; fixed-window deletion ranks the rear constant largest among tested terms, with smaller non-zero rear feedback and weak/mixed tested front correction.
+
+**Amber — bounded interpretation:** the law can be organized as a Kármán/cloaking control skeleton with a persistent rear component and smaller observation-dependent corrections; 400-step and named-condition runs are finite/pointwise evidence only.
+
+**Red — prohibited without new evidence:** necessity, sufficiency, causal ablation, mechanism, uniqueness, global optimality, universal Reynolds behavior, DTW as physical delay, SR action fit as field matching, or positive Illusion SR transfer/target tracking/generalization.
+
+## Figure, equation and table contracts
+
+- **Canonical law equation:** defines the deployed surrogate and action signs; it does not establish a governing flow law, mechanism, uniqueness or PPO equivariance.
+- **Symmetry schematic:** makes native order and imposed `G` mapping auditable; it does not prove plant symmetry or training equivariance.
+- **Compact Stage-3 arbitration panel:** separates short gate, standard validation and finite extension from spatial L2 field matching; failed formula candidates remain visible. It does not imply PPO/SR/zero role equivalence.
+- **PPO/SR/zero trace/table:** if included, must use standardized same-protocol role artifacts and state case, role, window, phase and metric; exact comparator data are currently a future panel dependency, not invented here.
+- **Deletion table:** reports parent, variant, deleted term, native DTW, delta, case and 40-step window; caption must say parent-relative term deletion, not causal ablation.
+
+## Imports, exports and bridge
+
+**Imports, provisional pending coordinator audit:** Module 03 plant/action/Re/role/metric definitions and Module 04 Kármán performance/comparator framing. This section imports no modern V5, CCD, OID or positive Illusion SR evidence.
+
+**Exports:** canonical law identity; exact symmetry map and action order/sign; distinction between formula-parent SR deployment and standardized PPO/SR/zero acquisition; finite 200/400 DTW observations; and fixed-window deletion ranking. Module 06 receives the field-accounting question without inheriting a causal interpretation.
+
+**Steady bridge:** the separate steady chain provides contextual magnitude correspondence for the rear constant; detailed steady sweep and theory interpretation are exported to Module 08.
+
+## Coverage, conflicts and deferrals
+
+| Coverage status | Sources/reason | Evidence status |
+|---|---|---|
+| `read` | Revised Round-3 rule/master prompt/quality gate; Round-2 freeze/style guide; Round-1 SR, DRL and coverage dossiers; current SR README/CLAIMS/HANDOFF/PIPELINE/results README/INDEX/catalog; Legacy interface README; SR deletion/RMSE/steady/per-case reader views | Read-full reader authorities; not independently verified or artifact-audited |
+| `mapped` | Archive and historical Illusion navigation, used only for boundary status; current formula/role leaf families identified through authority/catalog | Indexed/mapped is not read or independently verified |
+| `not-covered` | No new literature or CFD; no exhaustive leaf hash recomputation; no complete standardized PPO/SR/zero table asserted | Literature cannot establish internal results; formula identity remains current authority documentation |
+| `external-inaccessible` | External/symlink-backed standardized role payloads and hidden raw fields are not exposed in the normal workspace | Authority-led `artifact-audit deferred`; not a blocker unless needed for a proposed comparator/panel or conflict |
+
+**Conflict/gap:** no authority conflict is currently identified for the canonical Kármán law or deletion ranking. Exact Module 03/04 interface reconciliation remains a coordinator task. A proposed PPO/SR/zero visual additionally requires the complete standardized role comparator/panel data; this module does not invent them.
+
+## Revised-gate self-audit
+
+- **Two layers:** passed structurally; Layer 1 uses short source tags, Layer 2 carries exact provenance and numbers.
+- **Narrative cadence:** five shorter paragraphs, with transitions and local caveats; steady detail removed from the narrative.
+- **Validation vocabulary:** uses Stage-3 short gate/standard validation; spatial L2 remains Module 04’s field metric.
+- **Comparator boundary:** formula-parent SR deployment is separated from standardized Target/PPO/SR/Zero acquisition; missing direct role numbers are marked as a future panel dependency.
+- **Claim discipline:** green wording follows documented authority; hash/leaf audit is deferred, not a blocker; Illusion SR is historical/negative.
+- **Scope/interfaces:** steady is a one-sentence bridge to Module 08; field/CCD interpretation is exported to Module 06; Module 03/04 imports remain provisional.
+- **Status:** revised candidate for second coordinator audit; not self-declared ready.

docs/JFM_WYQ/ROUND3_BOTTOM_UP/06_KARMAN_FIELD_CCD.md

@@ -0,0 +1,111 @@
+# Round 3 revised candidate — Section 6: Kármán correction fields and action-correlated modes
+
+**Status:** REVISE completed; candidate for second audit, not ready.
+
+## Section question and bounded answer
+
+Under the exact Legacy Kármán field-error estimand, what share of the declared zero-to-DRL improvement is associated with the constant reference, what smaller residual remains, and what do CCD modes organize?
+
+Under the stated Legacy ROI, common mask and mean-field window, the constant mean accounts for `93.42%` of the measured zero-to-DRL target-error reduction, leaving a `6.58%` constant-to-DRL increment. A separately centred, non-paired residual is phase-coherent and descriptively organized into **action-correlated correction-field modes**; in the preferred pinball-inclusive derived sibling, raw weighted POD spans essentially the same rank-3 field subspace. These are scalar estimand and descriptive association statements, not energy or explained-variance partitions, paired counterfactuals, causal effects, response times, mechanisms or CCD superiority.
+
+## Layer 1 — six concise near-prose paragraphs
+
+### 1. Four roles define the comparison
+
+The field comparison starts with four roles: the target mean, passive zero-action mean, constant-mean control and DRL control. In the Legacy `karman_re100` case, with `Re_D=Re_code/2` under the Legacy `2D` reference and body order `(front, upper, lower)`, all four means are compared on the inclusive wake ROI `34 <= x/D <= 54`, `|y/D| <= 5`, using the exact intersection of their solver-fluid masks and a coordinate-weighted velocity error relative to the target mean. The constant arm is therefore a reference comparator in one declared estimand, not a paired intervention. [M03; CCD authority]
+
+Although Module 04's standardized PPO errors and this corrected CCD ledger have the same nominal x extent and numerically close values, they use different role acquisitions and role sets, retained windows, a geometry guard versus an exact four-role common mask, and an unweighted `U0`-normalized metric versus a coordinate-weighted metric on already nondimensional velocity. They are different estimands: their values are never combined or described as replication, validation, agreement or triangulation. [M04–M06 reconciliation]
+
+This fixes the denominator before the fields or modes are interpreted: the comparison is Legacy-specific and must not be pooled with the V5 contract or with steady, transient or Illusion metrics.
+
+### 2. The mean fields show where the measured reduction occurs
+
+The target error is zero by construction. The corresponding ROI mean errors are `E_zero=0.4694352273`, `E_constant=0.1110140739` and `E_DRL=0.0857787860`. The mean-field comparison therefore shows a large zero-to-constant change and a smaller constant-to-DRL adjustment under the same target comparator. The statement is about registered mean-field error only: it does not identify an actuator cause, establish feedback necessity or close a momentum or energy budget. [CCD ROI authority]
+
+The mean-field panel and scalar ledger should precede the residual and mode panels, so that the dynamic analysis answers a narrower follow-up question rather than replacing the primary field result.
+
+### 3. The 93.42/6.58 values are scalar error-difference shares
+
+The proportions are defined from the ordered scalar differences, not from field energy. With `E_0`, `E_C` and `E_D` denoting the three errors above,
+\[
+S_C=100\frac{E_0-E_C}{E_0-E_D}=93.42\%,
+\qquad
+S_D=100\frac{E_C-E_D}{E_0-E_D}=6.58\% .
+\]
+Thus the precise wording is that the constant mean accounts for `93.42%` of the measured zero-to-DRL ROI target-error reduction, while the remaining DRL increment is `6.58%`. The denominator is the scalar error difference `E_0-E_D`; these percentages are not explained variance, energy share, causal contribution or term attribution. [CCD ROI authority; M03 metric boundary]
+
+The ledger consequently motivates a reference-residual analysis, but does not turn the residual into an isolated physical response.
+
+### 4. Separate centring leaves a non-paired phase-conditioned residual
+
+The corrected CCD uses `19` retained DRL cycles and `10` phase bins (`N=190`). The constant phase template is its own 19-cycle ensemble mean, and each role is centred over its own retained ensemble before subtraction:
+\[
+U_{c,b}=(q^D_{c,b}-\bar q_D)-(\hat q^C_b-\bar q_C).
+\]
+The DRL and constant cycles are not paired. The construction uses no cross-role cycle pairing, Fourier resampling, lag, wrap or temporal-PSD interpretation. It removes the constant-template backbone in the declared reference sense and supplies a descriptive phase-conditioned field for association with the same-boundary DRL effective action; it does not isolate a causal DRL field, additive force or actuator response. [CCD math authority]
+
+The residual is therefore the input to organization analysis, not a counterfactual claim about what DRL changed in a matched realization.
+
+### 5. CCD organizes residual fields by action association and phase
+
+Weighted CCD relates the centred residual snapshots to the centred same-boundary DRL effective-action observable. Its supported output is a rank-3 descriptive field subspace whose action-side coordinates are presented as `front`, `rear-symmetric` and `rear-antisymmetric`, with phase reconstructions used to show how the residual is organized over the retained periodic coordinate. The physical output term is **action-correlated correction-field modes**. Mode signs are globally arbitrary, and body-local rotation or parity displays are diagnostics rather than wall-vorticity or mechanism measurements. [CCD result/figure authority]
+
+This is complementary to, not a derivation of, the SR law. Module 05’s rear-constant and rear-feedback ranking is an observation-to-action result under its own deletion and DTW contracts; CCD action coordinates are field/action association coordinates under the residual contract. The two estimands may be compared only as a brief, descriptive bridge.
+
+### 6. Raw POD sets the boundary on CCD’s added field information
+
+For the same centred residual object in the pinball-inclusive derived sibling, raw full-fit weighted POD is computed without using the action observable to choose its basis. The leading rank-3 principal cosines are approximately `1`, so CCD and POD span essentially the same tested field subspace. CCD’s bounded added information is therefore the action-coordinate association and phase organization of that subspace, not a distinct or superior field basis. This remains one corrected parent lineage and does not establish causality, response time, mechanism, generalization or CCD superiority. [CCD derived publication]
+
+The section exports a mean-dominant field ledger plus a smaller descriptive residual organization; later synthesis must preserve the separate SR, steady, V5 and OID boundaries.
+
+## Layer 2 — compact provenance and diagnostics ledger
+
+| Claim or object | Exact provenance and identity | Contract: chain, solver/geometry, case/role, Re | Comparator, metric/estimand, ROI/window | Status and audit state |
+|---|---|---|---|---|
+| Four-role mean comparison | `src/CCD_analysis/RESULTS_INDEX.md`; `FINAL_RESULTS.md`; ROI result schema `ccd-karman-roi-mean/v2`; manifest `f7f6141042678e52314a4d8f76d589108da8e844c8ce504f8f523a6f88a17e73` | LegacyCelerisLab; Legacy fluidic pinball; `karman_re100`; target, zero, constant mean, DRL; `Re_code=100`, `Re_D=50` | Coordinate-weighted RMS target error of retained role means; exact four-role common solver-fluid mask; inclusive `34<=x/D<=54`, `|y/D|<=5`; retained mean window | Current authority; artifact-audit deferred because authority binds result and manifest without conflict |
+| `E_zero`, `E_constant`, `E_DRL` | Same ROI-mean result and `FINAL_RESULTS.md` | Same Legacy contract; zero, constant mean and DRL roles | `0.4694352273`, `0.1110140739`, `0.0857787860`; target error is zero by construction | Current sourced values; derived share arithmetic is bounded and not independently triangulated |
+| `93.42%` and `6.58%` | Derived scalar ledger from the three current ROI errors; definitions in `MATHEMATICAL_DERIVATION.md` | Same Legacy contract | `(E_0-E_C)/(E_0-E_D)` and `(E_C-E_D)/(E_0-E_D)`; scalar error differences only | Bounded current interpretation; not energy, explained variance, causality or term share |
+| Residual definition | `MATHEMATICAL_DERIVATION.md`; `CORRECTED_MATH_CONTRACT.md`; corrected result schema `ccd-karman-cycle-template/v1`; manifest `b114fbcc5f821034829ca3771ac5ba13df4b75df312b0615b662f81a9e5737d8` | LegacyCelerisLab; Legacy `karman_re100`; 19 DRL cycles, 19 constant cycles, 10 bins; `N=190`, `Q=1`; DRL and constant roles not paired | Separately centred `U_(c,b)` against constant 19-cycle phase template; corrected wake-only field dimension `M=160400`; retained phase/cycle window | Current validated wake-only sibling; artifact-audit deferred, nonblocking |
+| CCD organization | Corrected wake-only result above; preferred derived pinball result `pinball-domain-20260809T1925+0800`, schema `ccd-karman-pinball-domain/v1`, result manifest `7e107a895364ed3e34abf2be513a3ce9e12736766ec6383c5b831891802d5828` | Same corrected Legacy parent lineage; pinball-inclusive domain `29<=x/D<=54`, `|y/D|<=5`; same `19×10` samples and action observable | Weighted CCD action association; descriptive phase reconstructions; coordinates `front`, `rear-symmetric`, `rear-antisymmetric` | Preferred derived result; not independent triangulation or new CFD |
+| Reader-facing mode panels | Preferred derived publication `publication-pinball-domain-20260809T1925+0800`; publication manifest `31888f8eeb6c70dd9f0ff3449fa879f44abf381338ec261b0e45f69603691fdd`; `01`–`04` figure groups | Derived from corrected parent; pinball-inclusive reader domain; plot arrays use `(x index,y index)` | `01_domain_estimand` precedes `02_ccd_mode_identity`, `03_phase_reconstruction`, `04_raw_pod_vs_ccd`; fixed partitions are registered coordinates | Preferred reader view; source arrays remain authority-led `artifact-audit deferred`, nonblocking while documented result/panel claims remain bound |
+| CCD diagnostics moved out of Layer 1 | `RESULTS_INDEX.md`, `FINAL_RESULTS.md`, `EXECUTION_CHECKPOINT.json` | Same Legacy corrected lineage; wake-only `M=160400`; pinball sibling `M=198528` | Wake strengths `0.0174141920`, `0.0069694895`, `0.0010500356`; least-favourable rank-3 projector cosine `0.9947903592`; leave-one-cycle-out leading cosine `0.9997247674`; pinball principal cosines `0.9999616609`, `0.9998971005`, `0.9976439804`; minimum declared pinball stability cosine `0.9974651867` | Descriptive diagnostics; singular strengths are not energy or explained variance; no independent triangulation |
+| Array and sign safeguards | `src/CCD_analysis/FIGURES_AND_DATA.md`; `src/pv_plot/README.md`; `src/OID_analysis/PHYSICS_CONTRACT.md` | Legacy CCD stored field orientation is `txy` (`T,NX,NY`); modern V5 evaluator orientation is `tyx` (`T,NY,NX`); body order `(front, upper, lower)` | Publication arrays `(x,y)` are displayed with transpose; action sign conflict in OID material remains unresolved and is not imported as CCD evidence | Current boundary; no silent transpose or sign reconciliation |
+| SR and performance bridge | Module 05 `05_KARMAN_SR_LAW.md`, especially its exports and deletion boundary; Module 03 `03_CONTRACTS_METRICS.md`; Module 04 standardized PPO mean/phase8 contract | Legacy SR only; executable observation-to-action surrogate, `alpha=omega/U0`, `Re_D=re_code/2`; CCD same Legacy case but distinct residual/action estimand | SR uses formula deployment, native DTW and parent-relative fixed-window deletions; CCD uses weighted field/action association under M06's corrected four-role coordinate-weighted mean-field estimand. M04 standardized PPO `E_mean` and phase8 are separate estimands and cannot validate or numerically close the M06 ledger | Reconciled interface; no SR derivation or numerical SR result duplicated here. Module 05’s direct standardized PPO/SR/zero panel remains a separate dependency if later proposed |
+| OID boundary | `src/OID_analysis/README.md`, `SCIENTIFIC_RESET.md`, `PHYSICS_CONTRACT.md`, `HANDOFF.md` | Active OID method core is claim-free; archived Legacy material remains historical/conditional | No active positive OID result, action importance, mechanism or control-authority comparison | Claim-free boundary; unresolved historical sign conflict retained |
+
+**Source versus derived identity.** The validated wake-only corrected sibling is the primary CCD result authority (`cycle-template-ccd-review-fix-per-cycle-20260808T1736+0800`). The pinball-inclusive result and `publication-pinball-domain-20260809T1925+0800` are preferred derived result and reader publication siblings. The latter do not replace the former, and their near-equivalence does not constitute independent triangulation. Canonical alias/promotion choice is not needed for the manuscript-facing wording because both exact identities are stated.
+
+## Figures, equations and table contracts
+
+- **Mean-field figure and scalar ledger, first.** Shows target/zero/constant/DRL means and the exact error-difference ledger. It cannot establish causality, energy share, mechanism or necessity.
+- **Residual definition/equation.** Defines the separately centred, non-paired phase-conditioned residual. It cannot establish a paired counterfactual, additive force or lag.
+- **Mode identity and phase reconstruction.** Shows action-correlated correction-field modes, action coordinates and phase organization. It cannot establish wall vorticity, action sign authority, response time or mechanism; the rotation proxy is visualization-only.
+- **Raw POD comparison.** Shows near-equivalent rank-3 subspaces in the derived pinball-inclusive object. It cannot establish CCD superiority or explained variance from CCD singular strengths.
+
+## Claim ladder
+
+**Green:** the exact Legacy mean errors, the scalar difference-share definition, the corrected non-paired residual contract, and the documented rank-3/action-coordinate organization under the named artifacts.
+
+**Amber:** the constant mean accounts for most measured reduction; the residual is phase-coherent and organized descriptively by action-correlated correction-field modes; CCD adds action-coordinate interpretation to a POD-equivalent tested field subspace.
+
+**Red without new evidence:** causal response, paired counterfactual, mechanism, response time, momentum restoration, energy/explained variance, term necessity, unique physical modes, CCD superiority, generalization, uncertainty, positive OID, or transfer to V5/steady/Illusion.
+
+## Interfaces, coverage and genuine blockers
+
+**Imports reconciled.** Module 03 fixes the Legacy/V5 separation, Legacy body/action/Re conventions, role semantics and spatial-L2/DTW distinction. Module 05 exports the Legacy SR law, its fixed-window deletion boundary and a brief control-skeleton context. The only interface mismatch requiring parent awareness is scope, not evidence: Module 05 owns the observation-to-action law and term ranking, whereas this module owns M06's corrected four-role coordinate-weighted mean-field estimand and residual organization. The SR law and CCD action coordinates are different estimands and must not be equated. Module 04's standardized PPO `E_mean` and phase8 are also separate estimands and cannot validate or numerically close the M06 ledger.
+
+**Exports.** Exact mean errors and `93.42/6.58` semantics; Legacy ROI/mask/window; non-paired `19×10` residual; rank-3 and action-coordinate identity; source-versus-derived lineage; POD boundary; and the bridge that the residual is descriptively action-correlated. No positive OID export.
+
+**Coverage ledger.** `read`: revised Round3 rule/master/quality; current Modules 03 and 05; Round2 freeze/style; Round1 field/CCD/OID and coverage dossiers; current CCD README, handoff, results index, derivation, claims, figures/data, final results, contracts, checkpoint and catalog; active OID reset/contract docs; current SR claims/handoff. `mapped`: current authority references to external dense result/publication parents and their reader copies; mapping is not read or independently verified. `not-covered`: raw dense arrays and universal leaf-hash recomputation, deferred because current authorities bind exact roles, results and publication identities. `external-inaccessible`: external dense payloads not exposed in the normal workspace; this is an authority-led `artifact-audit deferred` status, not a blocker.
+
+**Genuine blockers.** None for the bounded wording. A blocker would arise only if a proposed figure could not be sourced/redrawn from the documented publication/data contract, or if a source conflict changed the role, comparator, window or claim. The OID sign conflict remains quarantined and does not affect this CCD claim ceiling. Canonical alias/promotion remains outside the wording requirement and is nonblocking.
+
+## Revised-candidate audit
+
+- Layer 1 now contains six concise near-prose paragraphs with short source tags and no inline evidence-contract blocks.
+- Exact hashes, dimensions, singular strengths, stability diagnostics, orientation and source/derived identities are confined to Layer 2 or compact contracts.
+- Module 03/05 coverage is corrected to `read` and reconciled; SR derivation and detailed deletion results are not duplicated.
+- External artifact audit and canonical promotion are recorded as nonblocking deferrals, not blockers.
+- The `93.42/6.58`, non-pairing, descriptive CCD, POD, OID and `tyx/txy` semantics remain explicit.
+
+**Disposition:** revised candidate for second audit; not ready.

docs/JFM_WYQ/ROUND3_BOTTOM_UP/07_CROSS_CASE_EXTENSIONS.md

@@ -0,0 +1,137 @@
+# Round 3 — Cross-case extensions beyond periodic cloaking
+
+**Coordinator disposition — PASS, ready for downstream import but not final prose.** Vortex event-L2 and Erase quantitative panels remain blocked/conditional on the exact dependencies recorded below. This module asks what changes when the periodic Kármán cloaking task is made transient, when the declared target is non-zero, and when the objective is changed again to erasure. It preserves the two-layer architecture: Layer 1 is an author-readable section map; Layer 2 is the local provenance and audit ledger.
+
+## Section question and bounded answer
+
+**Question.** What changes when periodic Kármán cloaking is extended to a nonperiodic incoming vortex and when the objective changes from cloaking to illusion or erase?
+
+**Bounded answer.** Named PPO demonstrations support two distinct extensions: event-relative control beyond periodic phase for comparable-scale vortex disturbances, and retargeting or erasure beyond background matching under their own case-specific roles and metrics. They do not establish arbitrary-flow universality, transfer of the Kármán control skeleton, positive Illusion SR, or one common efficacy score.
+
+## Layer 1 — concise near-prose section draft
+
+### 1. Transition: the task changes in two independent ways
+
+The cross-case comparison should change one contract dimension at a time. Kármán cloaking compares a periodic disturbance with a declared background/reference; the Vortex cases remove the usable cycle coordinate and replace it with an incoming-event coordinate, whereas Illusion keeps a periodic target construction but replaces the near-zero/background objective with a non-zero target signature. Erase changes the reference once more: the desired state is a declared clean or erased signature rather than a Kármán target. The action interface and sparse wake-signature observations remain useful descriptors, but the comparison is no longer a common score. This is a transition from phase-indexed wake-signature matching to event-relative or target-field matching, not a generalized theory of control. [R1-DRL/cases; R1-metrics; R2-freeze; M03]
+
+**Fixed/change ledger for the manuscript.** Keep the Legacy scene and its action/normalization conventions attached to the named demonstrations; keep force and sparse downstream sensing as execution context; and keep spatial L2 primary whenever a valid field role exists. Change the temporal coordinate from complete-cycle phase to event-relative acquisition for Vortex, and change the target from background/near-zero to a declared non-zero field/signature for Illusion. For Erase, change the comparator and target semantics again. Do not import Kármán SR or CCD modes as interpretations of Vortex, Illusion or Erase: Modules 05–06 provide comparison context only. [M03; M05; M06]
+
+### 2. Kármán → Vortex: event-relative extension without arbitrary-flow generality
+
+The Vortex demonstration is best read as a pulse-response extension of the Legacy Kármán execution interface. The current case labels are `vortex_lamb` and `vortex_taylor`, with acquired variants including `vortex_lamb_y000`, `vortex_taylor_ym2L`, `vortex_taylor_ym1L`, `vortex_taylor_y000`, `vortex_taylor_yp1L` and `vortex_taylor_yp2L`; the action scale is the Vortex-specific `\(\omega/U_0=4a+[0,-4,+4]\)`, and the Legacy acquisition is transient with `SI=800` and `MAX_STEPS=150`. These labels and offsets must remain literal. The event artifact is `event_fields.npz`, whose `relative_offsets` are not periodic phase slots; the current plot navigation exposes `relative_offsets=[-10,-5,0,5,10]` around the controlled-role canonical event peak. These are event-relative sample offsets, not periodic phases. Layer 1 therefore retains only the named visual capability and pulse-like action interpretation; any quantitative event comparison is conditional on a later role audit and offline evaluator. [R1-DRL/case matrix; Legacy DATA_DICTIONARY.md; reproduction-plot manifest]
+
+A future quantitative statistic would be a downstream-region event-snapshot L2 at each named event-relative offset, but no quantitative Vortex L2 belongs in Layer 1 now: current `wake_l2.py` does not support `event_fields.npz`. Computing it would require new offline code plus audited target, controlled and physical-zero event roles, target identity, common ROI/mask and event window. The existing/new DTW remains complementary and must retain the Vortex contract: the current Legacy implementation has no lag and rolls the final window by `current_step+1`; it is a signal comparison, not physical delay or field equivalence. If those exact roles are absent, the visual capability remains a bounded observation and the event-L2 result is a drafting gap, not a number inferred from a plot. [R1-metrics; legacy_test/core/dtw_metrics.py; wake_l2.py boundary]
+
+Pulse-like actions make a bounded comparable-scale hypothesis plausible: an incoming Lamb or Taylor disturbance can be met by transient actuation in the named scenes. They do not show that arbitrary incoming flow, arbitrary amplitude, arbitrary offset or unseen vortex family is controlled, and they do not transfer the periodic Kármán action skeleton. The relevant next question is whether matched event acquisitions across disturbance scale and offset preserve the event-relative field comparison—not whether one transient panel establishes universality. [R1-DRL/cases; R2-freeze]
+
+### 3. Cloak → Illusion: a non-zero target changes both reward reading and field evaluation
+
+Illusion should be defined locally as matching a declared non-zero target signature or field, not as reducing a wake toward the physical zero. The target is a separately constructed periodic target-cylinder signature, including target force/harmonic information and the target sensor trajectory; the controlled pinball trajectory is evaluated against that target, with physical zero retained only as a separate comparator where its exact role is acquired. The named historical V5 PPO demonstration `ill_075L` (retained seed 43) is suitable as the main-text capability example. Its documented main-text evidence is capability-level reward/force/DTW and visual material; an independent field-L2 claim or panel is conditional on current field-role authority and should not be implied here. The Legacy `article-joint-data-illusion-20260718` manifest is a separate historical target-construction authority and must not be used to identify or numerically continue the V5 run; the retained size points remain case-specific demonstrations. Their reward/DTW and force behavior show capability and difficulty, not a cross-size law. [V5 env_illusion.py/training authority; article-joint-data-illusion-20260718 manifest; DRL claim ledger]
+
+For a periodic Illusion field panel, report complete-cycle mean L2 and phase8 L2 separately, with target, controlled and physical-zero roles explicitly separated; add normalized DTW only under the exact role, channel-normalization and window contract. Mean L2 asks about the persistent target field, phase8 asks about independently phased nearest snapshots, and normalized DTW asks about the declared sensor/target signal. They are related diagnostics within an Illusion acquisition only if the role set and windows match; they are not one efficacy score and cannot be pooled with Kármán, Vortex event L2, M06’s corrected four-role mean ledger, or steady profiles. Apply the near-collision gate to any similar L2/DTW values: compare acquisition, role set, ROI/domain, mask, weighting/normalization and window before using the values, and classify them as the same estimand, a related diagnostic or a different estimand. [M03; M04; Illusion manifests; near-collision interface]
+
+The Illusion PPO capability should be paired locally with the negative SR boundary, without treating the two as a direct comparator. In a different Legacy chain and role acquisition, the standardized `illusion_1L` SR result was effectively equal to physical zero over the declared long window. This is historical/negative evidence against promoting Illusion SR, while the V5 `ill_075L` PPO result remains a separate named historical capability demonstration. [SR CLAIMS.md; standardized Illusion role authority]
+
+### 4. Illusion → Erase: objective extension, failure envelope and bridge
+
+Erase changes the target/reference again. A defensible Erase result would define the clean or declared erased mean field and compare controlled and physical-zero roles to that target using mean-field L2, with DTW complementary only where the exact target, control and zero streams, normalization and window are documented. The historical Legacy route is explicitly unsupported: its evolving controlled-force FIFO was used as a target and no exact independent target-scale mapping exists. The plot manifest’s `erase` entries therefore do not authorize exact roles merely because filenames are present, nor do they revive unsupported Legacy erase evidence. If an authority later binds exact role artifacts, the figure can include an Erase mean field and case-specific DTW; until then the result is a gap/visual-navigation boundary, not quantitative efficacy. [Legacy legacy_test/README.md; R1-metrics; reproduction-plot manifest]
+
+The force-centred extension is useful but deliberately approximate. Near-zero resultant force motivates cloak as background suppression; target-force matching motivates Illusion and Erase at the observer/reward level. This does not imply that force matching determines the downstream field, and unresolved small structures and higher-Reynolds-number effects bound the approximation. The cross-case lesson is thus about objective and evaluation contracts: the same broad force-plus-sparse-observation execution can be retargeted, but neither the Kármán control skeleton nor a common L2/DTW efficacy score follows with it. The later steady section should return to steady only as contextual reference for the Kármán near-constant component, not as a bridge that closes these transient or objective-transfer gaps. [R1-DRL reward rationale; M05–M06 scope boundaries; R2-freeze]
+
+## Layer 2 — local provenance ledger
+
+| Claim/object | Exact source or artifact | Chain, solver/geometry, case/seed/role, Re | Comparator, metric/estimand, ROI/window | Evidence status and boundary |
+|---|---|---|---|---|
+| Fixed/change architecture | `ROUND2_TOP_DOWN/05_REVISED_STRUCTURE_FREEZE.md` §§2,5; `03_CONTRACTS_METRICS.md` §§4–5 | Legacy/V5 kept separate; periodic Kármán, transient Vortex, periodic Illusion, Erase candidate | Periodic complete-cycle mean/phase8, event-offset L2, Illusion target metrics, Erase mean-field; no pooled score | `read`; definitions and manuscript boundary, not internal result authority |
+| Vortex scene and action labels | `ROUND1_BOTTOM_UP/01_DRL_CONTROL_AND_CASES.md` §§4,6; `src/drl_pinball/DATA_DICTIONARY.md`; `src/drl_pinball/legacy_test/README.md` | LegacyCelerisLab; `vortex_lamb`/`vortex_taylor`; offsets `y000`, `ym2L`, `ym1L`, `yp1L`, `yp2L`; Vortex action `\(\omega/U_0=4a+[0,-4,+4]\)`; Re inherited/documented from Legacy transfer, not reinterpreted here | `SI=800`, `MAX_STEPS=150`; transient event roles, not periodic phase | `read`; named contract. No arbitrary-flow or universal transfer claim |
+| Vortex event artifact/offset | `src/drl_pinball/data/reproduction_plots/manifest.json` entries `vortex_lamb_y000` and `vortex_taylor_*`; `DATA_DICTIONARY.md` `event_fields.npz` | Legacy plot navigation; target/controlled/physical-zero entries; `relative_offsets=[-10,-5,0,5,10]` around the controlled-role canonical event peak | Future per-event downstream ROI L2 at named offsets; current plot is derived navigation only; `wake_l2.py` lacks event-field support | `read`; manifest is navigation, not authority for audited quantitative role comparison; raw event NPZ/evaluator output not independently verified; `artifact-audit deferred` |
+| Vortex DTW semantics | `src/drl_pinball/legacy_test/core/dtw_metrics.py`, summarized in R1 metrics dossier | Legacy transient Vortex; target/state sensor histories; no periodic phase | Native Vortex DTW has no lag and rolls final window by `current_step+1`; complementary signal metric | `read`; code contract, not rerun. Not delay/causality/field equivalence |
+| Vortex quantitative gate | `ROUND1_BOTTOM_UP/08_EXPERIMENT_AND_CASE_METRICS.md` §§4.2,5.2,9; `ROUND1_BOTTOM_UP/01_DRL_CONTROL_AND_CASES.md` §§171–196 | Exact target/control/zero event roles required; same chain and target identity required | Per-event downstream ROI L2 at acquired offsets; event labels preserved; current `wake_l2.py` does not evaluate event fields | `read`; exact roles/evaluator authority are a module-specific blocker for any retained quantitative Vortex number/panel. Visual capability remains documentable |
+| Illusion PPO capability | `src/drl_pinball/train/env_illusion.py`, `TRAIN_PIPELINE.md`; `src/drl_pinball/CLAIMS.md` DRL-C4; `src/SR_analysis/data/runs/article-joint-data-illusion-20260718/illusion/illusion_0.75L/manifest.json`; retained V5 `latest_eval_summary` authority | V5 Illusion `ill_075L`, seed 43 retained demonstration; V5 target construction and controlled plant are separate; target radius ratio `0.75` (physical radius `15`, diameter `30` at `L0=20`); V5 code Re100 / `Re_D=50` where applicable, not Legacy-equivalent | PPO reward/force/DTW capability view; not independent field efficacy row unless exact field roles pass | `read`; retained historical named demonstration; target/model artifact audit deferred |
+| Illusion target definition | `env_illusion.py` contract; Illusion manifest above; `ROUND1_BOTTOM_UP/01_DRL_CONTROL_AND_CASES.md` §52 | V5/Legacy target construction differs by chain; target force and six sensor channels plus target `Cd,Cl` in V5 | Non-zero target signature/field; controlled target comparison; physical zero separate comparator | `read`; target identity must be retained per chain; no full-state equivalence claim |
+| Illusion geometry terminology | V5 `env_illusion.py`; `src/SR_analysis/data/runs/article-joint-data-illusion-20260718/illusion/illusion_0.75L/manifest.json`; `DATA_DICTIONARY.md` | `ill_075L`; `L0=20`; target radius ratio `0.75`, physical radius `15`, diameter `30`; historical labels may use `target_diam` or `D_t/L0` | Geometry identity only; no diameter-ratio relabelling | `read`; explain once that historical variable/plot names do not change the physical radius interpretation; V5/Legacy chains remain separate |
+| Periodic Illusion metrics | `ROUND1_BOTTOM_UP/08_EXPERIMENT_AND_CASE_METRICS.md` §§4.2–4.3; `src/drl_pinball/eval/wake_l2.py`; `DATA_DICTIONARY.md` | Only exact compatible target/controlled/zero roles; V5 or Legacy separately | Complete-cycle mean L2 and independent phase8 separately; normalized DTW only with exact target-channel normalization/window; registered ROI `-6..+14D`, `|y-y0|/D<=5` where evaluator applies | `read`; raw fields/exact masks authority-led `artifact-audit deferred`; no pooling |
+| Illusion historical/negative SR | `src/SR_analysis/CLAIMS.md` I1–I3; `src/SR_analysis/HANDOFF.md`; standardized role artifacts named there | LegacyCelerisLab; `illusion_1L`; SR vs physical zero, one standardized realization; independently phased roles; `Re_D=re_code/2` | Native DTW `0.9117016` vs `0.9119890`; normalized DTW `0.8940190` vs `0.8940537`; mean L2 `0.189990` vs `0.189414`, `eta=-0.00304`; phase8 `0.264176` vs `0.261832`, `eta=-0.00895`; `480+160` window | `read`; current authority-led negative/historical result, raw role audit deferred |
+| Erase role/failure boundary | `src/drl_pinball/legacy_test/README.md` line 13; `ROUND1_BOTTOM_UP/08` §§4.2,5.2,9; manifest `erase` entries | Legacy historical route; `erase`; filenames show navigation but do not establish exact independent roles | Candidate mean-field L2 plus DTW only under exact roles; historical route uses evolving controlled-force FIFO target | `read`; quantitative Erase unsupported/currently module-specific blocker; do not infer roles |
+| Force-centred conceptual bridge | `ROUND1_BOTTOM_UP/01_DRL_CONTROL_AND_CASES.md` §§1,3; `src/drl_pinball/train/env_karman.py`, `env_illusion.py` | Kármán force aggregation is case-specific; Illusion target force aggregation differs; Erase historical reward special | Force terms motivate near-zero cloak and target-force matching, but field L2 is independent | `read`; bounded interpretation only; small structures/higher Re unresolved; not generalized theory |
+| Near-collision gate | `00_QUALITY_GATE.md` §§14–18; `00_WAVE_A_COORDINATOR_AUDIT.md` §§28–37; M03–M06 exports | Compare within exact chain/case/role acquisition; never cross Legacy/V5 silently | Check role acquisition/set, ROI/domain, mask, weighting/normalization, window; classify same estimand/related diagnostic/different estimand | `read`; mandatory downstream control |
+
+## Figure, equation and table contracts
+
+### Cross-case extension figure (one contract, separated subpanels)
+
+**Role.** A single progression figure may contain: (i) Vortex event panels at named event labels and offsets, with action trace and downstream ROI-L2/DTW annotation only if exact roles exist; (ii) Illusion target, physical-zero and controlled panels for `ill_075L`, with visual target/controlled/physical-zero roles and any available complete-cycle mean L2, phase8 or normalized DTW visibly separated; field metrics are conditional on exact current role authority; and (iii) an Erase mean panel only if exact target/control/zero roles are later authority-bound. Each subpanel must state chain, case, role, seed, target/reference, metric, ROI/mask and window.
+
+**Caption limits.** Mark Vortex event labels as event-relative, not phase. Mark Illusion as PPO capability and keep negative Illusion SR visible locally. Mark Erase as unavailable/visual navigation if its quantitative role contract is absent. Similar numerical values across subpanels are not pooled.
+
+**Does not establish.** The figure does not establish a common efficacy score, arbitrary-flow universality, Kármán skeleton transfer, positive Illusion SR, causal force-to-field mapping, or Erase success from filenames/visual similarity.
+
+### Optional compact metric-definition block
+
+Define, only where roles permit,
+
+`E_event(o) = RMS_ROI(q_controlled(o) - q_target(o))`,
+
+and use the existing periodic `E_mean` and `E_phase8` definitions without treating event snapshots as phase fields. The equation defines a candidate event estimand; it does not create missing event roles, common masks or an evaluator result.
+
+### Supplementary case matrix
+
+The full matrix should retain all vortex labels and `relative_offsets=[-10,-5,0,5,10]`, Illusion target sizes, V5/Legacy identifiers, role acquisition status, native/normalized DTW, mean/phase/event/mean-field semantics, and artifact deferrals. It is a supplement contract, not a main-text inventory or pooled bar chart.
+
+## Claim ladder
+
+**Green — supported at named scope**
+
+- Current event artifacts and plot navigation support named transient Vortex visual roles, while the named historical V5 `ill_075L` PPO result supports capability-level Illusion evidence. Neither currently supports a quantitative Vortex target/control/zero event comparison without audited exact role artifacts.
+- Periodic Illusion is a non-zero target problem: target, controlled and physical-zero roles and complete-cycle mean/phase8 metrics are conceptually distinct and can be reported only where their exact authorities exist.
+- Illusion SR is historical/negative under the standardized long-window 1L comparison; Erase’s historical Legacy route is explicitly unsupported for quantitative efficacy.
+
+**Amber — bounded interpretation**
+
+- Pulse-like actuation supports a comparable-scale event-response hypothesis for the named Vortex disturbances.
+- A force-centred observer/reward view motivates the distinction between near-zero cloaking and target-force retargeting, while unresolved small structures and higher-Re effects limit that approximation.
+- Cross-case breadth changes the temporal/target contract, not the evidentiary status of the Kármán control skeleton.
+
+**Red — prohibited without new evidence**
+
+- Arbitrary incoming-flow capability, universal vortex control, continuous scale/offset generalization, or transfer of the Kármán SR/CCD interpretation to Vortex, Illusion or Erase.
+- Positive Illusion SR, target tracking by SR, efficacy over physical zero from old finite-run similarity alone, or one common L2/DTW efficacy score.
+- Quantitative Erase success inferred from filenames, plot existence or unsupported Legacy reward targets.
+- Causal force-to-field theory, exact small-structure suppression, higher-Re universality, DTW as delay, or cross-module numerical agreement/validation from near-collision values.
+
+## Imports, exports and bridge
+
+**Imports.** M03 supplies Legacy/V5, action, role, ROI and L2/DTW semantics. M04 supplies the Kármán performance comparator but not a transferable cross-case score. M05 supplies the positive Kármán/cloak control-skeleton boundary. M06 supplies descriptive Kármán field organization only. Round-1 Vortex, case-metrics and coverage dossiers supply event labels and role gaps; current package authorities control the facts.
+
+**Exports to Module 08 synthesis.** Named PPO demonstrations support event-relative control beyond periodic phase for comparable-scale vortex disturbances and objective retargeting/erasure under case-specific contracts; they do not establish arbitrary-flow universality, Kármán skeleton transfer, positive Illusion SR or one common efficacy score. Export the exact Illusion negative SR comparison and the Erase role/evaluator gap. No Vortex number should be exported until exact event roles and an evaluator authority are bound.
+
+**Exports to Module 10 conclusion.** Cross-case results show capability breadth with asymmetric interpretation: execution can be retargeted under named cases, but the positive control-skeleton reading remains Kármán/cloak-specific. Export the failure envelope—Vortex event-role/evaluator gap if unresolved, Illusion SR negative result, Erase unsupported route, and no pooled score—and the requirement for matched event/target-role acquisitions as the decisive next test.
+
+**Bridge to later steady context.** After these extensions, the manuscript may return to steady as contextual reference for the Kármán near-constant component. That return must not imply that steady evidence closes Vortex transient transfer, Illusion target-field transfer or Erase efficacy.
+
+## Conflicts, deferrals and genuine blockers
+
+- **Authority-led deferrals, nonblocking for bounded prose:** external/symlink-backed raw Vortex and Illusion field payloads, model/normalizer bundles, solver-exact masks and leaf-hash recomputation. The current authorities bind the stated roles or claim boundaries without a conflict; artifact audit is deferred.
+- **Module-specific blocker:** a quantitative Vortex event-L2 panel is blocked unless exact target/control/zero event roles, target identity, event offsets, ROI/mask and evaluator authority are available. Visual capability can remain as a bounded observation.
+- **Module-specific blocker:** an Erase quantitative panel is blocked by the explicitly unsupported Legacy route and absent independently mapped target-scale contract. Do not fill this gap from the plot manifest.
+- **No scientific blocker to the bounded section text:** Illusion can be represented with PPO capability plus locally visible negative SR, provided the metrics are not pooled and role/window contracts remain explicit.
+- **Systematic control already present:** the near-collision gate is a coordinator-level control inherited from Wave A. Applying it here is mandatory; it is not a new module-specific interpretation.
+
+## Coverage ledger
+
+| Status | Sources inspected and reason | Evidence state |
+|---|---|---|
+| `read` | Current Round-3 rule/master/quality/interface ledger and Wave-A audit; Round-2 freeze/style; Modules 03–06; Round-1 `01_DRL_CONTROL_AND_CASES.md`, `06_CURRENT_DOCUMENT_COVERAGE_AUDIT.md`, `08_EXPERIMENT_AND_CASE_METRICS.md`; current DRL README/CLAIMS/RESULTS/DATA_DICTIONARY/WRITING_HANDOFF; Legacy acquisition README; current SR CLAIMS/HANDOFF; Vortex/Illusion manifests and current environment/result authorities | Read directly; not independently verified or artifact-audited |
+| `mapped` | Current plot navigation and package links to external role roots; archive/old Illusion and Vortex routes used only to understand status/history | Mapped is not read, verified or audited; no claim depends on archive leaf material |
+| `not-covered` | Raw event/phase NPZ arrays, exact Vortex event evaluator output, complete Erase target/control/zero independent roles, new offline event adapter, new CFD/training, new figures and external literature | Not covered because outside first-pass drafting; quantitative Vortex/Erase panels remain conditional |
+| `external-inaccessible` | Symlink/external storage roots for selected reproduction fields and some role payloads; model/normalizer bundles where not exposed in ordinary inspection | Authority-led `artifact-audit deferred`; no missing payload is silently treated as absent or present |
+
+## Self-audit under the current gate
+
+**Two layers and cadence.** Layer 1 contains four argumentative modules with local caveats and substantive transitions; Layer 2 carries exact contracts, roles, metrics, windows and status. It is a revised first-pass section dossier for second audit, not polished manuscript prose and not a readiness declaration.
+
+**Systematic versus module-specific defects.** The systematic risk is metric pooling/near-collision language across heterogeneous cases; the draft applies the Wave-A role/ROI/mask/weight/window gate and prohibits arithmetic or validation language. A second systematic risk is source-path dominance; narrative uses short tags and keeps inventories in Layer 2. Module-specific defects are the missing authoritative Vortex event-L2 evaluator/role result and unsupported quantitative Erase route. They are preserved as gaps rather than repaired by inference. No new systematic rule defect was found in this pass.
+
+**Gate checks.** Legacy/V5, periodic/transient/steady and capability/interpretation chains remain separate. Spatial L2 is not a validation Level/Stage; DTW remains complementary and case-specific. Illusion SR is locally visible as negative/historical. Kármán SR/CCD modes are not imported as Vortex/Illusion/Erase interpretations. Figure contracts visibly separate metrics and prohibit pooled bars. No citation, number, causal mechanism, universal claim or artifact was invented.
+
+**Disposition.** Second-audit PASS; ready for Wave C/downstream imports, not final manuscript prose. Quantitative Vortex event-L2 and Erase panels remain blocked/conditional as specified in the figure and blocker contracts.

docs/JFM_WYQ/ROUND3_BOTTOM_UP/08_STEADY_SYNTHESIS.md

@@ -0,0 +1,87 @@
+# Round 3 first-pass module — Section 8: steady context, physical synthesis and theory limits
+
+**Status:** revised candidate for second audit; not self-declared ready. This module is a two-layer author-facing section draft and evidence dossier for the 6% steady return. It does not reverse the historical order: steady discovery is recalled briefly after the Kármán deep analysis, and the full steady/theory analysis appears here retrospectively.
+**Coordinator second-audit disposition:** PASS — ready for downstream Wave C imports, but not final manuscript prose.
+
+## Section question and bounded answer
+
+**Question.** After the Kármán deep analysis, what bounded steady context and theory constraints contextualize why the learned periodic law retains a near-constant rear-rotation core, and where does that interpretation stop?
+
+**Bounded answer.** The independently defined steady V5/free-slip branch supplies historical discovery context, sign and magnitude calibration, and a settling-qualified low-deficit endpoint compatible with the Kármán near-constant core; direct wall-sign checks, the failed potential-circulation sign route and incomplete finite-Re budgets prevent mechanism, stability, optimum and cross-chain validation claims.
+
+## Layer 1 — narrative draft
+
+The steady result returns here because it was discovered first, not because it is a continuation of the Legacy Kármán analysis. The early capability section records the short discovery seed; the Kármán section then shows a dominant near-constant rear-rotation component and a smaller dynamic correction under its own Legacy contract. The present return asks the narrower retrospective question: whether an independently defined steady branch contextualizes why that recurrence is plausible. It does not make the V5 steady result evidence for the Legacy SR or CCD numbers, and the two chains are never numerically pooled. [Round-2 freeze; current steady README/RESULTS; Modules 05–06]
+
+Under the modern V5 uniform-inlet, free-slip-channel contract at `Re_D=50`, the body order is `(front, rear_y_plus, rear_y_minus)` and the wall law is `(U_w,V_w)=(-\omega r_y,\omega r_x)`. The accepted branch is `[0,+\Omega,-\Omega]`, with `\Omega>0` and `s=\Omega R/U_\infty`; the direct cardinal check gives downstream velocity on both gap-facing rear cardinals and upstream velocity on the two outer cardinals. The accepted `s=3.55` endpoint is settling-qualified and has the authoritative downstream full-cross-section relative velocity-deficit `E_inf_vector(q_ctl,q_in)=0.0231672176752314` at `x/D=10`; nearby values are approximately `0.0232525`, `0.0231671` and `0.0312490` at `s=3.45,3.55,3.65`. The spatial profile L2 at `x/D=10` is `0.0153919` and remains a diagnostic, while `q_blk` is non-ranking. Thus `s=3.55` is a qualified endpoint, not a continuous optimum or universal law; the apparent local minimum remains provisional pending restarted bracket, continuation, convergence and held-out qualification. [V5 endpoint manifest; steady v3b report; display/profile authority]
+
+The sign check gives the shortest physical reading available from the endpoint. Positive rotation of the upper rear cylinder drives its gap-bottom cardinal downstream and the lower rear cylinder’s negative rotation drives its gap-top cardinal downstream; the accepted actuation therefore belongs to a base-bleed/gap-jet family, not a boat-tail family. Endpoint gap-flux, near-wake, recirculation, vorticity and downstream-profile diagnostics are compatible with describing a gap-facing or near-wake reorganisation and a low-deficit downstream profile. They remain observations and bounded hypotheses: the available package does not establish the arrows from wall motion through shear or separation to gap momentum, base pressure, force or downstream deficit, and the accepted `s=3.55` and reverse `s=5.2` states are not amplitude-matched. [V5 wall oracle; NS ledger; steady display; Round-1 theory dossier]
+
+The shortest validated theory reference is an outer impermeability calculation, not a finite-Re rotating-wall model. For one circle, the exact uniform-flow-plus-circulation trace is `u_\theta(a,\theta)=-2U\sin\theta+\Gamma/(2\pi a)`: impermeability leaves `\Gamma` independent, and for `U\ne0` no constant circulation can impose the full rotating no-slip tangential trace. The MFS/strip construction therefore supplies a reference family only after its validation gates; the current derivation records the exact-circle and independent unbounded Laurent checks as validated bounded mathematics, while dense off-grid, sector, conditioning, source-radius/order and image-layer gates are required before interpreting a new MFS field, and no publication-ready MFS field is implied here. The attractive cancellation route fails its decisive sign check: the strip outer reference requires `g_opt=+6.147944`, whereas the accepted finite-Re endpoint’s descriptive fit is `g=-12.200554`. The failed route is retained because it limits the interpretation; it cannot be silently replaced by a rotating no-slip closure. [DERIVATION.md; current strip and finite-Re authority; LITERATURE.md]
+
+The integrated map therefore constrains, without explaining, one set of compatible, non-equivalent observations. The Legacy Kármán SR deletion test ranks its persistent rear constant above the tested rear-lift feedback under a fixed parent-relative DTW window; the corrected Legacy CCD ledger says its constant-control mean-field reference accounts for `93.42%` of the measured zero-to-DRL ROI target-error reduction, leaving a `6.58%` increment organized descriptively by action-correlated correction-field modes; the separate V5 steady branch supplies the accepted sign and a low-deficit endpoint at a rear-rotation scale of order `s=3.55`. These analyses organize separate action-space and field-role observations around a persistent/constant reference, without a common estimator or independent triangulation. They use different chains, roles, domains, masks, windows and estimands: they are compatible observations, not replication, validation, agreement, a pooled metric or a closed control skeleton. This bounded synthesis therefore contextualizes, but does not explain, the persistent/constant reference before bridging qualitatively to the planned open-loop experiment—steady operation may be inspected for straighter dye streaklines and the illusion operation for changed shedding frequency. Experiment remains qualitative, and the computational interpretation stops before mechanism, stability, optimum, momentum restoration, causal closure or cross-chain validation. A decisive next test would combine amplitude-matched branches with held-out endpoint evidence, a validated outer-reference comparison and a complete same-control-volume Navier–Stokes momentum/mechanical-energy budget, followed by synchronized quantitative experiment if the claim is to cross the CFD boundary. [Modules 05–06; steady NS/theory authorities; Round-2 experiment boundary]
+
+## Layer 2 — local provenance and claim-to-evidence ledger
+
+| Narrative claim | Exact current authority / artifact | Chain, solver, geometry, case and Re | Comparator, metric/estimand, ROI/window | Status / verification boundary |
+|---|---|---|---|---|
+| Historical steady discovery and distinct contract | Round-2 `05_REVISED_STRUCTURE_FREEZE.md` §§2, 3, 5; `src/steady_pinball_theory/README.md`, `RESULTS.md`; `src/SR_analysis/results/runs/article2-steady-analysis-20260720/interpretation.md` | V5 steady/free-slip versus LegacyCelerisLab Kármán; three-cylinder pinball, body order explicitly separate; V5 `Re_D=50`, Legacy `Re_code=100` means `Re_D=50` only within Legacy’s own convention | Steady is contextual; Legacy SR/CCD retain their own role-local estimands | Current authority read; not independent CFD recomputation or universal artifact audit |
+| Accepted branch and direct sign | `steady-contract-v5-torque-d20-20260808/accepted-s3.55/manifest.json`; `steady-article-publication-v3b-20260809/REPORT.md`; `contract.py`/current steady README | V5 CelerisLab/LBM, `D=20`, `R=10`, uniform inlet, free-slip lateral walls, rotating no-slip cylinders; accepted action `[0,+\Omega,-\Omega]`, `\Omega=0.00355`, x-right/y-up | Cardinal wall velocities; branch/sign oracle, not a field metric | Current manifest-bound contract; direct wall check recorded by authority; artifact leaf audit deferred |
+| Low-deficit endpoint | Same accepted endpoint manifest and publication v3b; `figures/summary/profile_Einf_vector.csv` | V5 free-slip `Re_D=50`, accepted controlled endpoint, settling window 10, final `tU_\infty/D=250` | Full-cross-section two-component `E_inf_vector(q_ctl,q_in)` at `x/D=10`: `0.0231672176752314`; `E_L2_vector=0.0153918657` is diagnostic; nearby `s={3.45,3.55,3.65}` `E_inf_vector≈{0.0232525,0.0231671,0.0312490}`; `q_blk` non-ranking | Qualified endpoint observation; not optimum, uniqueness, attraction, stability or causal intervention; endpoint artifact audit deferred |
+| Reverse boundary | `steady-results-display-v2-20260809/README.md`; current steady `RESULTS.md`; profile CSV | V5 same nominal plant family, reverse branch `s=5.2`, but distinct amplitude and reverse sign | `E_inf_vector(x/D=10)=0.6898418` versus accepted `0.0231672`; this is not an amplitude-matched controlled comparison | Current display/CSV authority; descriptive branch contrast only, no mechanism or stability inference |
+| Gap-facing physical reading | V5 manifest `wall_oracle`; `steady_pinball_theory/LITERATURE.md` sign master map; `ns-ledger-v2-oracle-bound-d20-20260809/analysis.json` | V5 accepted branch, rear gap/outer cardinal samples; endpoint NS diagnostic ledger | Signed gap fluxes, near-wake/recirculation/vorticity and profile diagnostics are endpoint spatial observations; no single pooled score | Bounded observation/hypothesis; no causal arrow, force ownership or closed budget |
+| One-circle outer reference | `DERIVATION.md` §§5, 11; `LITERATURE.md` Crowdy (2006), DOI `10.1016/j.euromechflu.2005.11.005`; Ueda et al. (2003), DOI `10.1017/S002211200300627X` | Irrotational Euler/Neumann outer problem; one-circle exact identity, not V5 finite-Re no-slip solver | Wall normal/tangential trace identity; no downstream V5 metric | Existing derivation; exact-circle mathematics separately checked in the derivation record; no finite-Re mechanism transfer |
+| MFS/strip gate and failed sign | `steady-strip-modal-analysis-20260804-v4/analysis.json`; `steady-finite-re-bridge-analysis-20260804-v4/analysis.json`; `DERIVATION.md` §§11–13 | Outer strip/unbounded potential references with finite-image MFS; finite-Re descriptive contour projection from accepted V5 endpoint | Strip cancellation `g_opt=+6.147944`; finite-Re descriptive fit `g=-12.200554`; `same_sign=false`; strip reference `BOUNDED`, confinement mechanism `FAIL` | Current authority; sign conflict is a genuine interpretation blocker for that route; MFS field interpretation remains gate-dependent and no new field is promoted |
+| Partial NS/budget boundary | `ns-ledger-v2-oracle-bound-d20-20260809/analysis.json`; v3b report/display README | V5 accepted/reverse endpoint diagnostics, one body-enclosing rectangular CV | Endpoint spatial fluxes, dissipation, gap flux and actuator-on-fluid power; storage and body-adjacent terms unavailable, residual `null` | Current partial ledger; mechanical-energy closure `NOT_CLOSED`, causality `NOT_ESTABLISHED`, force ownership not established; authority-led raw-term deferral |
+| Legacy SR magnitude context | `src/SR_analysis/results/runs/article2-steady-analysis-20260720/interpretation.md`, `steady_constant_correspondence.json`; SR current claims and Module 05 | LegacyCelerisLab disturbance-free contextual roles; separate Legacy geometry/API and action convention | Steady sweep `A=0…6`, best listed `A=5`, Legacy DTW similarity `0.9872116`; Kármán law deployment `0.9664944`; rear mean `(-3.430589,+3.431828)`, fitted magnitude `3.431209` | Contextual calibration only; not pooled with V5 `E_inf_vector`, not momentum equality or mechanism |
+| Legacy CCD compatible observation | Module 06 current corrected authority; `src/CCD_analysis/RESULTS_INDEX.md`, `FINAL_RESULTS.md`, corrected manifests | Legacy `karman_re100`, target/zero/constant/DRL roles; `Re_D=50` under Legacy `Re_code=100`; exact four-role common mask | Coordinate-weighted ROI mean errors `E_zero=0.4694352273`, `E_constant=0.1110140739`, `E_DRL=0.0857787860`; `93.42/6.58` scalar error-difference share; separate `19×10` residual | Current descriptive/noncausal result; not steady evidence and not a paired intervention; dense arrays/leaf audit deferred |
+
+### Existing derivation, independent checks and open conjectures
+
+- **Existing derivation:** one-circle Euler trace, fixed-circulation outer uniqueness under stated hypotheses, rear-opposite potential/dipole algebra, MFS construction, strip finite-image relations, `E_inf_vector` definition and partial NS-ledger identity.
+- **Independently checked/validated at bounded level:** current authority records exact-circle failure controls, bounded agreement of independently discretized unbounded MFS and Laurent/Fourier checks, finite-image convergence observations without an asymptotic tail claim, direct wall-cardinal sign, accepted endpoint settling and the finite-Re descriptive sign fit. These do not validate an MFS field as finite-Re physics.
+- **Open conjecture:** accepted gap-facing motion may participate in a base-bleed/gap-jet-family near-wake organisation and may be compatible with a low-deficit endpoint; a persistent rear rotation may provide a mean compensation-like contribution with a smaller dynamic correction. Neither conjecture is a mechanism, momentum restoration, stability statement, optimum or causal law.
+
+## Figures, equations and table contracts
+
+1. **Accepted/reverse profiles and fields plus `E_inf_vector(s)` curve.** Source from the current steady display/profile CSV and endpoint fields; show accepted `s=3.55`, reverse `s=5.2`, stationary reference and nearby accepted sweep values. Caption must state V5/free-slip `Re_D=50`, full-cross-section `q_ctl` versus `q_in`, `x/D=10`, diagnostic L2, reverse amplitude mismatch, and endpoint—not optimum/stability—status. Redraw is dependent on current manifests/field payloads; no new publication-ready figure is claimed.
+2. **Action-sign schematic.** Redraw the body order and four rear cardinal velocities from the accepted manifest. It establishes the sign and gap-facing/outer-facing observation; it does not establish circulation, base-bleed causality or boat-tail equivalence.
+3. **Shortest outer-reference/sign-conflict panel.** Use the validated one-circle trace and a compact `g_opt=+6.147944` versus `g=-12.200554` sign comparison. If an MFS field is retained, all dense off-grid, sector, conditioning, source-radius/order, image-layer, one-circle and independent-reference gates must be visibly complete; otherwise leave the field in the supplement/open ledger. This panel does not establish rotating no-slip finite-Re physics.
+4. **Integrated evidence map.** Nodes are Legacy SR term ranking, Legacy corrected mean/CCD organization, V5 steady endpoint and qualitative experiment bridge. Solid arrows mean same-contract definition or result; dashed arrows mean compatible observation across non-equivalent estimands; broken/unresolved arrows mean mechanism, causality or closure is open. The map must print chain, metric and status and must not be rendered as a causal flow diagram.
+
+**Equation budget:** reuse the wall law/action sign and `s`, the steady `E_inf_vector` definition, and either the one-circle trace or the partial NS-ledger identity. The full MFS, strip-image, multipole and CV derivations belong in the supplement unless their current validation gates are explicitly closed.
+
+## Claim ladder and interfaces
+
+**Green / supported:** accepted V5/free-slip sign contract; direct wall-cardinal check; settling-qualified `s=3.55` endpoint; `E_inf_vector` at `x/D=10`; profile L2 as diagnostic; reverse high-deficit endpoint as a distinct non-amplitude-matched observation; failed potential-cancellation sign; partial rather than closed NS ledger.
+
+**Amber / bounded:** low-deficit endpoint is compatible with the recurrent Kármán near-constant rear-rotation scale; accepted wall motion is in a base-bleed/gap-jet family; steady, SR and corrected mean/CCD observations can be integrated as compatible, non-equivalent context; qualitative experiment is a prospective bridge.
+
+**Red / prohibited without new evidence:** mechanism, causal gap-jet or momentum-restoration claim; `Gamma(Omega)` closure; MFS-to-no-slip transfer; stability, branch attraction, optimum or universal `s` law; Legacy/V5 numerical continuation/equivalence; arithmetic pooling or replication/validation/agreement/triangulation language; quantitative experiment validation.
+
+**Imports:** Modules 03–06 supply chain, metric, Legacy SR, corrected mean/dynamic and CCD boundaries. The current steady authority supplies the V5 contract, accepted sign, endpoint, theory failure and incomplete budget.
+
+**Exports to Module 09/10:** a bounded physical synthesis (`persistent rear rotation + smaller correction` as compatible organization only), the failed potential-sign route and incomplete-budget decisive tests, and a qualitative experiment bridge. No quantitative steady number, sign convention or endpoint metric may be transferred to Legacy without its own contract.
+
+## Near-collision check
+
+The Legacy SR fitted constant magnitude (`3.431209` in `alpha=omega/U_0`) and the V5 steady endpoint (`s=3.55`) are numerically similar dimensionless labels but use different action parameterizations, chains, plants and objectives; they are related contextual diagnostics only, not amplitude agreement, validation or an arithmetic comparison. CCD has no actuator-magnitude comparison here: its constant mean is a field-role reference under the corrected four-role estimand and can join the synthesis only as a non-equivalent mean-field observation. No arithmetic combination, replication, validation, agreement or triangulation wording is permitted.
+
+## Defects, gates and coverage ledger
+
+**Systematic/module-specific classification:** the current Round-3 controls require a cross-module near-collision check; this module applies it explicitly. Module-specific risks are the V5/Legacy sign temptation, provisional `s` minimum, MFS validation status and partial NS budget. No control-plane edit is authorized in this one-file task.
+
+**Genuine blockers:** none for the bounded section text. The potential sign conflict is a genuine blocker to that mechanism route, not to this bounded synthesis. A retained MFS field panel becomes blocked until its listed validation gates are complete; an exact artifact-audit claim becomes blocked if authority cannot bind the panel source.
+
+**Authority-led deferrals:** raw endpoint fields, external/symlink-backed roots, leaf hash recomputation, complete MFS raw arrays and full NS terms are deferred because current authorities bind the roles, numbers, contract and status without conflict. These deferrals do not authorize stronger claims.
+
+| Coverage status | Sources and reason | Evidence status |
+|---|---|---|
+| `read` | Current Round-3 master/rule/quality/interface ledger and Wave-A audit; Round-2 freeze/style; current Modules 03–06; Round-1 steady, SR, CCD/OID and coverage dossiers; current steady README, RESULTS, DERIVATION, LITERATURE, evidence README/INDEX, v3b report, display README, V5 endpoint manifest, strip/finite-Re authorities, NS ledger, validation authority and SR steady correspondence | Read directly; not equivalent to independent recomputation or universal artifact audit |
+| `mapped` | Current figure manifest/CSV and authority-linked external roots used to identify redraw dependencies; archive/history only for replacement navigation | Mapped/indexed, not read as independent evidence; not promoted |
+| `not-covered` | New CFD, new MFS computation, full raw-field/NPZ audit, universal hashes, new literature search, synchronized experiment analysis | Outside first-pass scope; no result inferred |
+| `external-inaccessible` | Symlink-backed/external endpoint and dense theory/NS payloads not fully exposed through ordinary repository reading | Authority-led `artifact-audit deferred`; nonblocking for bounded text, blocking for any retained unsupported panel or reproducibility claim |
+
+## Anti-index / anti-AI audit
+
+The narrative has five argumentative paragraphs with local caveats, not a package inventory. Numbers appear only where they change the inference; chain, comparator and metric distinctions are repeated only at the necessary bridge. No archive result is promoted, no V5 sign is transferred to Legacy, no spatial L2 is pooled with DTW or CCD, no potential field is presented as a mechanism, and no first-pass readiness claim is made.

docs/JFM_WYQ/ROUND3_BOTTOM_UP/09_EXPERIMENT_AUTHOR_INTERFACE.md

@@ -0,0 +1,151 @@
+# Round 3 — Qualitative experiment author interface
+
+**Status:** first pass; author interface for Frank, not manuscript prose and not self-ready. Frank retains authorship of the subsection. This module reserves a concise qualitative open-loop experiment subsection and defines the evidence and provenance needed before any prose, caption or figure is promoted.
+
+**Coordinator second-audit disposition — PASS:** interface ready for Frank/Wave C imports. Final prose and any figure remain blocked by the author-supply/provenance gate; this is not final manuscript prose.
+
+## Section question and bounded answer
+
+**Question.** What does the physical deployment demonstrate qualitatively, and what evidence is still required before it can be presented as an experimental result rather than a feasibility demonstration?
+
+**Answer ceiling.** Under the author-described apparatus and open-loop operation, the experiment can support only bounded qualitative feasibility observations: steady operation appears to produce straighter dye streaklines and illusion operation visibly changes shedding frequency. It does not currently support synchronized force/PIV validation, quantitative field matching, closed-loop experimental validation, or a causal explanation of the observations.
+
+## Source boundary and coverage ledger
+
+The author-described facts below are preserved as `[AUTHOR]`, not independently verified. External facility literature may motivate measurement cautions or future design choices, but cannot establish the DynamisLab apparatus, its conditions or its results.
+
+| Status | Sources inspected | Meaning |
+|---|---|---|
+| `read` | Round-2 `05_REVISED_STRUCTURE_FREEZE.md`, `06_SECTION_LEVEL_JFM_STYLE_GUIDE.md`; Round-3 rule, quality gate and interface ledger; ready M07 `07_CROSS_CASE_EXTENSIONS.md`; ready M08 `08_STEADY_SYNTHESIS.md`; Round-1 `08_EXPERIMENT_AND_CASE_METRICS.md`; Round-1 `06_CURRENT_DOCUMENT_COVERAGE_AUDIT.md` | Read as narrative/claim controls. Read is not independent verification or artifact audit. |
+| `mapped` | Repository experiment references and current case/metric interfaces named by those authorities | Mapped/indexed material is not read evidence and does not establish an experimental result. |
+| `not-covered` | Complete local raw experiment package, synchronized force/PIV data, calibration records, repeatability analysis and new figure generation | Not available or outside this first pass; do not fill gaps from memory, plots or CFD. |
+| `external-inaccessible` | Undermind Oil-tunnel background material and hidden/external raw facility artifacts not exposed in the repository | Context only after identity/reading level is resolved; cannot establish internal apparatus/results; raw audit deferred. |
+
+No current internal experiment document with an auditable apparatus/data package was located beyond the Round-1 authority above. Current status is author-described and provenance-incomplete, not independently verified.
+
+## Layer 1 — author-facing paragraph scaffold
+
+### Paragraph 1 — purpose, facility and apparatus boundary
+
+**Opening job:** establish deployability and the experimental boundary in one compact paragraph, without turning the subsection into a construction diary.
+
+> To assess whether the learned actuation could be deployed physically, [we/Frank] built a system from scratch in a recirculating water tunnel, using an isolated test section as a still-water towing section. A linear guide/rack and gantry carriage moved the [declared geometry] while [force sensor, ADC, Raspberry Pi, motor drivers, cylinders and power system] provided the actuation and acquisition hardware; Python was used for hardware/software deployment. The trials were operated **open loop**, with [exact motor/action waveform and sign convention to be supplied]. Because the towing speed and resulting hydrodynamic force were very low, the available force feedback was below a useful closed-loop level under the present conditions. [Insert only facility dimensions, water conditions, geometry, speed and trial identifiers after the provenance gate passes.]
+
+**Evidence level:** `[AUTHOR]` apparatus and deployment description; not independently verified. This establishes physical feasibility of deployment, not measurement accuracy or policy validation.
+
+**Computational-to-experiment bridge:** this is physical deployment of the steady and illusion actuation objectives, not a reproduction of the Legacy/V5 CFD chain. Do not transfer CFD Reynolds conventions, solver fields, L2/DTW values, SR laws or CCD interpretation without a separate same-plant, calibrated mapping.
+
+### Paragraph 2 — bounded qualitative observations
+
+**Opening job:** state only what archived raw media and trial identity can support, with comparator and observation kept local.
+
+> In the steady comparison, the dye streaklines in [trial/control/reference conditions] appeared [straighter/less deflected] over [declared observation region and time window]. In the illusion comparison, the dye visualization showed a visible change in shedding frequency under [declared open-loop action and trial ID]. These are qualitative image observations, not velocity or force measurements: no synchronized calibrated PIV/force package was available for the present comparison, and no quantitative field matching was performed. [Insert whether the observation was repeated, how frames were selected, and exact raw-media provenance.]
+
+**Evidence level:** `[AUTHOR]` observation, conditional on raw image/video provenance. Keep “appeared” unless a declared image-analysis protocol and uncertainty support stronger wording. “Changed shedding frequency” remains visual/qualitative unless frame rate, timing and event extraction support a physical frequency estimate.
+
+**Comparator rule:** name the actual trial/reference pair. Do not call dye a CFD-to-experiment match, field restoration, force reduction or experimental validation. Steady straightness and illusion shedding-frequency change are separate observations, not one efficacy score.
+
+### Paragraph 3 — limitation and next decisive experiment
+
+**Opening job:** place the main limitation beside the observation and define the next evidence threshold.
+
+> The present setup therefore provides a qualitative open-loop feasibility demonstration rather than a closed-loop experimental validation. Thermal convection is a **suspected confound** under the low-speed, low-force conditions, but the present record does not prove that it was the sole, or even dominant, source of background motion or force contamination. A quantitative comparison would require [synchronized calibrated force and PIV or an explicitly bounded image/sensor protocol], background blank controls, repeat trials, calibration and uncertainty records, and a traceable mapping between the computational target/reference and the physical acquisition. Until those records are available, the experiment should remain a qualitative main-text/supplement decision under the figure gate below.
+
+**Evidence level:** `[AUTHOR]` limitation and `[OPEN]` next-test requirements. The bridge to M10 is feasibility → evidence ceiling → decisive future test, not feasibility → validation.
+
+## Layer 2 — author supply checklist and provenance contract
+
+Frank must supply, or explicitly mark unrecoverable, the following before final prose or a figure request. “Not available” is acceptable; it must not be silently replaced by an estimate.
+
+### A. Facility, fluid and geometry
+
+- [ ] Facility identity and test-section dimensions: tunnel/channel, isolated still-water section, walls/free surface and relevant boundaries.
+- [ ] Water conditions: temperature/variation, preparation, fill/settling procedure, pump state and whether circulation was stopped or isolated.
+- [ ] Cylinder/geometry identity: diameter(s), spacing, body order, span/depth, mounting, blockage and coordinates; distinguish experiment from Legacy/V5 geometry.
+- [ ] Towing direction, carriage origin, travel range, acceleration/deceleration and settling interval.
+- [ ] Nominal and measured towing speed, speed trace if available, intended/estimated experimental `Re_D`, definition and viscosity source. Do not infer it from CFD labels.
+
+### B. Actuation and trial identity
+
+- [ ] Exact motor/action waveform for steady and illusion: amplitude, frequency/timing, phase/offset, duration, command units and physical-motion conversion.
+- [ ] Action signs and body ordering in experimental coordinates; identify directly documented versus reconstructed signs.
+- [ ] Trial IDs, date/time, case label, control/reference condition, geometry/setup version and excluded/failed trials.
+- [ ] Repeat count per condition and rule for selecting the shown trial; state whether the observation is one sequence or replicated.
+- [ ] Open-loop timing: command start, carriage motion, dye injection and what was/was not synchronized.
+
+### C. Raw media and image provenance
+
+- [ ] Raw image/video files, immutable filenames or hashes, camera, view, field of view and trial-ID linkage.
+- [ ] Spatial scale/calibration, frame rate, exposure/shutter, resolution, lens/view geometry and cropping/stabilization/enhancement/frame dropping.
+- [ ] Dye identity/concentration, injection location, volume/rate, timing and repeatability of injection protocol.
+- [ ] Blank/background media for no-tow/no-actuation and, if available, residual-motion and post-settling conditions.
+- [ ] Observation window/region; identify visual judgment versus a declared analysis procedure for “straighter” or “changed shedding frequency”.
+- [ ] Figure permissions, ownership, consent where relevant and permission to archive raw media in the supplement.
+
+### D. Sensors, calibration and synchronization
+
+- [ ] Force sensor model, mounting, axes, range, sampling rate, tare/zero, calibration curve, drift, noise estimate and units.
+- [ ] ADC configuration, Raspberry Pi clock/sample timing, motor-driver/encoder timing and clock offsets/dropped samples.
+- [ ] Carriage position/speed measurement and motor-command logging; distinguish commanded waveform from realized motion.
+- [ ] Synchronization marker shared by camera, force/ADC, carriage and actuation streams; if absent, state so.
+- [ ] Explicitly state that no synchronized calibrated force/PIV data were available if that remains true. Never infer force feedback from deployment software, reward, dye or uncalibrated ADC.
+
+### E. Background controls, uncertainty and recoverability
+
+- [ ] Blank runs, settling/wait times, pump/carriage state and temperature/thermistor records near relevant walls/surfaces if available.
+- [ ] Measured background velocity/dye displacement and method; do not subtract or attribute it without a declared control protocol.
+- [ ] Calibration, timing, scale, registration and repeatability uncertainty actually supported.
+- [ ] What cannot be recovered: raw media, trial IDs, speed, synchronization, calibration, repeat count or geometry. List each explicitly.
+
+### Provenance acceptance rule
+
+A figure or quantitative sentence may use only values/observations whose raw source, trial identity, condition, processing history and permission are traceable. Author memory can describe the setup as author-provided, but cannot supply missing numerical values. Authority-led deferral is acceptable for hidden artifacts only when an authority binds their role without conflict; it does not authorize a new panel or independent-verification claim.
+
+## Figure gate
+
+**Raw provenance must pass before any experiment image enters the main text.** The gate requires raw file/linkage, trial ID, condition, scale, frame rate/exposure, dye timing, processing history, repeat status and permissions. If any required provenance is missing, do not include the image as a main-text result. A supplement-only panel is permitted only if clearly labelled as an uncalibrated qualitative/open-loop demonstration, with available provenance and missing fields stated in the caption, and no quantitative inference. If raw media cannot be recovered, omit the panel rather than digitizing or reconstructing it from a plotted image or memory.
+
+### Candidate figure contracts
+
+1. **Optional apparatus schematic:** may show isolated still-water towing section, guide/rack, gantry, geometry and deployment signal path. It documents apparatus context and deployment feasibility; it does not prove calibration, accuracy, closed-loop operation or force-feedback quality. Avoid a component-by-component hardware diary.
+2. **Optional steady/illusion dye plate:** may show raw, minimally processed media with separate trial IDs. It supports only qualitative streakline/shedding observations; it does not prove field matching, velocity/vorticity agreement, force reduction, causal shedding control or experimental validation.
+3. **No synchronized force/PIV panel in this pass:** reserve it for later calibrated acquisition with clock alignment, common coordinates/mask and uncertainty. Do not register CFD fields onto dye images or imply a quantitative bridge without these inputs.
+
+## Claim ladder and exact nonclaims
+
+### Green — author-described, named qualitative scope
+
+- System built from scratch in a recirculating water tunnel with isolated still-water towing section, linear guide/rack, gantry carriage and listed deployment hardware, **as author-described**.
+- Python hardware/software deployment enabled learned actuation to run on the apparatus, **as author-described**.
+- Operation was open loop; useful closed-loop force feedback was not achieved under described low-speed/small-force conditions, **as author-described**.
+- In named steady media, dye streaklines appeared straighter; in named illusion media, shedding frequency visibly changed, **only after raw-media provenance passes**.
+
+### Amber — bounded interpretation
+
+- Apparatus provides qualitative physical deployment/feasibility demonstration for steady and illusion actuation.
+- Observations are consistent with qualitative wake modification under named open-loop conditions.
+- Background thermal convection is a suspected confound motivating controls, not established as sole cause of force/image contamination.
+- Demonstration motivates future synchronized, calibrated, repeated experiment; it does not raise the computational claim ceiling.
+
+### Red — exact nonclaims without new evidence
+
+Do not write or imply: “experimentally validated cloak/illusion”; “closed-loop experimental control”; “force feedback achieved”; “PIV confirms the CFD”; CFD-field matching; force reduction or field restoration from dye; quantitative shedding frequency, L2, DTW, uncertainty or efficacy percentage without its acquisition contract; thermal convection as proven sole cause; causal mechanism, momentum restoration, stability, optimality, robustness or generalization; or one common metric across dye, CFD, steady profiles, periodic phase and transient events.
+
+## Imports, exports and M10 handoff
+
+**Imports from M03–M08:** use frozen vocabulary `hydrodynamic cloaking`, `hydrodynamic illusion`, `wake-signature matching`, `field matching` and `qualitative open-loop experiment`; preserve Legacy/V5 separation, case-specific metrics, positive SR limited to Kármán/cloaking, negative Illusion SR, descriptive/noncausal CCD and steady as contextual. M07’s Illusion result is PPO capability plus historical/negative SR boundary; M08 supplies only a prospective qualitative bridge. None of these computational results validates apparatus or dye media.
+
+**Export to M10:** the experiment contributes a bounded feasibility observation and provenance-dependent figure option. Carry forward: no synchronized calibrated force/PIV validation, no quantitative field match, no closed-loop experimental validation, low-speed/background susceptibility and unresolved recoverability. The decisive next test is same-geometry, declared-condition, repeated, synchronized and calibrated acquisition with blank/background controls and uncertainty accounting.
+
+**Final bridge:** conclusion must separate “physical deployment was attempted and qualitative wake changes were observed” from “the computational target/reference was quantitatively matched.” The latter remains open.
+
+## Genuine blockers and first-pass disposition
+
+- **No blocker to this author interface:** section can be reserved with author-described facts and explicit slots.
+- **Blocker to final prose/figure inclusion:** missing raw-media provenance, trial linkage or permission; missing conditions needed to identify the observation; or any quantitative/validation claim without synchronized calibration and a case-specific metric contract.
+- **Module-specific evidence gap:** internal apparatus and raw experiment package were not independently audited. Oil-tunnel literature, CFD artifacts, memory and plotted images cannot repair this.
+- **Disposition:** submit for parent audit/revision. Do not mark `ready` or self-ready.
+
+## Anti-pattern audit
+
+No values, hardware performance, citations or raw artifacts are invented; author memory is not upgraded to independent evidence; the section is not a hardware diary; suspected thermal convection is distinct from proven cause; steady/illusion/CFD metrics are not pooled; and limitations sit beside each possible observation. Final manuscript prose remains Frank’s responsibility after the provenance gate is resolved.

docs/JFM_WYQ/ROUND3_BOTTOM_UP/10_LIMITATIONS_CONCLUSION.md

@@ -0,0 +1,88 @@
+# Round 3 revised candidate — Section 9: limitations and conclusions
+
+**Status:** revised candidate for second audit; not ready. This module answers the integrated claim-ceiling question after Modules 03–09. Module 09 is imported as a ready author interface, while final experiment prose and any panel remain blocked by the author-supply/provenance gate.
+
+## Section question and bounded answer
+
+**Question.** What integrated conclusion survives the Legacy/V5, chain, metric, causality, theory and experiment boundaries, and which decisive tests would raise the claim ceiling?
+
+**Bounded answer.** Named learned control provides capability demonstrations across the declared tasks; independent field matching is supported only for named Legacy Kármán Re100; and the Legacy Kármán action and field evidence supports a bounded control-skeleton interpretation organized by dominant persistent rear rotation plus a smaller dynamic correction. Cross-case Vortex and Illusion evidence remains named wake-signature or visual PPO capability under distinct contracts, with no Vortex event L2, conditional Illusion field L2 and unsupported Erase quantification; the author-described experiment interface remains provenance-gated. These levels do not establish a unbounded law or causal explanation.
+
+## Layer 1 — narrative and conclusion map
+
+### Discussion paragraph 1 — capability to Legacy Kármán field matching
+
+The strongest integrated result begins with capability, but its evidence is not a single pooled score. Under the named Legacy and V5 contracts, learned force-plus-sparse-observation control provides named wake-signature capability for the declared Kármán task and extends to named transient Vortex and non-zero-target Illusion demonstrations under distinct contracts. Independent spatial field matching is supported only for the named Legacy Kármán Re100 case: the controlled trajectory is closer to its target than physical zero in both complete-cycle mean-field and eight-slot phase-resolved L2 within the registered downstream ROI. DTW remains complementary, signal-level and case-specific; it does not replace L2, represent a physical delay or establish causality. The V5 seed/retention evidence and the Legacy Kármán field result therefore remain separate support chains, not a pooled convergence claim. [M03; M04; Wave-A audit]
+
+### Discussion paragraph 2 — executable SR and corrected mean/dynamic CCD
+
+The Kármán interpretation becomes narrower, rather than stronger, when the action and field analyses are placed in their own estimands. In the Legacy executable SR route, the mapped-shared law passes its named deployment checks, and parent-relative fixed-window deletions rank the persistent rear counter-rotation as the largest tested contribution, with smaller rear-lift feedback and a weak or mixed tested front correction. These are deletion tests, not causal ablations. In the corrected four-role Legacy CCD ledger, the constant mean accounts for 93.42% of the measured zero-to-DRL target-error reduction and the remaining 6.58% is organized by a separately centred, non-paired residual represented as action-correlated correction-field modes. Together these are compatible, non-equivalent action-space and field-role organizations that motivate a control-skeleton reading—dominant persistent rear rotation plus smaller dynamic correction—but they are not two independent confirmations and neither establishes term necessity, a unique law, a response mechanism or causal field attribution. The standardized PPO and corrected CCD mean errors are different estimands: they use different role acquisitions and role sets, retained windows, masks and weighting/normalization contracts, and are not combined arithmetically or described as replication, validation, agreement or triangulation. [M05; M06; M04–M06 reconciliation]
+
+### Discussion paragraph 3 — breadth, steady context and experiment boundary
+
+The interpretation is asymmetric across cases. The Vortex material supports only a comparable-scale, event-relative extension; no Vortex event L2 is currently available. Illusion PPO remains separate capability evidence, while Illusion SR is historical/negative and must not be used to negate or explain the PPO demonstration. Erase has no supported quantitative route under the historical evolving-force-FIFO target. The steady branch is separate contextual evidence: SR’s fitted constant magnitude 3.431 and the V5 endpoint at s=3.55 are related diagnostics only, not an amplitude comparison, cross-chain continuation or proof of the Kármán law. CCD’s constant mean is a field-role comparator, not an action magnitude. The failed potential sign and incomplete Navier–Stokes/momentum/mechanical-energy accounting leave the physical reading bounded. Finally, Module 09 reserves an author-described qualitative open-loop experiment interface: steady streaklines were described as straighter and Illusion shedding as visibly changed, but those observations remain conditional on raw-media provenance and trial linkage. They are not established paper results and do not provide quantitative field matching, calibrated force evidence or experimental validation. [M07; M08; M09; R1 experiment/case-metrics dossier]
+
+### Discussion paragraph 4 — decisive tests that change the ceiling
+
+The next tests should target the exact unresolved links rather than add cases without matched estimands. First, propose same-protocol Legacy Kármán interventions with predeclared compatible roles, windows and estimands for target, PPO, executable SR, constant and zero; do not assume that all roles can share identical phase semantics. Paired or independently parented realizations should be used where the design requires them. The action roles should then be compared at matched amplitude and sign, with the field evaluator carrying the same-chain L2 and case-specific DTW contracts. Second, Vortex requires an audited target/control/zero event acquisition and an event-relative downstream-ROI evaluator at the declared offsets; Erase requires an independently defined target scale and supported roles before any quantitative panel. Third, the persistent/dynamic interpretation requires a complete control-volume Navier–Stokes, momentum and mechanical-energy budget, including the terms currently missing from the partial ledger. Finally, crossing from CFD to experiment requires synchronized, calibrated PIV and body-resolved force measurements with registration, clocks, background blanks, declared repeats and the Module 09 provenance gate. These tests would raise specific claim ceilings; they would not retroactively merge chains or make non-equivalent metrics interchangeable. [M03; M05–M09; R1 experiment/case-metrics dossier]
+
+## Three-paragraph conclusion map
+
+### Conclusion paragraph 1 — answer in evidence order
+
+Open with the named capability: learned control provides wake-signature or visual PPO capability for the declared Kármán, Vortex and Illusion task demonstrations, with the cross-case contracts kept separate. Move immediately to the independent Legacy Kármán spatial result: field matching is supported only for that named case by the complete-cycle mean and phase-resolved L2 comparisons, while DTW is complementary signal evidence. Do not introduce a new number, citation or figure here.
+
+### Conclusion paragraph 2 — strongest interpretation and ceiling
+
+State that the Legacy Kármán action and field evidence supports a bounded control-skeleton interpretation: dominant persistent rear rotation plus a smaller dynamic correction. Name the SR deletion ranking and corrected four-role mean/dynamic CCD organization only as non-equivalent evidence. Make the mandatory boundary explicit: the M04 standardized PPO and M06 corrected CCD ledgers are different estimands and do not provide independent triangulation. The interpretation remains descriptive and bounded; it does not establish a unbounded law, causal explanation, necessity, sufficiency, uniqueness or claim of time-invariant behaviour.
+
+### Conclusion paragraph 3 — limits and one decisive testing program
+
+Close with the asymmetric envelope: Vortex has no event L2 yet; Illusion PPO is separate from negative Illusion SR; Erase lacks a supported quantitative route; steady 3.431 and s=3.55 are contextual diagnostics; theory and budgets remain incomplete; the experiment interface remains author-described and provenance-gated. Name the decisive program compactly: matched same-protocol interventions and roles, event evaluator/acquisitions, same-chain amplitude/sign comparisons, independent parent realizations where needed, complete budgets, and synchronized calibrated PIV/force measurements. The paper therefore establishes named capability across tasks, Legacy Kármán field matching, and a bounded Legacy Kármán control-skeleton interpretation, with the next ceiling set by direct intervention, closure and measurement tests.
+
+## Layer 2 — claim-to-evidence and interface ledger
+
+| Claim or boundary | Exact source/artifact | Chain, solver/geometry, case/role, Re | Comparator, metric/estimand, ROI/window | Status |
+|---|---|---|---|---|
+| Named capability and Legacy Kármán field matching | `ROUND3_BOTTOM_UP/04_KARMAN_PERFORMANCE.md`; `03_CONTRACTS_METRICS.md` | Legacy Re100 Kármán principal; separate V5 Re100 secondary | Named wake-signature/visual PPO capability under distinct cross-case contracts; Legacy target/controlled/zero complete-cycle mean and eight-slot phase L2 in registered ROI; DTW complement | Read; raw fields/masks artifact-audit deferred |
+| Persistent rear term and smaller feedback | `ROUND3_BOTTOM_UP/05_KARMAN_SR_LAW.md`; `tables/term_deletion.md` | Legacy Kármán re-codes 50/100/200/400; native Legacy action | Parent-relative fixed 40-step native DTW deletion; rear constant largest tested | Read; deletion is not causal ablation |
+| Corrected mean/dynamic organization | `ROUND3_BOTTOM_UP/06_KARMAN_FIELD_CCD.md` | Legacy Re100; target, zero, constant mean, DRL; exact common mask | Coordinate-weighted mean error; separately centred non-paired 19×10 residual; CCD action association | Read; descriptive/noncausal |
+| M04/M06 non-equivalence | `00_WAVE_A_COORDINATOR_AUDIT.md` §§28–37; M04/M06 exports | Distinct acquisitions, role sets, windows, masks and chains/contracts | M04 unweighted U0-normalized geometry guard versus M06 coordinate-weighted exact common mask | Read; never arithmetic combination/triangulation |
+| Cross-case ceiling | `ROUND3_BOTTOM_UP/07_CROSS_CASE_EXTENSIONS.md` | Legacy Vortex; V5 Illusion PPO; Legacy Illusion SR/Erase history | No Vortex event L2; Illusion PPO capability; Illusion SR negative; Erase unsupported | Read; gaps preserved |
+| Steady/theory/experiment limits | `ROUND3_BOTTOM_UP/08_STEADY_SYNTHESIS.md`; `ROUND3_BOTTOM_UP/09_EXPERIMENT_AUTHOR_INTERFACE.md`; `ROUND1_BOTTOM_UP/08_EXPERIMENT_AND_CASE_METRICS.md` | Separate V5 steady and Legacy Kármán; author-described open-loop water-tunnel interface | SR 3.431 and V5 s=3.55 contextual; failed sign; incomplete budgets; experiment observations conditional on raw-media provenance | Read; M09 interface ready, final prose/panel blocked |
+| Decisive tests | M03, M05–M09 exports; R1 experiment/case metrics | Matched same-chain roles/interventions; exact event roles; calibrated experiment | Same-protocol metrics, event L2, amplitude/sign, complete budgets, synchronized PIV/force | Open tests, not results |
+
+### Interface ledger
+
+- **Imports:** M03 chain, role, action, Reynolds and metric contracts; M04 Legacy Kármán field-matching result; M05 executable SR/deletion ceiling; M06 corrected four-role mean/dynamic and CCD boundary; M07 cross-case gaps; M08 steady/theory bridge; M09 author-described experiment interface and provenance gate.
+- **Export to final audit:** named capability across tasks, Legacy Kármán field matching, and a bounded Legacy Kármán control-skeleton interpretation. No export carries a pooled metric, cross-chain continuation, causal mechanism, unbounded law or quantitative experiment result.
+- **Genuine blockers:** none for this bounded first-pass text. Retained quantitative Vortex/Erase panels remain blocked by role/evaluator contracts; quantitative experiment remains blocked by missing synchronized calibrated PIV/force data. Hidden artifacts are `artifact-audit deferred` where authorities bind the current claim without conflict.
+
+### Coverage ledger
+
+- **`read`:** Round-3 controls/interface and Wave-A audit; Round-2 freeze/style; ready Modules 03–09; Round-1 experiment/case-specific metric dossier; current authorities cited above. Evidence is read, not independently recomputed or universally artifact-audited.
+- **`mapped`:** package manifests, figure navigation and authority-linked external result roots referenced by Modules 03–09. Mapped is not read, independently verified or artifact-audited.
+- **`not-covered`:** new literature, CFD, retraining, event evaluator, Vortex/Erase acquisitions, complete budgets, synchronized experiment processing and new figures.
+- **`external-inaccessible`:** hidden/symlink-backed raw fields, selected model/normalizer bundles and dense theory/CCD payloads where current authorities bind bounded statements; artifact audit deferred.
+
+## Figure, table and equation contracts
+
+1. **Integrated evidence map:** orders capability, Legacy L2, SR deletion, corrected mean/dynamic CCD, cross-case, steady, theory and experiment. Dashed links mean compatible but non-equivalent observations; unresolved links mark causality and closure gaps. It does not depict a causal chain or pooled score.
+2. **Compact limitations table:** lists chain, case, role, estimand and blocked claim for Vortex, Illusion SR, Erase, steady, theory and experiment. It does not add results or convert missing artifacts into negative findings.
+3. **No new equation or numerical figure is required for the conclusion.** Existing panels retain their own captions and limitations; this first pass does not imply that all proposed panels already exist.
+
+## Claim ladder
+
+**Supported / green:** named learned control provides wake-signature or visual PPO capability under the declared cross-case contracts; independent Legacy Kármán field matching shows the controlled field is closer to target than physical zero under its mean and phase L2 estimands; executable Legacy SR deletion ranks the persistent rear term largest among tested terms; corrected CCD organizes a smaller residual descriptively as action-correlated correction-field modes.
+
+**Bounded / amber:** these results support a Legacy Kármán control-skeleton interpretation consisting of dominant persistent rear rotation plus a smaller dynamic correction; separate steady diagnostics provide contextual information; cross-case demonstrations show named capability breadth under changed contracts.
+
+**Prohibited / red without new evidence:** unbounded transfer, causal explanation, necessity, sufficiency, uniqueness, claim of time-invariant behaviour, momentum restoration, independent triangulation, cross-chain continuation, quantitative Vortex or Erase efficacy, positive Illusion SR, or quantitative experiment claims.
+
+## Anti-repetition and first-pass audit
+
+- **Evidence order:** named capability → Legacy Kármán L2 → executable SR/deletion → corrected four-role mean/dynamic CCD → cross-case breadth → separate steady context → author-described qualitative experiment interface; the conclusion map follows the same order.
+- **Near-collision:** M04/M06 are explicitly different estimands; no arithmetic, replication, validation, agreement or triangulation wording is used. SR 3.431 and steady s=3.55 are related contextual diagnostics only; CCD constant mean is not action magnitude.
+- **Boundary:** no Vortex event L2 is invented; Illusion PPO is not merged with negative Illusion SR; Erase is not repaired from filenames; steady/theory/experiment do not raise the Kármán ceiling.
+- **Terminology:** wake-signature matching, field matching, spatial L2, complementary DTW, control skeleton and action-correlated correction-field modes are used; causal and unbounded readings are excluded.
+- **Layer control:** Layer 1 has four discussion paragraphs plus a three-paragraph conclusion map; Layer 2 carries provenance, coverage, interfaces and contracts. This is a revised candidate for second audit, not a readiness declaration.

docs/JFM_WYQ/ROUND3_BOTTOM_UP/11_FINAL_CONSISTENCY_AUDIT.md

@@ -0,0 +1,128 @@
+# Round 3 coordinator-level final consistency audit
+
+**Status: audit, not manuscript approval.** This coordinator audit reads the Round-2 freeze/style controls, all active Round-3 controls, and Modules 01–10 in full. It checks cross-module terminology, claims, contracts, bridges, redundancy and readiness; it does not recompute scientific artifacts, edit source modules or ledgers, regenerate figures, reconstruct masks, verify hashes, or validate the experiment.
+
+## 1. Scope and complete coverage ledger
+
+**Controlling identity:** *Interpreting learned hydrodynamic cloaking and illusion in a fluidic pinball*. The active controls are the Round-2 `05_REVISED_STRUCTURE_FREEZE.md` and `06_SECTION_LEVEL_JFM_STYLE_GUIDE.md`; Round-3 `.cursor/rules/jfm-round3-section-drafting.mdc`; `00_MASTER_AGENT_PROMPT.md`; `00_QUALITY_GATE.md`; `00_MODULE_INTERFACE_LEDGER.md`; `00_WAVE_A_COORDINATOR_AUDIT.md`; and active Modules `01_INTRODUCTION.md`–`10_LIMITATIONS_CONCLUSION.md`. Module 11 is this audit, not a manuscript section.
+
+| Status | Complete coverage | Boundary |
+|---|---|---|
+| `read` | Round-2 freeze/style; all five Round-3 controls/audits; Modules 01–10, each in full | Read is not independently verified or artifact-audited |
+| `mapped` | Manifests, figure navigation, authority-linked external/symlink paths and artifact identities named by modules | Mapped is not read or numerical authority |
+| `not-covered` | New CFD/training/SR/CCD; raw NPZ/dense arrays; universal hashes; solver-exact mask reconstruction; event evaluator; new figures; complete Erase roles; synchronized experiment processing; new literature search | Outside consistency audit; no claim inferred |
+| `external-inaccessible` | Hidden/external fields, model/normalizer bundles, dense CCD/theory payloads and external/raw facility artifacts | Authority-led `artifact-audit deferred` only where current authority binds a conflict-free bounded statement; retained new panels remain gated |
+
+All active modules are included in the read ledger: M01 Introduction; M02 capability progression; M03 contracts/metrics; M04 Kármán performance; M05 Kármán SR law; M06 Kármán field/CCD; M07 cross-case extensions; M08 steady synthesis; M09 experiment interface; M10 limitations/conclusion. The interface ledger’s `ready` statuses mean ready for downstream dossier use, not manuscript approval or publication-ready artifacts.
+
+## 2. Global terminology audit
+
+Terminology is consistent and should remain frozen:
+
+- **Concept:** `hydrodynamic cloaking` means approach to a declared background/reference; `hydrodynamic illusion` means approach to a declared non-zero target/signature. Neither means universal invisibility or full-state equivalence.
+- **Execution:** `wake-signature matching using forces plus sparse downstream observations`. This does not establish observability or optimal sensing.
+- **Evaluation:** **field matching**, with spatial **L2** primary where supported. **DTW** is complementary, signal-level and case-specific; it is not spatial L2, physical delay, causal diagnosis or a universal cross-case score. Native/normalized DTW and similarity orientation must stay explicit; a higher similarity must never be called a lower error.
+- **Interpretation:** **control skeleton** is a bounded post-evidence interpretation, not a mechanism, unique law, necessity claim or cross-case result.
+- **Dynamic fields:** use exactly **action-correlated correction-field modes**. CCD is descriptive/noncausal and not superior or unique relative to POD.
+
+Pipeline `Level-2/Level-3` and `Stage-3 short/standard gate` remain distinct from spatial L2. Never call a validation level an “L2 gate.”
+
+## 3. Manuscript claim ladder audit
+
+The dossier preserves exactly this ladder:
+
+1. **Named cross-task capability:** named wake-signature or visual PPO capability across declared Kármán, Vortex and Illusion tasks under distinct contracts.
+2. **Legacy Kármán Re100-only independent field matching:** independent spatial field matching only for named Legacy `karman_re100`/Re100 target, controlled and physical-zero roles in the declared downstream ROI; mean and phase L2 remain separate, with DTW complementary.
+3. **Bounded Legacy Kármán/cloak-only control-skeleton interpretation:** executable Legacy SR, corrected four-role mean comparison and a separately centred non-paired residual organized by action-correlated correction-field modes support a bounded dominant persistent rear-rotation plus smaller dynamic-correction reading. This is not causal, necessary, sufficient, unique, optimal, stable, universal or transferable.
+
+M01 and M10 state this ladder; M02 and M04–M09 respect it. Positive SR remains Kármán/cloak-only. Illusion SR is historical/negative, CCD descriptive/noncausal, OID claim-free and steady contextual.
+
+## 4. Legacy/V5 and metric-boundary audit
+
+Legacy and V5 are separate executable chains: never pool values or imply numerical continuation/equivalence. Preserve solver and geometry, body order, action sign/calibration, case/seed/role, comparator, ROI/mask and window in every result. Legacy `Re_code=100` is `Re_D=50` under its Legacy `2D` convention; V5’s similarly labelled Re100 is its own chain. V5 Illusion `0.75` is a target radius ratio under the declared geometry, not silently a diameter ratio. Best-checkpoint discovery, retained evaluation and seed spread are not robustness.
+
+Periodic complete-cycle mean L2, eight-slot phase L2, transient event snapshots, Illusion mean/phase/DTW, Erase candidate mean fields and steady late/profile errors are non-equivalent estimands. Vortex offsets are event-relative, not phase. Steady profiles are not periodic means. Case-specific DTW windows/formulas cannot be pooled.
+
+### Explicit M04/M06 near-collision prohibition
+
+M04 standardized PPO mean errors and M06 corrected CCD mean errors are **different estimands**, despite close magnitudes and nominally coincident x extent. M04 uses three independently acquired target/controlled/physical-zero roles, a geometry guard, unweighted lattice-cell RMS after `U0` normalization and its complete-cycle/phase8 window. M06 uses target/zero/constant/DRL, a retained four-role window, an exact common solver-fluid mask and coordinate-weighted RMS on already nondimensional velocities. Never equate, subtract, average or cross-normalize them; never call them replication, validation, agreement, closure or independent triangulation. M06’s `93.42/6.58` share exists only in its corrected ledger.
+
+## 5. Section sequence and bridge audit
+
+The frozen order is:
+
+`Introduction → contracts/metrics → capability progression → Kármán performance → Kármán SR law → Kármán correction fields/CCD → Vortex/Illusion/Erase extensions → steady context/theory → qualitative experiment → limitations/conclusion`.
+
+The physical progression is explicitly **steady discovery → Kármán → Vortex → Illusion/Erase → qualitative experiment**. The full steady analysis returns later, after Kármán depth and cross-case extensions; this two-pass use preserves chronology without placing the full steady analysis first.
+
+Bridges are coherent: M01 converts gap to contracts; M03 enables bounded progression; M02 selects Kármán depth; M04 turns performance into action-law complexity; M05 turns term ranking into field-accounting; M06 prevents Kármán interpretation transfer; M07 changes temporal/target contracts; M08 contextualizes the persistent reference and theory limits; M09 separates deployment from validation; M10 closes with claim ceiling and decisive tests.
+
+### Duplicate ownership/deletion recommendations
+
+These are synthesis-stage recommendations only; no module is edited:
+
+- M01 P5, M02 P5 and M10 repeat the three-level ladder. M01 owns the full definition; M02 keeps only progression scope; M10 restates it once in conclusion order. Delete package-list repetition.
+- M02 P3 and M04 P1–3 repeat Kármán L2 values. M02 owns the short Fig. 1 preview; M04 owns all detailed values, roles, ROI, windows and interpretation; remove duplicate result sentences during assembly.
+- M04/M06 interface text, M06 P1–2, M08 P5 and M10 P2 repeat the near-collision rule. Keep the full technical statement in the coordinator/M06 interface and shorten later reminders.
+- M05 owns SR formula/deletion; M08 owns V5 steady sign/magnitude context. Remove repeated formula/sweep or any implication of amplitude agreement.
+- M06 owns residual/CCD/POD detail; M08/M10 retain only synthesis-level noncausal limits.
+- M07 owns Illusion PPO versus negative SR; M10 states only the boundary. M09 owns the experiment provenance checklist; M10 retains only the qualitative/open-loop handoff.
+
+## 6. Figure/table/equation readiness matrix
+
+No status below implies panels already exist.
+
+- **Fig. 1:** text-ready as Kármán-dominant progression; steady, Kármán, Vortex and V5 `ill_075L` panels are dependency-gated. No experiment panel. Optional Re400 marker is drop-first.
+- **Plant/action/ROI contracts:** text/table-ready; redraw is production work and does not prove observability, chain equivalence or exact masks.
+- **Kármán L2:** text-ready from bound summaries; panel-ready only with raw role/target/manifest/rendering support. Geometry-guarded ROI and mask limitation remain captioned.
+- **SR:** equation/text-ready; deletion table is bounded. Standardized PPO/SR/Zero panel is dependency-gated and must be omitted without exact same-protocol artifacts.
+- **CCD:** text-ready under corrected four-role estimand; panels are dependency-gated on source publication arrays. Mean precedes residual/modes; POD cannot imply CCD superiority.
+- **Cross-case:** text-ready for bounded Vortex/Illusion capability. Vortex quantitative event L2 is **blocked** pending exact roles, ROI/mask, offsets and evaluator; Illusion field L2 is conditional; Erase quantitative panel is **blocked**.
+- **Steady:** text-ready as contextual V5 endpoint/sign/theory limit; profiles, reverse/MFS panels are dependency-gated, with MFS blocked until gates close.
+- **Experiment:** author-supply/provenance-gated; no quantitative validation panel.
+- **Integrated map/limitations table:** text-ready as noncausal synthesis aid; not a pooled score or causal graph. Conclusion needs no new equation/figure.
+
+## 7. Literature/citation identity audit
+
+The style guide records ten exact JFM identities at `full-PDF` level: Deng et al. (2020), Cornejo Maceda et al. (2021), Rabault et al. (2019), Paris et al. (2021), Li & Zhang (2022), Xia et al. (2024), Wang et al. (2024), Loiseau et al. (2018), Gautier et al. (2015) and Mao et al. (2015), with DOIs retained there. M01 records the exact DOI identities for Pendry et al. (2006), Leonhardt (2006), Vasquez et al. (2009), Ma et al. (2013) and Urzhumov & Smith (2012). Repository dossiers are labelled `full-report`/`read` where applicable; literature supplies context/style precedent only, never internal evidence or priority.
+
+Unresolved gaps: `Ren19b`, exhaustive sentence-level JFM coding, broader facility literature, and any metadata/read-level identity not already established in the controlling dossiers. Do not invent citations, DOI, quotations or priority language. No new citation or verification is added here.
+
+## 8. Experiment audit
+
+The experiment is **author-described qualitative open-loop dye only**, and provenance-gated. Allowed observations are that steady operation appeared to produce straighter dye streaklines and Illusion operation visibly changed shedding frequency, subject to raw-media provenance and trial linkage. There is no synchronized calibrated force/PIV validation, quantitative field match, closed-loop experimental validation, quantitative shedding frequency or causal explanation. Thermal convection is a suspected confound, not a proven sole cause. Missing apparatus, conditions, action waveform/sign, trial IDs, raw media, scale/timing, calibration, synchronization, repeats and permissions must be supplied or marked unrecoverable; otherwise omit the panel.
+
+## 9. Findings with severity and exact location
+
+`PASS` means text-ready for author synthesis, not manuscript approval. `REVISE` names the targeted file/location. `BLOCK` applies only to required content that cannot responsibly be drafted; artifact blockers are separated from module-text readiness.
+
+### PASS — module text readiness
+
+- M01 `01_INTRODUCTION.md`; M02 `02_CAPABILITY_PROGRESSION.md`; M03 `03_CONTRACTS_METRICS.md`; M04 `04_KARMAN_PERFORMANCE.md`; M05 `05_KARMAN_SR_LAW.md`; M06 `06_KARMAN_FIELD_CCD.md`; M07 `07_CROSS_CASE_EXTENSIONS.md`; M08 `08_STEADY_SYNTHESIS.md`; M09 `09_EXPERIMENT_AUTHOR_INTERFACE.md` as an author interface; and M10 `10_LIMITATIONS_CONCLUSION.md` all preserve their scope and claim ceilings. M09’s final prose is not yet supplied, by design.
+
+### REVISE — exact synthesis changes
+
+- `01_INTRODUCTION.md` P5 and `10_LIMITATIONS_CONCLUSION.md` Discussion/Conclusion: retain the exact three-level ladder once as full definition and once as concise conclusion; delete repeated package-list versions from M02/M04/M08/M10.
+- `02_CAPABILITY_PROGRESSION.md` P3 and `04_KARMAN_PERFORMANCE.md` P1–3: M02 may preview Fig. 1; M04 owns detailed Legacy L2/DTW values and contracts. Delete duplicate numerical result prose during assembly.
+- `06_KARMAN_FIELD_CCD.md` P1–2 and `08_STEADY_SYNTHESIS.md` P5/`10_LIMITATIONS_CONCLUSION.md` P2: retain the full M04/M06 different-estimand prohibition once; later mentions must not imply agreement or independent validation.
+- `05_KARMAN_SR_LAW.md` P3–5 and `08_STEADY_SYNTHESIS.md` P5: keep SR formula/deletion and steady sign/magnitude in separate ownership; delete any amplitude-agreement or continuation implication.
+- `09_EXPERIMENT_AUTHOR_INTERFACE.md` §§2–Figure gate and `10_LIMITATIONS_CONCLUSION.md` P3: M09 owns provenance; M10 keeps only qualitative/open-loop/provenance-gated status.
+- Final manuscript assembly across all files: run a whitespace and LaTeX terminology pass for `Kármán`, `Re_D`, `Re_code`, `L2`, `DTW`, `E_mean`, `E_phase8`, `\text{}` labels, action-order notation and escaped `%`; choose display notation once without changing chain-local conventions.
+
+### BLOCK — retained content only
+
+- `02_CAPABILITY_PROGRESSION.md` Fig. 1 Vortex event-L2 panel: blocked until exact target/control/zero roles, target identity, common ROI/mask, offsets and evaluator output exist.
+- `07_CROSS_CASE_EXTENSIONS.md` Erase quantitative panel/efficacy: blocked by unsupported target-scale and independent-role contract; filenames do not establish results.
+- `05_KARMAN_SR_LAW.md` standardized PPO/SR/Zero numerical panel: blocked unless exact same-protocol role artifacts and comparator/window values are bound; text is unaffected if dropped.
+- `09_EXPERIMENT_AUTHOR_INTERFACE.md` image/result panel: blocked until raw-media provenance, trial linkage, conditions and permission pass; no quantitative validation claim.
+- `08_STEADY_SYNTHESIS.md` new MFS field panel: blocked until listed validation gates close; the bounded theory/sign failure remains usable.
+
+These are panel/artifact blockers, not blockers to the bounded text. No current blocker prevents author synthesis of the text dossier.
+
+## 10. Final disposition
+
+**READY FOR AUTHOR MANUSCRIPT SYNTHESIS, with this file remaining an audit rather than manuscript approval.** Before polished prose, the author must adopt the fixed terminology and exact claim ladder; assemble the frozen sequence and delete duplicated result paragraphs according to the ownership map; make M04 the sole detailed owner of Legacy Kármán independent field matching; preserve the M04/M06 prohibition; decide which optional panels are dropped or sourced; complete the LaTeX/whitespace pass; and supply or omit the experiment provenance-gated material.
+
+**Author supply list:** authority-bound source/rendering for each retained panel; same-protocol PPO/SR/Zero artifacts if retained; audited Vortex event roles/evaluator; independent Erase roles if proposed; validated MFS artifacts if retained; experiment facility/geometry/conditions, action waveform/sign, trial IDs, raw media, calibration/synchronization, repeats and permissions; and final bibliography/read-level confirmation without invented metadata or priority claims.
+
+This is a final **consistency audit**, not manuscript approval.

docs/JFM_WYQ/ROUND3_BOTTOM_UP/README.md

@@ -0,0 +1,16 @@
+# Round 3 — Bottom-Up Section Modules
+
+Round 3 converts the Round-1 evidence dossiers and current package authority into author-readable, section-level drafts/dossiers approaching JFM prose. It is neither another raw index nor polished final manuscript prose. The parent agent coordinates dispatch, interfaces, audits and revisions; section agents do not self-declare manuscript readiness.
+
+## Controlling inputs
+
+Every module reads, in full or with explicit scoped coverage, `ROUND2_TOP_DOWN/05_REVISED_STRUCTURE_FREEZE.md`, `06_SECTION_LEVEL_JFM_STYLE_GUIDE.md`, relevant Round-1 dossiers, and relevant current authority documents/manifests. Hidden/raw/external artifacts may be deferred only when an authority document describes the deferral. Legacy and V5 remain separate, and every number retains its artifact, comparator, metric and window.
+
+## Phases
+
+- **Wave A — hard evidence:** dispatch contracts/metrics, Kármán performance, Kármán SR law, and Kármán field/CCD modules. The parent reviews first passes for evidence completeness, narrative coherence/JFM style, interfaces, claim ceiling and redundancy, then returns targeted revision prompts. A module is ready only after the second audit.
+- **Wave B — extensions and synthesis:** dispatch cross-case vortex/illusion/erase, then steady context/physical synthesis. Keep case-specific metrics distinct; transfer only explicitly licensed definitions or bounded context.
+- **Wave C — framing:** draft introduction, capability/progression, qualitative experiment author interface, and limitations/conclusion after the analytical modules stabilize. Frank retains authorship of the experiment subsection.
+- **Final audit:** coordinator performs whole-manuscript terminology, chain, metric, bridge, claim-ceiling and redundancy consistency audit; no new evidence is invented during synthesis.
+
+See `00_MASTER_AGENT_PROMPT.md`, `00_QUALITY_GATE.md`, and `00_MODULE_INTERFACE_LEDGER.md` for dispatch, review and interface contracts.

docs/JFM_WYQ/ROUND4_TOP_DOWN/00_DISPLAY_EQUATION_INVENTORY.md

@@ -0,0 +1,33 @@
+# Round 4 Display and Equation Inventory
+
+**Inventory only.** No candidate panel, source array, rendering, table, or publication-ready figure exists merely because it is inventoried. Text readiness and panel/artifact readiness are independent.
+
+Classifications: `required`, `optional`, `dependency-gated`, `blocked`, `author-supply-gated`.
+
+| Display/equation family | Classification | Owner | Argumentative job | Readiness/dependency | Fallback |
+|---|---|---|---|---|---|
+| **Fig. 1 capability/progression** | `dependency-gated` | Capability | Steady discovery → Kármán → Vortex → `ill_075L`, Kármán dominant, configurations and evidence definitions distinct | Text architecture ready; each panel needs chain/case/seed/role/comparator/coordinate/metric/source/render provenance. Vortex event L2 blocked; Illusion PPO only; no experiment. Re400 optional/drop-first | Retain sourced panels; omit blocked annotations/Re400 |
+| **Section 2 two-configuration geometry and DRL-flow schematics/tables** | `required` | Problem/method/evaluation | Decode parabolic/no-slip and uniform/free-slip configurations; body/action/sign; $x/D=10$, $y/D=0,\pm2$, radius-$0.25D$ two-component sensing; target/reward flow; ROI/roles/$Re_D$/metrics; compact CFD | Architecture frozen; sensor note is `read` and fixes geometry/ROI/metric boundaries; CelerisLab dossier is `read/full-report` and fixes K2 `partial pass` / S2 `pass` for current CelerisLab only. Redraw, per-result rows and artifact binding remain production dependencies. Does not prove observability, optimal sensing, configuration equivalence, canonical rerun completion, an exact mask, or older-solver validation; makes no Fig. 1 decision | Minimal geometry + DRL flow schematic and compact tables; provenance/detail to supplement |
+| **Kármán performance display** | `dependency-gated` | Kármán performance | L2-first parabolic/no-slip mean/phase result; DTW complementary; uniform/free-slip retention secondary | Text/numbers bound; fields need exact roles, manifests, ROI/mask/window, rendering. Separate from M06 | Compact result table; omit unsourced redraws |
+| **Section 2 method/evaluation equations** | `required` minimum | Problem/method/evaluation; reused by performance | Define bulk/mean-based $Re_D$, physical actuation, objective roles, `E_mean`, phase-resolved L2 and optional comparator-relative form; give DTW recurrence/path once because uncommon | Definition-ready at architecture level; current defaults source-bound; exact per-result settings and values pending, including learning-rate variation across retained runs. Spatial L2 is not Level/Stage; DTW is not delay/field equivalence | Keep only reused equations; move algorithm/settings detail to supplement |
+| **SR law/deletion** | law/deletion `required`; PPO/SR/Zero panel `dependency-gated` | SR | Exact executable law/symmetry; deployment; largest-tested persistent term | Law/deletion text-ready. Same-protocol panel blocked without exact roles/comparator/window. Deletion noncausal | Keep law, symmetry, deletion; omit direct role panel |
+| **CCD mean/share display** | `dependency-gated` | Field/CCD | Target/zero/constant/DRL means before residual; scalar error-difference accounting | Ledger bound; redraw needs publication arrays and exact mask/window. Cannot validate M04 | Compact scalar ledger if fields unavailable |
+| **CCD residual/modes/POD sequence** | `dependency-gated` | Field/CCD | Define non-paired residual; show action-correlated modes, phase, POD boundary | Equation/text ready; panels need corrected parent/derived arrays, orientation, lineage, captions. No causality/lag/energy/superiority | Keep residual equation and concise statement/table |
+| **Cross-case display** | overall `dependency-gated`; quantitative Vortex/Erase `blocked` | Cross-case | Event-relative Vortex, non-zero-target Illusion PPO, Erase boundary | Vortex, Illusion and Erase physical-variable role fields are available/mapped candidates. Quantitative Vortex and Erase products remain blocked pending versioned evaluators; Illusion is bounded PPO capability and its exact scalar binding is conditional | Use sourced visuals; omit quantitative Vortex/Erase and state gaps |
+| **Cross-case event equation** | `optional` | Cross-case | Define conditional event-relative ROI error without calling offsets phase | Definition creates no roles/mask/evaluator/result | Use prose/table semantics |
+| **Steady/theory display** | `dependency-gated` | Steady/theory | Separate uniform/free-slip endpoint/sign context; accepted/reverse boundary; failed closure | Existing finite-Re and outer-reference assets are bounded redraw sources; reverse amplitude mismatch must remain in the caption. No new MFS computation is authorized; Frank’s main equation/composition choice remains open | Sign schematic + endpoint table + one-circle/sign conflict; omit MFS field |
+| **Steady/theory equations** | `required` minimum | Steady/theory | Wall law/spin ratio, steady metric, and shortest outer-reference or partial-budget boundary | Existing derivation/check states distinct; no inviscid rotating no-slip closure | Full MFS/multipole/strip/CV derivations to supplement |
+| **Experiment display** | `author-supply-gated` | Experiment | Qualitative open-loop apparatus and, if gated, steady/Illusion dye observations | Needs Frank's conditions, action/sign, trials, raw media, scale/frame rate, processing, repeats, permissions. No calibrated force/PIV panel | Omit images; retain short author-written interface or bounded supplement media |
+| **Conclusion display/equation** | `omitted` | Conclusion | Text-only synthesis under current author correction | No conclusion-owned display, table, equation, citation or new number | No display; optional author-owned limitations/outlook remains outside the frozen synthesis |
+
+## Equation ownership
+1. Problem/method/evaluation owns reusable bulk/mean-based $Re_D$, action/sign, objective/role, DTW, and field-metric definitions.
+2. SR owns reflection map and canonical law; no full reprint elsewhere.
+3. Field/CCD owns scalar shares and residual; CCD operator only if interpretively necessary.
+4. Cross-case event equation is conditional, never a result.
+5. Steady/theory owns `s`, steady metric, and selected outer-reference/budget equation.
+6. Conclusion introduces no equation, numerical result, citation, or artifact.
+7. Freeze notation once for `Kármán`, `Re_D`, provenance-only `Re_code`, spatial L2, DTW orientation, ROI/mask, action order/sign, and percentages; delete duplicates or move them to supplement. The reference parabolic-source case carries no reader-facing Reynolds value until inherited-label conversion closes.
+
+## Panel readiness check
+Before calling any item panel-ready, record exact authority/artifact; chain/solver/geometry; case/seed/role; comparator; metric/estimand; ROI/domain/mask; weighting/normalization; window/phase/event coordinate; array orientation; rendering provenance; caption limitations; and artifact-audit state. Missing required fields mean `dependency-gated`, `blocked`, or `author-supply-gated`, with the listed fallback.

docs/JFM_WYQ/ROUND4_TOP_DOWN/00_GLOBAL_DECISION_LEDGER.md

@@ -0,0 +1,152 @@
+# Round 4 Global Decision Ledger
+
+**Owner:** Round-4 coordinator.
+**Initial status:** `discuss`; unresolved author choices are not frozen.
+**Role:** architecture/ownership/claim/supersession source of truth, not prose or evidence.
+
+## Controlling decisions carried from R2/R3
+| Decision | Disposition | Boundary |
+|---|---|---|
+| Provisional title: *Interpreting learned hydrodynamic cloaking and illusion in a fluidic pinball* | carried, provisional | No colon/opening “A”; no positive Illusion SR implied |
+| Concept = cloaking/illusion; execution = wake-signature matching; evaluation = field matching with L2 primary/DTW complementary; interpretation = control skeleton | frozen input | R2 freeze/style and R3 audit |
+| The periodic Kármán cloak under the parabolic-inflow/no-slip-wall configuration is the sole deep case | frozen input | Cross-cases do not inherit interpretation; Reynolds naming is owned by Section 2 and bare `Re100` is prohibited |
+| The parabolic-inflow/no-slip-wall and uniform-inflow/free-slip-wall configurations are separate | frozen input | Never pool, continue, validate, or imply equivalence; internal chain IDs remain provenance-only |
+| Physical progression: steady discovery → Kármán → Vortex → Illusion/Erase → qualitative experiment | frozen input | Early steady seed; later full steady/theory return; experiment is a short author-owned main-text subsection near the end |
+| Compact CFD qualification | frozen input | Selected observables only |
+| Positive SR Kármán/cloak-only; Illusion SR historical/negative | frozen input | No revived positive claims |
+| `action-correlated correction-field modes`; OID claim-free | frozen input | CCD descriptive/noncausal; no OID result |
+| Case-specific metrics | frozen input | Periodic/event/Erase/steady/DTW non-equivalent |
+| Experiment qualitative, open-loop, author-written, provenance-gated | frozen input | No quantitative validation |
+| R3 modules text-ready, not manuscript/panel-ready | frozen input | R3 final audit |
+
+## Global claim ladder
+1. **Named cross-task capability:** named wake-signature or visual PPO capability across declared Kármán, Vortex, and Illusion tasks under distinct contracts.
+2. **Periodic Kármán cloak under the parabolic-inflow/no-slip-wall configuration — independent field matching:** spatial field matching only for the named target, controlled, and physical-zero roles; mean and phase L2 separate, DTW complementary.
+3. **Periodic Kármán cloak under the parabolic-inflow/no-slip-wall configuration — bounded control-skeleton interpretation:** executable SR, corrected four-role mean comparison, and non-paired residual organized by action-correlated correction-field modes support dominant persistent rear rotation plus smaller dynamic correction.
+
+Level 3 is not causal, necessary, sufficient, unique, optimal, stable, robust, universal, transferable, or independent triangulation. Cross-case capability cannot inherit Levels 2–3.
+
+## Frozen section order
+Round-4 freeze order:
+`problem/method/evaluation → capability → Kármán performance → SR → field/CCD → cross-case → steady/theory → experiment → conclusion → Introduction (last)`
+
+Assembled manuscript order:
+`Introduction → problem/method/evaluation → capability → Kármán performance → SR → field/CCD → cross-case → steady/theory → experiment → conclusion`.
+
+Introduction freezes last so framing cannot outrun settled results.
+
+## Ownership and duplication
+| Owner | Owns | Allowed elsewhere | Prohibited duplication/inference |
+|---|---|---|---|
+| Introduction | Full ladder, concept/gap, contribution | Conclusion concise restatement | Detailed values, packages, priority/mechanism |
+| Problem/method/evaluation | Configurations/action/Re/roles/objective/metrics/ROI/windows/compact CFD | Results import definitions | Results narrative or configuration/evidence transfer |
+| Capability | Fig.1 progression, short Kármán preview, steady seed | Performance owns values; steady owns detail | Pooled efficacy, full result prose, experiment panel |
+| Kármán performance | Detailed parabolic-inflow/no-slip-wall L2/DTW; separate uniform-inflow/free-slip-wall seed/retention boundary | Capability headline only | M06 as replication/agreement |
+| SR | Parabolic-inflow/no-slip-wall formula/symmetry/deployment/deletion | Steady recurrence context only | Full steady/field detail, causal ablation, positive Illusion SR |
+| Field/CCD | Corrected means/shares, residual, CCD/POD boundary | Synthesis brief statement | SR derivation, causal response, M04 closure |
+| Cross-case | Vortex boundary, Illusion PPO/negative SR, Erase gap | Capability preview; conclusion envelope | Kármán transfer, common metric |
+| Steady/theory | Uniform-inflow/free-slip-wall endpoint/sign/context, theory failure/budgets | Capability seed; conclusion limit | Cross-configuration continuation, amplitude agreement, mechanism/optimum |
+| Experiment | Author provenance and qualitative observation | Conclusion gate only | Quantitative CFD validation/numbers |
+| Conclusion | Text-only integrated established answer | Optional visibly author-owned limitations/outlook outside frozen synthesis | Detailed replay, limitations table, evidence map, decisive-tests programme, new equations/citations/numbers/results |
+
+Near-collision: M04 standardized PPO and M06 corrected mean values are **different estimands**. SR action magnitude and V5 steady spin ratio are **related contextual diagnostics**, not amplitude agreement. No arithmetic or replication/validation/agreement/triangulation language.
+
+## Fixed scientific boundaries
+Cloaking/illusion is declared-reference matching, not universal invisibility. Positive SR is Kármán/cloak-only; Illusion SR is negative/historical. CCD is descriptive/noncausal; OID claim-free. Vortex is named comparable-scale transient capability; Erase quantification stays blocked. Steady is separate-chain kinematics-first viscous context, not a base-bleed/gap-jet mechanism label, boat-tail analogy, or mechanism. Observed gap-jet-family compatibility is allowed only after wall-motion direction is established and with a local hypothesis ceiling. Theory constrains without finite-Re closure. Experiment is qualitative/open-loop/provenance-gated. No mechanism, momentum restoration, stability, optimality, robustness, universality, priority, or quantitative experimental validation is licensed.
+
+## Open author choices
+| Choice | Recommendation / safe default | Status | Consequence |
+|---|---|---|---|
+| Final title | Retain provisional title unless reopened after section freezes | `discuss` | Any change needs terminology/claim audit |
+| Fig.1 set and optional Re400 marker | Author accepted the two-figure safe default: steady profiles plus a Kármán-dominant wake composition with one source-bound Vortex and one `ill_075L` preview; no Re400 marker | `author-choice-closed` | Exact panel source/provenance/caption/artwork remain gated; ladder unchanged |
+| Standardized PPO/SR/Zero panel | Omit; mismatched same-target acquisition blocks PPO inclusion | `blocked` | Use Target/SR/physical-zero only where exact local roles bind |
+| CCD manuscript label | Use `action-correlated correction-field modes` for outputs and `CCD analysis` generically | `revise` | Exact output terminology is mandatory |
+| Quantitative Vortex/Erase panels | Omit all scalars/efficacy wording until a versioned evaluator binds target, roles, ROI/mask, normalization, window and provenance | `blocked` | Named target/controlled/physical-zero role-field material may remain |
+| MFS field panel | Default omit until validation gates close | `blocked` | Use one-circle/sign-failure boundary |
+| Experiment image | Author-provenance-gated; omit if the image gate fails | `author-interface` | This dependency does not block the author interface or authorize an observation |
+| Experiment wording | Frank-reserved substantive wording; no agent fill | `author-interface` | If new provenance-passing material arrives, import requires a new targeted provenance audit and explicit coordinator reopening decision |
+| Final display/equation compression | Minimum argumentative set; S8 uses the Hodge/period display with adjacent one-circle incompatibility and one integrated figure | `partially closed—S8 only` | S8 source/provenance/caption/artwork remain gated; other compression remains drafting-level |
+
+Defaults do not make unresolved rows frozen.
+
+## Supersession rules
+1. Current authority/manifests/immutable artifacts/claims supersede summaries and memory.
+2. R2 `05_REVISED_STRUCTURE_FREEZE.md` supersedes conflicting R2 `01`–`04`; R2 `06` controls style.
+3. R3 `11_FINAL_CONSISTENCY_AUDIT.md` and reconciled exports control R3 conflicts; detail remains with owning module.
+4. A Round-4 change records old/new decision, reason, authority/author instruction, affected units, date, and re-audits; never erase provenance.
+5. Author choices cannot override evidence, merge chains/metrics, fabricate readiness, or raise claim ceilings.
+6. Reopen `frozen` only for new authority, discovered conflict, or explicit author redirection; dependent sections are re-audited.
+7. Missing optional panels do not reopen text when omission/supplement fallback exists; missing required evidence does.
+
+## Author feedback decisions — 2026-08-10
+- Overall manuscript structure and logic are accepted.
+- Section proportions are provisionally accepted and will be refined after all section discussions.
+- A distinct short Conclusion is required.
+- The experiment remains a short author-owned main-text subsection near the end.
+- The Introduction architecture is the accepted five-paragraph sequence: fascination/gap → pinball/force mediation → nonlinear feedback/PPO/analytical hooks → realization, independent evaluation, action-side and field-side analysis → narrowed question/contribution ladder/contracts bridge.
+- Reader-facing `Legacy`/`V5` naming is prohibited. Use `parabolic-inflow/no-slip-wall channel configuration` and `uniform-inflow/free-slip-wall channel configuration`; retain internal chain IDs only for provenance and ownership separation.
+- Reynolds naming is frozen in Section 2: use article-wide $Re_D=U_{ref}D/\nu$ with bulk/mean inlet speed for the parabolic configuration and uniform inlet speed for the uniform configuration; state $U_{max}\simeq1.5U_{ref}$ for the parabolic profile, convert a $2D$ code label by one half only when the owning source establishes the same velocity reference; parabolic inherited labels remain gated where historical `U0` may be centreline or bulk/mean; prohibit bare `Re100`; never infer equality from inherited labels.
+- Section 2 reader-facing title is `Problem formulation, control framework and evaluation`; `contract` remains internal governance language only.
+- Section 2 architecture is frozen. Main text foregrounds the canonical improved method and uses one compact neutral sentence that later deep analyses use data generated with an earlier implementation under the same parabolic-inflow/no-slip-wall physical configuration; implementation differences and the absence of a complete canonical rerun are documented in the supplement. No canonical-generation, equivalence, reproduction, or rerun-completion claim is licensed.
+- The sensor note is `read`: placement is $x/D=10$, $y/D=0,\pm2$, radius $0.25D$, with both velocity components, and its ROI/metric boundaries are current. Raibaudo supports the streamwise precedent, Cornejo Maceda and Li support downstream phase/state sensing, and Marra (`mapped/full record`, DOI `10.1017/jfm.2024.593`) supports only low-dimensional controlled-flow plausibility; none establishes optimality or observability.
+- The CelerisLab validation dossier is `read/full-report`: current-solver Kan99b K2 retains the authority's exact `partial pass` label and Sah04 S2 is `pass`; the Kan99b mean-lift sign difference is understood as the CFD coordinate/force-direction convention, not unresolved physics or suspected wrong rotation. No older-solver validation is licensed.
+- Fig. 1 remains undiscussed; no capability-panel decision is inferred; Section 2 geometry and DRL-flow schematics are required method displays.
+- The Introduction may be revisited after detailed sections, but its current Round-4 architecture is `frozen`; cultural-hook and citation wording remain drafting choices.
+- Round-4/5 work must prioritize author-facing argument and design decisions over dossier-scale governance; retain only a compact controlling-read/gap ledger.
+- Future section work uses a two-section cadence: the coordinator presents two detailed minimum context packets together, records author feedback, launches two tightly scoped continuation agents in parallel, then performs a coordinator second audit before freeze or R5 handoff.
+- Every minimum context packet lists exact controlling reads, accepted author feedback, available materials and readiness, requested output decision, exclusions, target length/section share, paragraph jobs, concrete display layout, ownership and exact questions. Continuation agents use targeted Nowledge recall first when prior decisions matter, verify against repository authorities, and do not rediscover the project.
+
+## Provenance separation
+Internal IDs `Legacy` and `V5` remain valid only in contracts, artifact paths, ledgers, and audit trails where they identify distinct evidence chains. Replacing manuscript-facing labels does not merge, erase, or transfer ownership between those chains.
+
+## Coordinator update record
+Format: `date | item | previous | new | authority/author instruction | affected units | required audit`.
+
+| Date | Item | Previous | New | Authority / author instruction | Affected units | Required audit |
+|---|---|---|---|---|---|---|
+| 2026-08-10 | Overall architecture | Under discussion | Structure/logic accepted; proportions provisional pending all section discussions | Author feedback | All Round-4 section freezes | Recheck final assembled balance |
+| 2026-08-10 | Conclusion and experiment | Conclusion/experiment placement carried but not restated as author decision | Distinct short Conclusion required; short author-owned experiment subsection retained near manuscript end | Author feedback | Experiment, Conclusion, assembled order | Preserve qualitative/provenance gates |
+| 2026-08-10 | Introduction | Freeze candidate with unresolved reader adjectives | Five-paragraph architecture frozen; exact boundary-condition configuration labels adopted | Author feedback plus repository configuration/kernel evidence | Introduction, Contracts, captions/ledgers | Revisit wording after detailed sections without reopening architecture |
+| 2026-08-10 | Reynolds naming | Configuration-local convention with parabolic reference unresolved | Section 2 freezes article-wide bulk/mean-based $Re_D=U_{ref}D/\nu$; uniform speed for uniform inlet; $U_{max}\simeq1.5U_{ref}$ for parabolic inlet; source-bound $2D$ labels convert by one half only when the velocity reference matches; historical parabolic labels remain gated | Author feedback and repository evidence | Section 2, Introduction, labels/captions | Confirm every displayed inherited label against source authority |
+| 2026-08-10 | Fig. 1 | Open choice | Undiscussed; no new decision | Author feedback | Capability/display planning | Keep open |
+| 2026-08-10 | Section 2 formulation and provenance | Freeze candidate; sensor/validation authorities misclassified; provenance too defensive | Architecture frozen; canonical main text plus compact neutral earlier-implementation sentence; differences and no-complete-rerun fact supplementary; sensor note `read`; CelerisLab dossier `read/full-report`; no equivalence claim | Author correction plus current training, sensor and validation authorities | Section 2, captions, supplement provenance, inventories | Keep run values and production dependency-gated; no Fig. 1 decision |
+| 2026-08-10 | Agent-control priority | Round-1-style comprehensive coverage remained prominent in R4 controls | Author-facing argument/design and decision value take priority; retain compact controlling-read/gap ledger | Author process correction | All R4/R5 prompts, packets and audits | Reject dossier-scale rediscovery or governance-heavy outputs |
+| 2026-08-10 | Section cadence | Sections researched/frozen individually | Present two detailed packets together; after author feedback launch two scoped continuation agents in parallel; coordinator second-audits both | Author process decision | Remaining R4 sections and R5 handoffs | Audit pair interfaces, duplication, ownership and accepted feedback |
+| 2026-08-10 | Continuation context | Agents assembled context from broad controls/dossiers | Coordinator minimum packet freezes exact reads, feedback, materials/readiness, decision, exclusions, length/share, paragraph jobs, display layout, ownership and questions; targeted Nowledge recall precedes repository verification when relevant | Author process decision | Coordinator, all continuation agents | Memory never overrides artifacts; no whole-project rediscovery |
+| 2026-08-10 | Kan99b mean-lift sign | `partial pass` with sign/convention described as unresolved | Preserve exact `partial pass`; sign difference is understood as CFD coordinate/force-direction convention, not unresolved physics or suspected wrong rotation | Author correction and qualification interpretation | Section 2, controls, future prose/captions | Remove unresolved-sign wording while preserving dossier label |
+
+## Integration closeout record — 2026-08-11
+
+All five pair audits and the final reconciliation were completed on 2026-08-11. Round 4 closed with S3/S8 author choices open and S9 as an author-interface. For R5 production, the author has now accepted the recorded S3 and S8 safe recommendations and permits explicit author-field placeholder TeX for S9. These decisions close author-choice status only; they do not retroactively approve artifacts, provenance or experimental content.
+
+| Section | Current integrated choice/default | Status after integration | Recommended disposition | Coordinator second-audit focus |
+|---|---|---|---|---|
+| 3 | Capability/progression preview only; one representative Vortex and one Illusion source-bound preview; steady exact endpoint detail deleted; Fig. 1 composition remains author choice | `discuss` | `discuss` | Final reconciliation complete; Fig. 1 composition remains an open author choice |
+| 4 | Reference parabolic-source case uses exact configuration wording and no reader-facing Reynolds value; body jobs are inference/comparator-led; layout items are coordinator defaults | `frozen` | R5 eligible; artifacts gated | Final reconciliation passed; no figure generated |
+| 5 | Historical effective applied $\alpha=\omega/U_0$ retains its source convention; general $s=\alpha R(U_0/U_{ref})$, reducing to $s=\alpha R$ only with source-established $U_0=U_{ref}$; neutral source-case labels and shared-lag six-channel native-DTW deletion contract fixed | `frozen` | R5 eligible; artifacts gated | Final reconciliation passed; parabolic velocity-reference conversion remains gated; no figure generated |
+| 6 | Exact configuration wording; target → physical zero → constant reference → DRL is the sole accounting-order exception; complementarity internal; POD supplemental; cross-cloak small multiples omitted; source-case Reynolds conversions removed from the Figure Plan | `frozen` | R5 eligible; artifacts gated | Final reconciliation passed; title/composition remain drafting-level; no figure generated |
+| 7 | Bounded role-identity/morphology text architecture; all provisional Vortex/Erase scalars and efficacy implications omitted; quantitative displays blocked and role-field visuals dependency-gated | `frozen` text contract | R5 text eligible | Final reconciliation passed; no figure generated |
+| 8 | Tangential wall velocity alone is directly prescribed; shear/vorticity, shear-layer, inter-cylinder and recirculation features are endpoint observations/hypotheses | `discuss` | `discuss` | Frank must decide the main equation/composition; no acceptance inferred |
+| 9 | Author-owned substantive wording remains reserved; image and physical command map are separately provenance-gated | `author-interface` | `author-interface` | Any future D3/C1 import requires a new targeted provenance audit and explicit coordinator reopening decision |
+| 10 | Text-only established synthesis; no Erase success, experiment result, display, equation, citation, number, limitations table, evidence map or decisive-tests programme | `frozen` | R5 eligible | Pair-E second audit passed; later author limitations/outlook remains outside the frozen synthesis |
+
+**Final reconciliation status:** Pair audits and final reconciliation completed on 2026-08-11. S1 is frozen/R5 eligible; S2 frozen/eligible with local gates; S3 remains `discuss` for the genuine Fig. 1 author choice; S4–S6 are frozen/eligible with artifacts gated; S7 text is frozen/eligible with quantitative Vortex/Erase displays blocked and role visuals gated; S8 remains `discuss` for the genuine author equation/composition choice; S9 remains author-interface, not agent drafting; S10 is frozen/eligible.
+
+**Safe defaults applied:** omit provisional Section-7 scalars; omit PPO from the mismatched Section-5 same-target comparison; keep Section-6 complementarity diagnostics internal; keep POD supplementary; omit cross-cloak small multiples; omit a conclusion display. These defaults support the frozen text contracts but do not resolve S3/S8 author choices or alter the S9 interface.
+
+**Supersessions:** current Sections 1 and 2 are the active integrated owners of former Style-01/02 content; deleted Style files are historical/superseded provenance only. The current Section-10 author correction supersedes any control that required a limitations table, evidence map, decisive-tests/future-work programme, or conclusion-owned display/equation/citation/number.
+
+
+## Final coordinator outcome — 2026-08-11
+
+Pair audits and final reconciliation are complete. Round 4 is closed for frozen units but is not globally all-sections frozen because Sections 3 and 8 remain `discuss` and Section 9 remains an author-interface. The figure-spec phase may proceed under recorded gates. Literature/style enrichment may cover all units, but it must not override the open S3 Fig. 1 choice, the open S8 equation/composition choice, or the S9 author interface.
+
+## Round-5 author-choice record — 2026-08-11
+
+| Date | Item | Previous | Accepted R5 control | Closed status | Gates explicitly retained |
+|---|---|---|---|---|---|
+| 2026-08-11 | S3 figure composition | Genuine author choice; two-figure recommendation only | Two-figure safe default: steady profile figure plus Kármán-dominant wake figure with one source-bound Vortex and one 0.75-radius Illusion preview | Author-choice gate closed | Exact panel selection, source/provenance, crop/scale, caption, artwork, rendering and publication readiness |
+| 2026-08-11 | S8 equation/composition | Genuine author choice; recommendation only | Multi-cylinder Hodge/period outer-solution display with prescribed periods, adjacent one-circle incompatibility, and one integrated CFD/outer-reference/sign-failure figure | Author-choice gate closed | Equation/source check, figure assets, provenance, caption, redraw/rendering, finite-image status and all finite-Re mechanism ceilings |
+| 2026-08-11 | S9 R5 output form | Author-interface; agent prose blocked | Explicit author-field placeholder TeX is permitted | Placeholder-format choice closed | All apparatus facts, execution details, trial IDs, command map, raw media, processing, permissions, morphology observations and result import remain author/provenance-gated |
+
+These accepted choices supersede only the corresponding `discuss`/output-form status in the earlier closeout rows. They do not close any artifact, evaluator, source, provenance, caption or publication-readiness gate.

docs/JFM_WYQ/ROUND4_TOP_DOWN/00_MASTER_AGENT_PROMPT.md

@@ -0,0 +1,49 @@
+# Round 4/5 Section Agent Prompt
+
+Use only after the coordinator has built the minimum context packet and the author has responded to the paired section discussion.
+
+## Coordinator minimum context packet
+- **Unit / round / output:** [SECTION], [R4 FREEZE or R5 DRAFT], `[EXACT FILE]`
+- **Decision requested:** [ONE BOUNDED OUTPUT DECISION]
+- **Exact controlling reads:** [R4 CONTROLS, OWNING DOSSIER, CURRENT AUTHORITIES/MANIFESTS]
+- **Accepted author feedback:** [DATED DECISIONS; distinguish accepted, provisional, open]
+- **Available materials:** [EXACT TEXT, DATA, SCRIPT, FIGURE/PANEL OR TABLE INVENTORY + `text-ready`/`artifact-ready`/`dependency-gated`/`blocked`]
+- **Target size:** [WORDS/PAGES] and [PERCENT SECTION/MANUSCRIPT SHARE]
+- **Paragraph jobs:** [P1 JOB → P2 JOB → ... → CLOSING BRIDGE]
+- **Concrete display layout:** [FIGURE/TABLE/EQUATION COUNT, PANEL GEOMETRY, PLACEMENT, EACH ARGUMENTATIVE JOB AND FALLBACK]
+- **Ownership:** imports [EXACT]; exports [EXACT]; prohibited duplication [EXACT]
+- **Exact questions:** [FINITE LIST ANSWERABLE FROM PACKET]
+- **Exclusions:** [OUT-OF-SCOPE SOURCES, CLAIMS, RUNS, EDITS, DISPLAYS]
+
+## Prompt
+You are the tightly scoped **[R4 SECTION-FREEZE / R5 SECTION-DRAFT]** continuation agent for **[UNIT]** in *Interpreting learned hydrodynamic cloaking and illusion in a fluidic pinball*.
+
+Start with targeted Nowledge `memory_search` for prior decisions relevant to the exact questions; use thread recall only when the exact earlier discussion is needed. Then verify operative decisions against the packet's repository authorities. Memory is decision/history navigation and never overrides current authorities, manifests, code/configuration or immutable artifacts.
+
+Read the minimum context packet above and only its exact controlling reads. Do not rediscover the whole project or recreate Round-1 coverage. Use repository tools for internal evidence. Use Undermind only for a stated external-literature gap and only after `get_orientation` then `list_workspaces`; literature cannot establish internal results. For any authorized read-only derivation/script, first discover the designated conda environment from repository authority or environment listings and record it—never guess. Do not run CFD, training, artifact generation or expensive analysis unless separately authorized.
+
+### Deliverable priorities
+1. Answer the section question with an evidence-bounded recommendation that gives the author a concrete decision.
+2. Honor target length/share and make every paragraph perform its assigned argumentative job.
+3. Specify the concrete display layout: panel arrangement, owner, source arrays/artifacts, placement, argumentative job, caption ceiling, readiness and fallback. Inventory is not artifact readiness.
+4. Respect result/terminology ownership, imports/exports, bridges and duplication deletions.
+5. State green/amber/red claim wording and preserve chain, configuration, Reynolds convention, role, comparator, metric/estimand, ROI/window and artifact boundaries.
+6. Include only a compact coverage ledger for the controlling reads and material gaps: `read`, `mapped`, `not-covered`, `external-inaccessible`, with verification/artifact-audit state where relevant.
+7. Separate evidence-governed conclusions from author choices. For each choice give recommendation, consequence and safe default.
+
+### Fixed boundaries
+- Positive SR is Kármán/cloak-only; Illusion SR is historical/negative; CCD is descriptive/noncausal and uses `action-correlated correction-field modes`; OID is claim-free; steady is contextual; experiment is qualitative/provenance-gated.
+- Internal Legacy/V5 chains remain separate and provenance-only in manuscript prose. Case-specific metrics and near-colliding estimands are never pooled or called replication, validation, agreement or triangulation without exact authority.
+- Kan99b K2 remains `partial pass` under the qualification authority. The mean-lift sign difference is the CFD coordinate/force-direction convention, not unresolved physics or suspected wrong rotation; preserve the overall label and state its reason accurately.
+- A first pass cannot freeze itself. Return `researching`, `discuss`, `revise` or `blocked`; only the coordinator's second audit may set `frozen` or authorize R5 prose.
+
+### Output order
+1. Recommended decision and bounded answer.
+2. Target size/share and paragraph-job map.
+3. Concrete display/equation layout and material readiness.
+4. Ownership, imports/exports, bridge and deletions.
+5. Claim ceiling, choices and exact answers.
+6. Compact coverage/gap ledger.
+7. Disposition.
+
+Edit exactly `[EXACT FILE]`; do not edit R1–R3 dossiers, evidence packages, immutable artifacts, archives, controls or other sections.

docs/JFM_WYQ/ROUND4_TOP_DOWN/00_QUALITY_GATE.md

@@ -0,0 +1,50 @@
+# Round 4/5 Quality Gate
+
+Round 4 freezes author-facing arguments and section designs; Round 5 drafts from those freezes. Neither round creates evidence. Decision value and manuscript design outrank procedural volume.
+
+## Entry gate — coordinator packet
+- [ ] Exact unit, round, output decision and file are named.
+- [ ] Exact controlling reads and accepted dated author feedback are listed.
+- [ ] Available material/artifact inventory distinguishes text-, data-, script-, panel- and dependency-readiness.
+- [ ] Target words/pages and section/manuscript share are stated.
+- [ ] Paragraph jobs and closing bridge form an argument, not an inventory.
+- [ ] Concrete display layout gives count, panel geometry, placement, job, owner, source, readiness and fallback.
+- [ ] Imports, exports, exclusions and exact bounded questions are explicit.
+
+A subagent failing this gate is not launched to rediscover the project.
+
+## Process gate — continuation and routing
+- [ ] When prior decisions matter, agent begins with targeted Nowledge memory/thread recall, then verifies against repository authorities; memory never overrides artifacts.
+- [ ] Repository tools handle internal evidence. Undermind is literature-only after `get_orientation` and `list_workspaces`.
+- [ ] Read-only scripts/derivations use a designated conda environment discovered from authority/listing; no environment is guessed.
+- [ ] No CFD, training, artifact generation or expensive analysis ran without authorization.
+- [ ] Coverage ledger is compact and limited to controlling reads/gaps (`read`, `mapped`, `not-covered`, `external-inaccessible`).
+
+## Section design/freeze gate
+- [ ] Recommendation answers the exact author-facing question and explains consequence/trade-off.
+- [ ] Length/share, paragraph progression, display cadence and evidence density are credible.
+- [ ] Claims retain owner, chain/configuration, Reynolds convention, role, comparator, metric/estimand, ROI/window and authority.
+- [ ] Imports stay brief; exports and deletion owners prevent duplication.
+- [ ] Displays/equations state what they establish and do not establish; text readiness is separate from artifact readiness.
+- [ ] Author choices remain `discuss`/`blocked` or receive an accepted decision; safe defaults are not falsely frozen.
+- [ ] Kan99b K2 stays exactly `partial pass`; the mean-lift sign difference is attributed to CFD coordinate/force-direction convention, never unresolved physics or suspected wrong rotation.
+- [ ] Fixed SR/CCD/OID/steady/experiment and cross-chain/estimand cautions pass.
+
+## Two-section cadence and second audit
+1. Coordinator presents two detailed packets together.
+2. Author responds to both; coordinator records accepted, provisional and open feedback.
+3. Coordinator launches two tightly scoped continuation agents in parallel.
+4. Coordinator re-reads both outputs and controlling authorities, checks interfaces/duplication/claims/displays/choices, and records a second audit.
+5. Only then may R4 become `frozen` or R5 prose become drafting-ready.
+
+## Statuses
+- `researching`: a named controlling read is incomplete.
+- `discuss`: a bounded author choice is decision-ready.
+- `revise`: targeted argument/design/interface correction is needed.
+- `frozen`: coordinator second audit locked the R4 text contract.
+- `blocked`: required authority, material, role, comparator, decision or conflict resolution is absent.
+- `author-interface`: substantive content is explicitly reserved for author supply and provenance binding; it is neither frozen nor an agent-filled gap.
+
+An integration correction pass returns recommendations only; it does not perform the coordinator second audit or set a new freeze.
+
+Final Round-4 audit checks the paired sections in the whole-manuscript order, ownership, bridges, claim ladder, display/equation inventory, accepted author decisions and Round-5 handoff. Optional omitted panels do not block text when a safe fallback preserves the argument.

docs/JFM_WYQ/ROUND4_TOP_DOWN/00_R4_CLOSEOUT_AUDIT.md

@@ -0,0 +1,126 @@
+# Round-4 horizontal closeout audit
+
+**State:** final reconciliation complete; pair audits and final reconciliation closed on 2026-08-11.
+
+**Role:** coordinator-auditable correction, final Round-5 admission decision and handoff control; not evidence, manuscript prose, artifact approval, or a section freeze.
+
+**Date:** 2026-08-11.
+
+## 1. Scope and consolidated correction categories
+
+This integration pass normalized the current Round-4 controls and Sections 1–10 under six categories:
+
+1. **Configuration and Reynolds convention:** article-wide $Re_D=U_{ref}D/\nu$ remains fixed; $U_{ref}$ is bulk/mean for parabolic inflow and uniform speed for uniform inflow. The reference parabolic-source Kármán case carries no reader-facing Reynolds value until its inherited `U0` centreline-versus-bulk/mean label is source-bound.
+2. **Action, order and signs:** computational order is $(F,U,L)=(\text{front},\text{rear}_{y+},\text{rear}_{y-})$; positive $\omega$ is counter-clockwise under $(U_w,V_w)=(-\omega r_y,\omega r_x)$. Historical SR $\alpha=\omega/U_0$ has inverse-length units unless radius is already nondimensionalized; the general article surface-speed command is $s=\omega R/U_{ref}=\alpha R(U_0/U_{ref})$; $s=\alpha R$ is allowed only when source authority establishes $U_0=U_{ref}$. Historical parabolic $U_0$ is not retroactively bound.
+3. **Ownership and preview/detail splits:** Section 3 previews progression; Section 7 owns detailed Vortex/Illusion/Erase material; Section 8 owns exact steady endpoint/theory context; Figure Plans own crop, scale and panel geometry.
+4. **Metric and interpretation boundaries:** generic metric definitions remain in Section 2; local Section-6 and steady estimands remain non-equivalent. DTW orientation/window/aggregation and deletion-delta sign are explicit. CCD outputs use exactly `action-correlated correction-field modes`.
+5. **Evidence ceilings and safe omissions:** provisional Vortex/Erase scalars and efficacy wording are omitted; PPO is omitted from the mismatched Section-5 comparison; Section-6 complementarity stays internal, POD supplemental, cross-cloak small multiples omitted; experiment remains author-interface.
+6. **Conclusion and supersession controls:** Section 10 is text-only established synthesis. Required limitations table, evidence map, decisive-tests/future-work programme and conclusion-owned display/equation/citation/number are superseded. Former Style-01/02 files are historical; current Sections 1/2 are active owners.
+
+## 2. Section 1–10 integration matrix
+
+| Section | Formal current status | Text readiness | Artifact readiness | Open author choices | Safe fallback | Recommended disposition |
+|---|---|---|---|---|---|---|
+| 1 Introduction | `frozen` | Architecture ready | No owned empirical display | Drafting-level cultural hook/optional example | Generic hook; omit optional example | R5 eligible; Pair A passed |
+| 2 Problem/method/evaluation | `frozen` | Architecture ready; local source gates explicit | Required schematics/tables dependency-gated | No architecture choice | Minimal schematic/tables; provenance detail supplementary | R5 eligible with local source/display gates; Pair A passed |
+| 3 Capability/progression | `discuss` | Preview/detail split complete | Fig. 1 dependency-gated | Genuine Fig. 1 author choice | One representative Vortex and one Illusion source-bound preview | Keep open; do not freeze Fig. 1 |
+| 4 Kármán performance | `frozen` | Text contract R5 eligible | Field redraw/caption binding dependency-gated | No scientific choice; production defaults recorded | Compact table; optional spectrum/disturbance material omitted first | Frozen text; artifacts gated; no figure generated |
+| 5 Kármán SR law | `frozen` | Text contract R5 eligible after final narrow fix | Target/SR/zero available; PPO mismatch gate retained | Optional deployment envelope is drafting-level | Omit PPO/envelope; retain law, Target/SR/zero and deletion | Frozen text; artifacts gated; no figure generated |
+| 6 Field/CCD | `frozen` | Text contract R5 eligible | Main composites dependency-gated; POD supplemental | Title/final composition choices are drafting-level | Scalar ledger + residual definition | Frozen text; artifacts gated; no figure generated |
+| 7 Cross-case | `frozen` text contract | Text R5 eligible | Quantitative Vortex/Erase blocked; role visuals dependency-gated | Final visual selection remains production-level | Physical-variable role fields only; positive Illusion PPO | Frozen text; displays remain gated; no figure generated |
+| 8 Steady/theory | `discuss` | Direct-prescription/endpoint-observation boundary repaired | Integrated composition dependency-gated | Frank: main equation/composition | Endpoint/sign table + concise outer boundary | Remain `discuss`; no acceptance invented |
+| 9 Experiment | `author-interface` | Scaffold only; substantive wording author-reserved | Image/command map separately provenance-gated | Frank supplies apparatus, trials, media and wording | Omit observation/media | Pair E passed as author-interface; not a frozen agent-written section |
+| 10 Discussion/conclusion | `frozen` | Text-only synthesis architecture passed | No display | Optional author-owned limitations/outlook outside synthesis | Omit experiment and Erase | R5 eligible; Pair E passed |
+
+## 3. Coordinator second-audit result — 2026-08-11
+
+- **Pair A — Sections 1–2: PASS.** Scientific content was not altered. S1 is frozen/R5 eligible; S2 is frozen/R5 eligible with its local source/display gates.
+- **Pair B — Sections 3–4: PASS AFTER REPAIR.** Section 3 remains `discuss` because Fig. 1 composition is a genuine author choice; it uses one representative Vortex preview and leaves event evolution to Section 7. Section 4 is frozen/R5 eligible with artifact gates retained.
+- **Pair C — Sections 5–6: PASS AFTER REPAIR.** Section 5 now binds historical effective applied $\alpha=\omega/U_0$, the conditional physical conversion, neutral source-case labels and the exact shared-lag six-channel deletion contract. Sections 5 and 6 are frozen/R5 eligible with artifact gates retained.
+- **Pair D — Sections 7–8: PASS WITH OPEN S8 AUTHOR CHOICE.** Section 7 text is frozen/R5 eligible; quantitative Vortex/Erase displays remain blocked and role visuals gated. Section 8 remains `discuss`, preserving the genuine author equation/composition choice and the direct-prescription versus endpoint-observation boundary.
+- **Pair E — Sections 9–10: PASS.** Section 9 remains an author-interface with image/provenance dependencies separated from author-reserved wording. Section 10 is frozen/R5 eligible. Any future experiment import requires a new targeted provenance audit and explicit coordinator reopening decision.
+
+Pair audits and the final reconciliation are complete. Frozen text units are Round-5 eligible under their recorded local and artifact gates; Sections 3 and 8 remain `discuss`, and Section 9 remains an author-interface.
+
+## 4. Post-R4 handoff fields
+
+- **Final coordinator record:** `2026-08-11 | Pairs A–E audited; final reconciliation complete; frozen units R5 eligible under recorded gates | coordinator pair audits and final reconciliation`.
+- **Accepted author choices:** none newly inferred in this repair.
+- **Open author choices:** `S3 FIG.1 COMPOSITION`; `S8 MAIN EQUATION/COMPOSITION (FRANK)`.
+- **Drafting-level production choices:** `S4 density/spectrum defaults`; `S5 optional envelope`; `S6 title/phase-bin composition`—these do not reopen the frozen text contracts.
+- **Remaining blocked authorities:** `PARABOLIC SOURCE VELOCITY-REFERENCE CONVERSION`; `S7 VERSIONED EVALUATOR FOR QUANTITATIVE VORTEX/ERASE`; `S9 AUTHOR PROVENANCE/COMMAND MAP/WORDING`.
+- **Final section statuses:** `S1 frozen/R5 eligible; S2 frozen/R5 eligible with local gates; S3 discuss—genuine Fig. 1 author choice; S4 frozen/R5 eligible—artifacts gated; S5 frozen/R5 eligible—artifacts gated; S6 frozen/R5 eligible—artifacts gated; S7 frozen text/R5 eligible—Vortex/Erase quantitative displays blocked and role visuals gated; S8 discuss—genuine author equation/composition; S9 author-interface—not agent drafting; S10 frozen/R5 eligible`.
+- **R5 authorization:** all frozen text units are eligible under their recorded gates; S3/S8 remain discuss and S9 remains author-interface.
+- **Figure-specification phase:** integration complete. Specifications, provenance fields, normalized readiness states and safe fallbacks are organized for Sections 1–10 under recorded artifact/evaluator/author gates. No panel, table, figure or artifact was generated, promoted, recomputed or newly independently audited.
+- **Paragraph-level literature/style enrichment:** may cover all units but must not override the open S3/S8 author choices or the S9 author interface; no new literature was used in this repair.
+
+## 5. Compact coverage ledger
+
+| State | Material | Boundary |
+|---|---|---|
+| `read` | Four current Cursor rules; all six R4 controls; current Sections 1–10 | Architecture, terminology, ownership and current claims; read is not independent scientific verification or artifact audit |
+| `read` | Targeted current SR authority/code for historical action definition and DTW deletion contract | Exact implementation binding only; no CFD, fitting, recomputation or artifact generation |
+| `memory-recalled` | Three narrow Nowledge searches for Reynolds/ownership, SR action history, and Sections 7/9/10 decisions | Navigation/history only; repository authorities controlled edits |
+| `mapped` | Owning R3/package authorities already named by current packets | Not reread except the concrete SR conflict; mapping is not verification |
+| `not-covered` | New literature, R1–R3 edits, artifacts, CFD/training/rendering/computation, experiment media | Explicitly excluded; targeted current Figure-Plan text was edited |
+| `external-inaccessible` | None required for the integration decisions | No accessibility claim beyond current packet records |
+
+## 6. Integration disposition
+
+All five pair audits and the final reconciliation are complete as of 2026-08-11. Round 4 is closed for frozen units, but it is not globally all-sections frozen: Sections 3 and 8 remain `discuss` for genuine author choices, and Section 9 remains an author-interface rather than agent drafting. The figure-specification integration is complete under recorded gates; literature/style enrichment may cover all units but cannot override those open choices or interface.
+
+## 7. Round-5 admission audit — 2026-08-11
+
+**Audit basis.** All ten current `SECTION_*.md` packets, including their dated paragraph-level literature/style enrichments, were read against the current global ledger, style crosswalk, Round-2 structure/style controls and terminology/writing rules. `PASS` admits the packet's authorized prose jobs; `CONDITIONAL` admits only the listed text scope while retaining local author, source or artifact exclusions; `BLOCKED` admits no agent drafting for that unit. Prose readiness is not artifact readiness.
+
+| Section | Admission | Packet/jobs, authority and claim ceiling | Terminology, displays/equations and provenance | Interfaces, author inputs and gates |
+|---|---|---|---|---|
+| 1 Introduction | **PASS** | Frozen five-job architecture; internal ladder/definitions remain authority-owned and literature remains framing/style only; no detailed result or raised claim | Configuration names and `Kármán` fixed; no owned empirical display, equation or table; optional hook/example safely omittable | Imports ladder and exports definitions bridge; cultural hook/citation wording are drafting choices, not blockers |
+| 2 Problem/method/evaluation | **CONDITIONAL** | Frozen jobs and claim ceiling; current method/configuration authorities own facts; exact run rows remain source-bound | Required method schematics/tables are production-gated; reusable equations admitted only with source-bound notation; inherited parabolic Reynolds conversion remains gated | Prose may draft definitions and bounded qualification; do not print unbound run values, imply canonical rerun/equivalence or claim display readiness |
+| 3 Capability/progression | **CONDITIONAL** | Paragraph jobs and preview/detail ownership are ready, but packet remains `discuss`; no pooled efficacy or transferred depth | Fig. 1 composition is a genuine author choice and all panels remain source/production-gated; no equation | Draft only figure-independent viewing contract, non-equivalence, bounded previews and Kármán-selection logic. Exclude final panel count/layout, source-bound representative choice and composition-dependent caption prose pending author choice |
+| 4 Kármán performance | **PASS** | Frozen text contract; authority-bound tested range, five-realization result and deep-case field comparison; local repeatability/metric ceilings explicit | Two figures are not generated/approved; field redraw, hash/scale/crop and exact shared spectrum window remain gated; no new equation ownership | Paragraph prose may begin using packet facts; use omission/supplement fallbacks and never present artifact readiness |
+| 5 Kármán SR law | **PASS** | Frozen five jobs; executable surrogate, imposed symmetry and parent-relative deletion ceilings explicit | Required owner equations are definition-ready; main figure production gated; PPO/SR common-target comparison blocked pending reconciliation; historical $U_0$ conversion gated | Draft formula-parent/standardized SR scopes separately; omit PPO from common panel and optional envelope unless exact gates close |
+| 6 Field/CCD | **PASS** | Frozen six jobs; exact four-role authority, scalar ceiling, non-paired residual and descriptive CCD/POD boundary explicit | Required share/residual equations ready; composites need redraw/recomposition; POD supplemental; cross-cloak audit omitted; optional CCD operator only if needed | Draft bounded mean/scalar/residual/modal prose; do not promote internal complementarity diagnostics or infer causality/triangulation |
+| 7 Cross-case extensions | **PASS** | Frozen text contract with role-field-only Vortex/Erase and positive PPO Illusion ceiling; local clocks/comparators and strict non-transfer explicit | Quantitative Vortex/Erase displays **BLOCKED** without versioned evaluators; role visuals dependency-gated; no scalar efficacy; source-bound fallbacks recorded | Draft changed-contract and morphology/role prose only; omit efficacy, common score, arbitrary-flow, transfer and Kármán interpretation |
+| 8 Steady/theory | **CONDITIONAL** | Paragraph jobs, endpoint authority, kinematics-first hypothesis and failed closure ceiling are ready, but packet remains `discuss` | Frank must select the main equation and integrated composition; source assets are redraw-ready, not approved; outer field retains bounded-reference status | Draft only figure/equation-independent context, endpoint and failure logic. Exclude final displayed equation, composition-dependent narration/caption and any acceptance of the recommendation |
+| 9 Experiment | **BLOCKED** for agent prose | Author-interface only; scaffold has jobs but Frank owns apparatus, execution, trial observations and wording; no experimental result currently exports | Image/command-map provenance unresolved; no equation/table; single frames cannot support temporal, field, force or validation claims | Agent may preserve placeholders/interface only. Manuscript paragraph drafting waits for author supply, targeted provenance audit and explicit reopening; omission is the current fallback |
+| 10 Discussion/conclusion | **PASS** | Frozen three-plus-two text-only synthesis; current Sections 1–8 own all imported meanings; no new evidence, values, citations or outlook programme | No display, equation, citation or number; exact terminology and non-equivalence controls ready | Draft from established retained results only; no Section-9 observation unless its separate reopening gate later passes; author-only limitations/outlook remains outside freeze |
+
+### Admission result and conditional exclusions
+
+**Round-5 paragraph drafting may begin now for Sections 1, 2, 4, 5, 6, 7 and 10, and only for the figure/equation-independent portions of Sections 3 and 8 described above. Section 9 is blocked for agent manuscript drafting.** The manuscript is therefore **prose-admitted with conditional exclusions**, not globally unconditional and not artifact-ready.
+
+The remaining conditions are exact:
+
+1. **S2 source/display gates:** bind per-result settings, selected policy/normalizer/target, qualification rows, displayed inherited Reynolds labels and required schematic/table sources before printing dependent detail.
+2. **S3 author gate:** Frank selects Fig. 1 composition; until then, no final panel selection, layout, source-bound representative choice or composition-dependent caption prose.
+3. **S4–S6 production gates:** redraw/recompose and provenance-bind figures; keep spectrum/window, PPO/SR target reconciliation, historical velocity-reference conversion, action-trace choice, optional operator/POD and cross-cloak material under their recorded gates/fallbacks.
+4. **S7 evaluator gates:** no quantitative Vortex or Erase efficacy, scalar or publication panel until a versioned evaluator binds target, roles, ROI/mask, normalization, window and provenance; role-field visuals remain production-gated.
+5. **S8 author gate:** Frank selects the main equation and composition; no composition-dependent prose or implied acceptance. MFS/outer material remains a prescribed-circulation reference, not finite-$Re$ mechanism validation.
+6. **S9 author/provenance gate:** no agent-authored apparatus/result prose and no promoted media until author facts, trial/media provenance, permissions, command map and morphology wording pass a targeted audit and explicit reopening. Otherwise omit.
+7. **Article-wide artifact gate:** no figure, table or panel is publication-ready merely because a textual specification or source exists. Every retained display still requires source/provenance, crop/scale/role binding, caption-ceiling and final production review.
+
+### Audit dimensions and final determination
+
+Packet status, paragraph jobs, evidence authority, claim ceiling, terminology, display provenance/readiness/fallback, equation ownership, literature style support, interfaces/bridges, author-only inputs and artifact gates are explicit for every section. No unresolved production gate was marked resolved. The operative distinction is:
+
+- **Prose-ready:** authorized paragraph jobs can be drafted within the owning packet's scientific and wording ceiling.
+- **Artifact-ready:** a retained display has passed source/provenance, role/metric, rendering and caption approval. No Round-4 planning statement grants this state; current main displays remain gated or absent as recorded above.
+
+**Final admission:** **CONDITIONAL PASS for Round 5.** Begin bounded paragraph drafting under the updated writing protocol and the section-by-section exclusions above; do not treat this as admission of S9 agent prose, approval of S3/S8 author choices, or publication readiness of any artifact.
+
+## 8. Final style-closeout coverage record
+
+- `read`: all current Round-4 Sections 1–10 in full, including every dated 2026-08-11 enrichment block; current global ledger, style crosswalk, closeout audit and README; Round-2 structure freeze/style guide; current terminology and writing rules. These reads govern architecture/style only and are not independent scientific or artifact audits.
+- `memory-recalled`: three narrow Nowledge searches for Round-4 admission/status, recurring paragraph style and artifact gates. Memory was navigation/history only; repository controls determined the audit.
+- `durable-memory write`: not used; this closeout records its decision in repository controls, and no separate durable memory was needed.
+- `thread recall`: not used; exact conversation wording was unnecessary after current repository authorities were read.
+- `literature`: no new search or PDF read was performed in this final synthesis. The section packets' dated literature work was **read as recorded planning evidence** and mapped to recurring global style contracts; it was not independently re-read, verified, or used as internal scientific authority.
+- `Working Memory`: not used.
+- `not-covered`: raw artifacts, scientific recomputation, CFD, training, SR/CCD/theory reruns, experiment media audit, figure generation and publication rendering.
+
+## 9. Round-5 author-choice update — 2026-08-11
+
+The author accepted the S3 two-figure safe default and the S8 recommended equation/composition, and authorized explicit author-field placeholder TeX for S9. R5 admission is updated only as follows: S3 may draft against the two-figure interface; S8 may draft the Hodge/period display with adjacent one-circle incompatibility and integrated-figure responsibilities; S9 may create placeholders containing author fields only.
+
+This update closes no scientific, source, evaluator, provenance, caption, artwork, rendering or publication-readiness gate. S7 quantitative Vortex/Erase remains blocked; S3/S8 displays remain production-gated; S9 remains blocked for agent-authored facts, observations and experimental results. The exact R5 controls and output paths are in `../ROUND5_BOTTOM_UP/README.md`.

docs/JFM_WYQ/ROUND4_TOP_DOWN/00_STYLE_CROSSWALK.md

@@ -0,0 +1,159 @@
+# Round 4 style synthesis and Round-5 crosswalk
+
+**Role:** historical Gate-A audit of the original ten style memos; its style findings are progressively integrated into active `SECTION_*.md` files, and parallel style memos are deleted after integration. This is style/architecture control, not manuscript prose, scientific evidence, artifact approval, or a section freeze.
+
+**Overall result: PASS for global style synthesis.** The historical Gate-A audit remains below; the 2026-08-11 paragraph-level enrichment in all ten active section packets is now consolidated into the operational Round-5 protocol without changing science or readiness.
+
+## 1. Scope and complete coverage ledger
+
+`read` is not independently verified or artifact-audited. `mapped` is not read. External literature is **style evidence only**; internal claims remain controlled by repository authorities, manifests, immutable artifacts, the Round-2 freeze, and audited Round-3 dossiers.
+
+| State | Inputs | Boundary |
+|---|---|---|
+| `read` | All Round-4 controls: `README.md`, `00_MASTER_AGENT_PROMPT.md`, `00_GLOBAL_DECISION_LEDGER.md`, `00_DISPLAY_EQUATION_INVENTORY.md`, `00_QUALITY_GATE.md` | Complete workflow/decision control set; not scientific verification |
+| `read` | All ten style memos, each in full | Style observations, recommendations, identities/read levels, interfaces and anti-patterns; no memo is internal scientific authority |
+| `read` | R2 `05_REVISED_STRUCTURE_FREEZE.md`, `06_SECTION_LEVEL_JFM_STYLE_GUIDE.md` | Author-ratified architecture/claims and controlling style synthesis |
+| `read` | R3 `00_MODULE_INTERFACE_LEDGER.md`, `00_QUALITY_GATE.md`, `11_FINAL_CONSISTENCY_AUDIT.md` | Readiness, ownership, near-collisions, blockers and assembly controls; no recomputation |
+| `mapped` | R3 Modules `01`–`10`; authorities/manifests/artifacts they name | Scientific detail remains with its owner; no mapped item is promoted to read or verified |
+| `mapped` | Literature records declared by the memos/R2 guide | Identity/read-level audit only; no new literature search or PDF reading |
+| `not-covered` | New literature; raw arrays/hashes; CFD, training, SR, CCD or OID recomputation; experiment analysis; figures; manuscript drafting | Outside Gate A; no citation, internal number, panel, or prose inferred |
+| `external-inaccessible` | Hidden/symlink payloads, model bundles, dense fields/theory arrays, raw experiment media | Authority-led artifact deferral; retained dependent panels/claims remain gated |
+
+Observed comparator order, cadence and caveat practice come from the memos' declared `full-PDF`, `full-report`, `mapped`, or `abstract/snippet` literature states. Project scope, terminology, results and readiness come only from internal controls and owning authorities.
+
+## 2. Gate-A result by memo
+
+| Integrated owner | Result | Exact reason | Targeted correction |
+|---|---|---|---|
+| `SECTION_01_INTRODUCTION_FREEZE.md` | **PASS** | Current owner of the integrated Introduction architecture and former Style-01 findings | Deleted `STYLE_01_INTRODUCTION.md` is historical/superseded provenance only |
+| `SECTION_02_PROBLEM_METHOD_EVALUATION.md` | **PASS** | Current owner of the integrated method/evaluation architecture and former Style-02 findings | Deleted `STYLE_02_CONTRACTS_METRICS.md` is historical/superseded provenance only |
+| `SECTION_03_CAPABILITY_PROGRESSION.md` | **PASS** | The former Style-03 findings, four exact identities/DOIs/read levels, comparator-led visual hierarchy, caption/body division, local limits, anti-patterns and fallbacks are integrated into the active Section file; the parallel style memo was deleted | None; publication-size event-sequence choice remains `discuss` |
+| `SECTION_04_KARMAN_RESULTS.md` | **PASS** | The former Style-04 findings, four exact identities/DOIs/read levels, comparator-led escalation, field/signal separation, caption practice and anti-patterns are integrated into the active Section file; the parallel style memo was deleted. Current author feedback replaces the old retention emphasis with five-realization high-performance discovery | None; final two-figure density choices remain `discuss` |
+| `SECTION_05_KARMAN_SR_LAW.md` | **PASS** | Gate-A style research passed: required R2/R3 controls were read, and panel status, ownership and Kármán/cloak-only scope were reconciled without changing the claim ceiling; these findings were integrated into the Section file and the parallel style memo was deleted | None; targeted controlling-read correction verified |
+| `SECTION_06_KARMAN_FIELD_CCD.md` | **PASS** | Gate-A style research passed: three exact full-PDF identities plus one exact mapped/inaccessible record; read levels are not inflated; mean-before-modes, non-pairing, fixed terminology, POD boundary and M04/M06 separation remain explicit; these findings were integrated into the Section file and the parallel style memo was deleted | None; keep Noack et al. (2016) mapped unless full text is read |
+| `SECTION_07_CROSS_CASE_EXTENSIONS.md` | **PASS** | Gate-A style research passed and was integrated with the author-corrected changed-contract architecture, actual Vortex/Illusion/Erase data audit, evidence boundaries and two-figure production interface; the parallel style memo was deleted | None; retain event-relative semantics, local comparators and strict non-transfer boundaries |
+| `SECTION_08_STEADY_THEORY.md` | **PASS** | Gate-A style research and author feedback were integrated into the Section file: finite-Re observation, improved kinematics-first physical hypothesis, prescribed-circulation outer solution, streamline display and failed sign/closure route remain distinct; the parallel style memo was deleted | None; retain outer-reference status and finite-Re mechanism ceiling |
+| `SECTION_09_EXPERIMENT.md` | **PASS** | Gate-A style research passed and was integrated: experiment media retain explicit author-supply gates and prose/omission fallbacks, with no observation promoted before provenance closes; the parallel style memo was deleted | None; retain author-supply gating, prose/omission fallback and no-observation-promotion ceiling |
+| `SECTION_10_DISCUSSION_CONCLUSION.md` | **PASS** | Gate-A style research passed and was integrated: exact Deng et al. and Cornejo Maceda literature identities were corrected, and the evidence-order/separate-Discussion-and-Conclusion rationale was retained; the current author instruction supersedes the former limitations/future-work design, and the parallel style memo was deleted | None; retain exact identities and the evidence-order/split rationale without restoring limitations or future-work design |
+
+No memo uses literature to establish an internal result, and no memo invents a new internal number. The corrected defects were limited to exact identity and control-coverage bookkeeping; they do not alter scientific claims.
+
+## 3. Shared versus section-specific JFM practice
+
+### Shared observed practice
+
+1. A target, baseline, reference role or simpler controller is legible before interpretation; extensions renew comparators locally.
+2. Evidence escalates from contract to principal result to richer spatial/modal interpretation; strong vocabulary follows dedicated tests.
+3. One paragraph and one display normally carry one main argumentative job.
+4. Captions decode roles, conditions and panels; body text states inference, limit and consequence.
+5. Dangerous limitations sit beside affected results and may be compressed again in Discussion.
+6. Equations define immediately/repeatedly used objects; figures carry empirical burden; tables compress contracts or comparable rows.
+7. Bridges pose the unresolved scientific question rather than announce a section number.
+8. Negative routes remain visible to motivate refinement or define the envelope.
+
+### Section-specific practice
+
+- **Introduction:** compact funnel; clustered citations; no equations or detailed project results; subsections only for genuine taxonomy.
+- **Contracts:** dependency order from plant/baseline through objective, metrics and qualification; tables expose contract shifts.
+- **Capability:** one dominant comparator-bearing case anchors breadth; panel order/area express hierarchy; no common score.
+- **Performance:** exact comparator and principal result precede interpretation; field and signal estimands stay separate.
+- **SR:** variables, objective and deployment precede law; execution arbitrates; deletion is not causal ablation.
+- **Field/CCD:** mean/scalar estimand precedes residual/modes; POD is a same-object boundary, not a superiority foil.
+- **Cross-case:** change one contract dimension where possible; renew comparator; label visual-only and negative evidence locally.
+- **Steady/theory:** late context is narrower than a second climax; attractive theory and failed sign/closure stay adjacent.
+- **Experiment:** apparatus/acquisition/provenance precede qualitative observation; omit media when provenance fails.
+- **Discussion/Conclusion:** complete interpretation before the short close and synthesize only established answers in evidence order. Limitations/outlook are optional, visibly author-owned, and outside the current synthesis architecture.
+
+## 4. Global Round-4 rules to freeze
+
+1. **Comparator-first:** bind target, controlled role, physical zero/baseline, chain and estimand before inference.
+2. **Evidence escalation:** named capability → Legacy Kármán independent field evaluation → bounded Kármán/cloak-only interpretation; cross-cases do not inherit deeper levels.
+3. **Local caveats:** put mask/window, acquisition, causal, transfer, theory and provenance limits with the affected sentence/caption.
+4. **One job per paragraph:** delete package chronology, inventory prose and panel narration.
+5. **Ownership:** Section 2 owns reusable definitions; Capability Fig. 1/progression; Performance detailed reference-case L2/DTW; SR deployment/law/deletion; Field/CCD corrected means/residual/modes/POD; Cross-case changed-contract extensions; Steady/theory separate-chain endpoint/sign/failure; Experiment provenance/observation; Conclusion text-only synthesis and no new evidence.
+6. **Caption/body division:** caption states chain, case/role, comparator, metric/coordinate, source and ceiling; body explains inferential consequence.
+7. **Display discipline:** inventory never means existence. Every retained display needs owner, job, source, readiness, caption ceiling and fallback; text and artifact readiness stay separate.
+8. **Equation discipline:** freeze reusable notation once; do not duplicate owner equations; Conclusion has none.
+9. **Bridges:** gap → contracts → capability → independent Kármán evaluation → law complexity → field accounting → non-transfer → steady context → qualitative deployability → established synthesis.
+10. **Claim ceilings:** positive SR Kármán/cloak only; Illusion SR historical/negative; CCD descriptive/noncausal using exactly **action-correlated correction-field modes**; OID claim-free; steady contextual; experiment qualitative, open-loop and provenance-gated.
+11. **Notation/chain:** preserve `Kármán`, `Re_D`, `Re_code`, spatial L2 versus Level/Stage, DTW orientation, action order/sign, ROI/mask/window and percentages. Legacy/V5 never pool, continue, validate or imply equivalence.
+12. **Metric boundary:** periodic mean/phase, event-relative, mean-only, steady profile/`E_inf_vector`, and native/normalized DTW are non-equivalent. M04/M06 means are **different estimands**; SR action magnitude/V5 spin are **related contextual diagnostics** only.
+
+## 5. Author decisions for macro discussion (maximum five)
+
+1. **Figure density and main/supplement split.** Recommend the minimum argumentative main set; move derivations, qualification matrices and optional diagnostics to supplement; omit before crowding. Decide final Fig. 1 panels and optional Re400/limit markers.
+2. **Gated displays.** Recommend omitting standardized PPO/SR/zero, quantitative Vortex/Erase, and MFS field panels unless exact gates close. Treat CCD operator/POD choices as coordinator production recommendations; the evidence-map proposal is superseded and is not an open decision.
+3. **Experiment inclusion.** Retain a short author-owned subsection only if provenance supports a named qualitative observation; otherwise use a limit/interface sentence and omit media. Quantitative validation remains prohibited.
+4. **Discussion/Conclusion structure.** Retain the accepted integrated Discussion followed by a distinct short Conclusion; no one-section fallback remains active.
+5. **Introduction freeze and title.** Freeze Introduction last; retain the provisional title unless reopened. Any title change triggers terminology/claim audit.
+
+## 6. Section-specific crosswalk
+
+| Section | Recommended sequence | Primary display contract | Ownership | Main fallback |
+|---|---|---|---|---|
+| Introduction | concept/boundary → difficulty → capability/gap → evidence-chain need → scope/bridge | No owned empirical display; optional prospective Fig. 1 reference | Full ladder, concept, gap, scope | No display; five-job funnel |
+| Contracts | comparison need → plant/chain → action/sign → observation/objective → metrics/roles/windows → qualification | Required contract table + minimal schematic; reusable metric equations | Reusable definitions and qualification | Compact table/schematic; supplement detail |
+| Capability | viewing contract → steady seed → dominant Kármán → bounded extensions → limit/deep-case selection | Kármán-dominant distinct-contract Fig. 1 | Progression, evidence levels, deep-case choice | Source-bound panels only or prose |
+| Performance | tested conditions and repeated learning → exact deep-case field comparison → mean/phase L2 → signal/action organization → SR question | Two figures: integrated condition/learning map and L2-first reference-case display | Condition range, five learning histories, detailed deep-case L2/DTW and PPO action/sensor evidence | Move disturbance-size group and spectrum/DTW detail to supplement |
+| SR | surrogate boundary → deployment → law → execution → deletion → field question | Required deployment map/law/deletion; direct role panel gated | Executable law, imposed structure, arbitration, tested deletion | Equations + deletion table |
+| Field/CCD | four roles → means/ledger → limit → residual → CCD → POD boundary | Mean before modes; residual equation required; CCD operator optional | Corrected means, residual, descriptive modes, POD boundary | Scalar ledger + residual equation |
+| Cross-case | framing → event-relative → non-zero target → blocked/failure route → envelope | Distinct-contract display; quantitative Vortex/Erase gated | Comparator renewal, evidence levels, negative routes | Case-contract table + sourced visuals |
+| Steady/theory | return → V5 endpoint → bounded description → outer reference/failure → export | Sign contract and failure required; endpoint gated; MFS blocked | Separate-chain endpoint/sign/context/theory failure | Sign schematic + endpoint table + trace |
+| Experiment | purpose → apparatus/provenance → qualitative comparison → limit/test | Author-supply-gated media; no quantitative panel/equation | Frank-owned provenance and observation | Prose interface or omission |
+| Discussion/Conclusion | breadth-to-depth → bounded interpretation → short close | No conclusion display/equation/citation/number | Integrated established answer only | Text-only synthesis; optional author-owned limitations/outlook outside frozen synthesis |
+
+## 7. Macro-discussion input
+
+### Author-ratified decisions carried forward
+
+- **Assembled order:** Introduction → contracts → capability → Kármán performance → SR → field/CCD → cross-case → steady/theory → experiment → Discussion/Conclusion.
+- **Freeze order:** contracts → capability → Kármán performance → SR → field/CCD → cross-case → steady/theory → experiment → Discussion/Conclusion → Introduction last.
+- **Claim ladder:** named cross-task capability; independent field matching for the reference periodic Kármán cloak case in the parabolic-inflow/no-slip-wall channel configuration; bounded Kármán/cloak-only control-skeleton interpretation. The internal source label remains provenance only.
+- **Historical planning shares:** Introduction 10%, contracts 13%, capability 11%, Kármán performance 12%, SR 17%, field/CCD 17%, cross-case 10%, steady/theory 6%, experiment/limitations/conclusion 4%. This is a superseded R2 allocation record, not an active requirement for a limitations unit and not a new measurement.
+- **Physical chronology:** steady discovery → Kármán → Vortex → Illusion/Erase → qualitative experiment, with steady as an early seed and later contextual return.
+
+### Coordinator recommendation; not author-ratified
+
+1. Preserve both orders and keep SR, field/CCD, capability and cross-case as distinct ownership units.
+2. Protect the Kármán block as the analytical centre; cross-case and steady define breadth/boundaries, not competing interpretation.
+3. Apply the claim ladder sentence by sentence; interpretation remains Kármán/cloak-only and noncausal.
+4. Retain the minimum display inventory: method table/schematic; Kármán-dominant Fig. 1; Kármán performance; SR law/deletion; mean then residual/CCD; bounded cross-case role-field display/table; steady sign/endpoint/failure; provenance-gated experiment if available. Conclusion has no display. Optional operators, insets, maps and blocked panels are drop-first.
+5. Retain the minimum equation inventory: reusable definitions in Contracts; SR deployment/law; scalar identity only if retained plus residual in Field/CCD; event equation only with a retained quantitative result; steady definition only if not imported plus one outer-reference boundary. Full derivations go to supplement.
+6. Keep result ownership exclusive: previews name conclusions without values; owners carry exact comparisons; Discussion compresses without replay.
+
+These recommendations remain `discuss` until recorded as author decisions; they do not become frozen by appearing here.
+
+## 8. Handoff and quality gate
+
+1. Present the five decision bundles to the author for macro discussion; record decisions or safe omissions in the global ledger. Do not falsely freeze unresolved choices.
+2. Freeze architecture only when ownership, bridges, ladder, notation, space and display/equation inventory agree. Optional panels may remain gated; retained required panels with missing source or roles remain unavailable.
+3. For section freezes, reread the owning R3 dossier and current authorities, not only this crosswalk; current Sections 1/2 replace deleted Style 01/02 as active reads; run local near-collision and caption-ceiling checks and require coordinator second audit.
+4. Before Round 5, audit `Kármán`, Legacy/V5, `Re_D`/`Re_code`, spatial L2 versus Level/Stage, DTW orientation, action order/sign, ROI/mask/window, percentages, equation ownership, duplicate detail and whitespace.
+
+**Gate-A handoff:** the ten-memo style set is **PASS** and ready for macro discussion through the five author-decision bundles. Optional display dependencies and later section-freeze checks retain their recorded gates. No manuscript prose, literature, internal number, artifact or panel is created by this audit.
+
+## 9. Final paragraph-enrichment synthesis and authority — 2026-08-11
+
+### Authority and supersession
+
+The dated 2026-08-11 paragraph-level enrichment blocks in `SECTION_01`–`SECTION_10` are the current section-local owners of paper identities, read states, paragraph-specific observations, local wording ladders, caveats, captions and anti-patterns. They supersede the corresponding globalized recommendations in the Round-2 style guide and this file wherever a section packet is more specific. The Round-2 structure freeze still controls author-ratified architecture where not superseded by recorded Round-4 author decisions; `00_GLOBAL_DECISION_LEDGER.md` controls Round-4 architecture, ownership, claim and supersession conflicts. The global `.cursor/rules/jfm-writing-style.mdc` contains only recurring Round-5 operations. It does not replace a section packet, establish evidence or carry paper lists.
+
+### Cross-section contracts extracted from all five dual-section enrichments
+
+| Recurring contract | Section-enrichment mapping | Global Round-5 operation |
+|---|---|---|
+| Paragraph move | S1–S2 define before inference; S3–S4 comparator and principal result before diagnostics; S5–S6 object/equation before execution or modes; S7–S8 renew changed contracts and keep failed routes adjacent; S9–S10 comparator/provenance before observation and synthesis in earned evidence order | One job; claim/question → comparator/definition → evidence → interpretation → local caveat → consequence/bridge |
+| Evidence and citations | Every pair separates repository-owned internal statements from literature-owned rhetorical/context statements | Cite the external proposition only; never use literature to certify an internal result, value, artifact or readiness; retain source-state labels |
+| Caveat timing and wording | Acquisition, metric, non-pairing, intervention, evaluator, theory-closure and provenance limits repeatedly sit beside the attractive result | Green direct; amber only with named test and adjacent condition; red prohibited without dedicated authority/test; no terminal disclaimer repair |
+| Caption/body division | Across field, signal, modal, theory and experiment displays, captions decode conditions and provenance while body states inference | Caption: configuration/roles/variable/coordinate/window/metric/source/local ceiling; body: comparison/inference/limit/consequence |
+| Equation/display discipline | S1/S9/S10 prohibit equations; S2 owns reusable definitions; S5/S6/S8 place only immediately used owner equations; all packets separate inventory from readiness | Define once with symbols/units/sign/domain; keep beside dependent evidence; no decorative duplication; require owner/source/readiness/fallback for displays |
+| Transitions | Every pair converts the established result into the next scientific question and preserves the manuscript interface | Scientific bridge, never section/package announcement; previews do not steal owner detail and synthesis does not replay it |
+| Anti-patterns | All pairs reject catalogues, chronology, panel narration, generic AI prose, metric pooling, delayed caveats and unsupported strong verbs | Run the anti-AI and paragraph self-audit in the global rule before accepting each paragraph |
+
+### Section-local material deliberately not globalized
+
+Paper identities and DOIs, paragraph-by-paragraph support sets, exact values, local role order exceptions, individual captions, figure compositions, equation choices, production defaults and case-specific green/amber/red phrases remain in their owning section packets. This prevents a paper catalogue or local scientific constraint from becoming a false manuscript-wide convention.
+
+### Round-5 handoff
+
+The global writing rule operationalizes these recurring contracts. Round 5 must read the owning R4 packet first, then use this crosswalk for manuscript-wide consistency. A global PASS here does not change `discuss`, `author-interface`, dependency-gated or blocked statuses and does not approve any artifact.

docs/JFM_WYQ/ROUND4_TOP_DOWN/README.md

@@ -0,0 +1,27 @@
+# Round 4 — Top-Down Decisions and Section Freezes
+
+## Purpose
+Round 4 converts the R2 architecture and audited R3 dossiers into author-facing section arguments, concrete display designs and frozen R5 drafting packets. It is not manuscript prose, new evidence, artifact generation, CFD or training. Round 4 should be lighter than the dossiers: preserve traceability in a compact ledger while spending most effort on argument, ownership, displays and decisions.
+
+## Authority and tool routing
+Current repository authorities, manifests, code/configuration and immutable artifacts override summaries and memory. A continuation agent begins with targeted Nowledge memory/thread recall when prior decisions matter, then verifies against exact repository authorities. Nowledge is for decisions/history; repository tools are for internal evidence; Undermind is for external literature only after `get_orientation` and `list_workspaces`. Read-only derivations/scripts use only a designated conda environment discovered from current authority or environment listings. Never guess an environment or run CFD/training without authorization.
+
+## Coordinator workflow
+1. Maintain the global decision ledger, whole-manuscript ownership and author feedback.
+2. Build a minimum context packet for each section: exact controlling reads, accepted feedback, available material/readiness inventory, requested output decision, exclusions, target length/share, paragraph jobs, concrete display layout, ownership and exact questions.
+3. Present two detailed section packets together to the author.
+4. After feedback, launch two tightly scoped continuation agents in parallel. Agents use the packet and targeted checks; they do not rediscover the project.
+5. Perform a coordinator second audit across both outputs before freezing either section.
+6. Repeat by section pairs; freeze the Introduction last, then run the whole-manuscript Round-4 audit.
+
+## Controlling inputs
+- `ROUND2_TOP_DOWN/05_REVISED_STRUCTURE_FREEZE.md` and `06_SECTION_LEVEL_JFM_STYLE_GUIDE.md`.
+- Relevant `ROUND3_BOTTOM_UP` owning dossiers, interface ledger, quality gate and final consistency audit.
+- Current repository authorities/manifests named by the packet.
+- This directory's master prompt, quality gate, global ledger and display/equation inventory.
+
+## Round-5 handoff
+Each frozen packet carries its bounded answer, target length/share, paragraph jobs, exact displays/equations, ownership/imports/exports, claim ceiling, accepted author decisions, exclusions, compact coverage/gap ledger and separate text/artifact readiness. Round 5 drafts only from that packet and its controlling authorities; it does not silently reopen decisions or fill evidence gaps.
+
+## Current horizontal integration status
+The consolidated correction, final style synthesis and Round-5 admission matrix is `00_R4_CLOSEOUT_AUDIT.md`; recurring enrichment contracts and supersession are mapped in `00_STYLE_CROSSWALK.md`. Round 5 has a **conditional pass**. The author has accepted the S3 two-figure safe default and S8 recommended equation/integrated composition, so those author-choice gates are closed for bounded R5 drafting; their source, provenance, caption, artwork and production gates remain open. Section 9 remains an author-owned provenance-gated interface and permits explicit author-field placeholder TeX only, not agent-authored facts or results. Prose admission does not approve any panel or artifact. Current Sections 1/2 own integrated former Style-01/02 content; deleted Style files remain historical/superseded provenance only.

docs/JFM_WYQ/ROUND4_TOP_DOWN/SECTION_01_INTRODUCTION_FREEZE.md

@@ -0,0 +1,280 @@
+# Round 4 frozen section architecture — Introduction
+
+## 1. Status and scope
+
+**Status:** `frozen` for Round-4 architecture; **not final manuscript prose**, not a citation-priority audit, and not scientific or artifact verification. Cultural-hook and sentence-level citation wording remain drafting choices and do not reopen the architecture.
+
+**Scope:** architecture for an unsplit five-paragraph Introduction. This file controls paragraph jobs, literature roles, terminology choices, evidence boundaries, ownership, and the bridge into the contracts section. It does not draft publishable paragraphs, define reusable equations, narrate results, or make any panel publication-ready. Reader-facing configuration labels are frozen below from inlet and lateral-wall boundary conditions; additional solver, outlet, geometry, and action differences remain owned by configuration-local contracts.
+
+**Bounded recommendation:** begin with the human idea of concealment and deception, move immediately to why direct linear/transformation cloaking does not transfer trivially to separated finite-Reynolds-number Navier–Stokes flow, introduce the fluidic pinball as the controlled channel-flow object, formulate force-mediated cloaking as a nonstationary feedback problem, and preview a layered realization–evaluation–interpretation design. Close with the narrowed question and three contribution levels, then hand the reader to configuration-local contracts. Do not organize the Introduction around internal chain labels, package history, or defensive nonclaims.
+
+## 2. Section question and answer ceiling
+
+**Section question.** Can rotational control of a fluidic pinball in channel flow use force feedback and sparse downstream observations to approach a declared hydrodynamic reference, and what evidence is required to distinguish learned execution, independent field evaluation, and bounded physical interpretation?
+
+**Answer ceiling.** The Introduction may motivate and announce three nested levels: (1) named learned-control capability under distinct task contracts; (2) independent spatial field matching for the periodic Kármán cloak under the parabolic-inflow/no-slip-wall channel configuration only; and (3) a bounded Kármán/cloak-only control-skeleton interpretation assembled from executable action-law and descriptive field-organization evidence. It may not imply universal invisibility, linear-cloak/Navier–Stokes equivalence, cross-chain continuation, a causal mechanism, transfer, robustness, optimality, priority, positive Illusion symbolic regression, or quantitative experimental validation.
+
+**Green:** define hydrodynamic cloaking as approach to a declared background and illusion as retargeting toward a declared non-zero signature; motivate force-plus-sparse-observation control; state the three levels without values.
+
+**Amber:** say that the later layered evidence is designed to test field matching beyond the policy observer/reward space and to support a bounded Kármán/cloak-only organization of action and field evidence.
+
+**Prohibited red:** all-observer invisibility; direct equivalence between transformation/linear cloaks and finite-Re Navier–Stokes control; force reduction as proof of field matching; causality, necessity, sufficiency, uniqueness, stability, robustness, optimality, universality, transfer, priority, independent triangulation, positive Illusion SR, causal CCD, an OID result, or quantitative experiment validation.
+
+## 3. Author feedback incorporated
+
+- **Reader-facing configuration terminology:** use the exact pair `parabolic-inflow/no-slip-wall channel configuration` and `uniform-inflow/free-slip-wall channel configuration`. Do not use historical/current, earlier/revised, or internal chain IDs as reader-facing adjectives. The configurations remain separate because solver, outlet, geometry, actuation, Reynolds convention, roles, and windows also differ; those details remain in contracts. Internal chain IDs (`Legacy`, `V5`) are provenance only.
+- **Reynolds language:** never use bare `Re100`. Prefer the reader-facing basis $Re_D=U_{ref}D/\nu$, with $U_{ref}$ defined separately for each configuration. For the uniform-inflow/free-slip-wall configuration, the configured uniform inlet speed supplies $U_{ref}$. For the parabolic-inflow/no-slip-wall configuration, the article-wide convention fixes $U_{ref}$ as the bulk/mean inlet speed and $U_{max}\simeq1.5U_{ref}$. Only conversion of inherited source labels remains gated where historical `U0` may denote centreline rather than bulk/mean speed; that source-specific gate does not reopen the article convention. Equal code labels must never be presented as equal physical Reynolds numbers.
+- **P1:** open from concealment as longstanding fiction/human fascination; use Harry Potter only with a precise source. Move to hydrodynamic cloaking and then illusion as hiding-to-deception. Radar stealth/deception is optional. End on the finite-Re gap after naming nonlinear advection, separated wakes, dissipation, boundary conditions, active forcing, and observer/target dependence. Delete literature disclaimers.
+- **P2:** start from the scale gap between shaping an extended Navier–Stokes field and controlling a compact rotating object. Select the pinball for rich nonlinear dynamics, prior rotation knowledge, and broad force/action authority. Distinguish force-target matching from scalar drag minimization. Include Noack lineage and Feng et al. only under exact roles. End with the natural force-mediated control question.
+- **P3:** bridge to nonstationary feedback; motivate DRL/PPO; treat sparse sensing as application-driven. Force feedback complements sparse observations without novelty or observability claims. Preview objective execution, target-field evaluation, and extractable/checkable action/field organization as reader hooks.
+- **P4:** use learned realization → independent evaluation → action-side and field-side analysis. Explain each layer through exact closed-loop ML, sparse/reduced-order, and modal/field precedents. Remove package names and defensive nonclaim lists.
+- **P5:** close confidently with the narrowed question, three contribution levels, and positive rationale for Kármán depth. Compress asymmetric scope into one sentence and bridge to contracts. Do not lead with limitations, Illusion SR, Erase, experiment caveats, or application disclaimers.
+- **Experiment:** reserve a short, author-owned qualitative open-loop subsection near manuscript end. Mention it in the Introduction only if structurally useful.
+- **Scientific cautions retained:** positive SR is Kármán/cloak-only; Illusion SR is historical/negative; CCD is descriptive/noncausal and uses exactly `action-correlated correction-field modes`; OID is claim-free; steady is contextual; experiment is qualitative, open-loop, author-owned, and provenance-gated.
+
+## 4. Recommended five-paragraph architecture
+
+### Paragraph 1 — fascination, hydrodynamic cloak, and finite-Re boundary
+
+**Rhetorical job.** Turn a familiar desire to hide or deceive into a precise fluid-mechanical gap without beginning with methods or internal evidence.
+
+**Logical moves.**
+1. Open with concealment/invisibility as a longstanding human and science-fiction fascination.
+2. Use Harry Potter only as a brief cultural hook if the author approves an exact citable source; otherwise keep the opening generic.
+3. Move from optical/field cloaking to hydrodynamic cloaking: an object seeks to leave a declared background flow/signature minimally disturbed for a specified observer or comparison.
+4. Introduce illusion as the natural extension from hiding to deception: replace the expected signature with a declared non-zero target. A compact radar analogy is optional.
+5. Explain why transformation-based or active linear-field constructions do not transfer directly: nonlinear advection couples scales and states; body boundaries generate separated shear layers and wakes; dissipation changes cancellation logic; boundary conditions and active forcing belong to the plant; success depends on declared observer and target.
+6. End with the gap: how can an actively forced body in nonlinear channel flow be controlled toward a declared background or target, and how should success be established?
+
+**Literature roles.** Pendry et al. (2006), Leonhardt (2006), Vasquez et al. (2009), Ma et al. (2013), and Urzhumov & Smith (2012) support cloak/active-cloak genealogy and restricted assumptions. Chen, Shen & Xu (2022) can supply an exact recent porous-medium hydrodynamic example, not finite-Re channel-flow evidence. Harry Potter/radar sources remain optional verification tasks.
+
+**Internal project imports.** Declared-background and declared-non-zero-target definitions; nonlinear channel-flow framing; no result values or chain IDs.
+
+**Excluded content.** History catalogue; packages; linear/NS equivalence; universal invisibility; priority; applications list; literature disclaimer.
+
+**Bridge/output.** The nonlinear-flow gap needs a concrete controllable body whose force and wake signatures are coupled.
+
+### Paragraph 2 — fluidic pinball and force-mediated cloaking
+
+**Rhetorical job.** Introduce the fluidic pinball as the object implementing the concept and distinguish the objective from ordinary drag reduction.
+
+**Logical moves.**
+1. Start from the scale mismatch: the desired outcome is an extended Navier–Stokes field, whereas control acts through three compact rotating bodies.
+2. Define a three-cylinder fluidic pinball in channel flow; defer solver, boundaries, Reynolds, geometry, sign, and action details.
+3. Explain that approaching a declared background may require reducing or reshaping the body's force signature, making force a meaningful signal and intermediary.
+4. Separate scalar drag minimization from potentially time-dependent, multi-component force-target matching; neither alone proves downstream field matching.
+5. Establish the Noack/Morzyński lineage: rich nonlinear regimes, three independent rotations, and reduced-order/open-loop/closed-loop/ML control history.
+6. Add exact rotational-force precedent: Feng et al. (2023) controls force extrema and references of a rotating pinball using DRL. This establishes nearby precedent and accessible authority, not this field-cloaking result.
+7. Explain selection by rich nonlinear dynamics, prior rotation knowledge, and broad accessible force/action range in a compact MIMO plant.
+8. End naturally: can rotations use force-mediated feedback to make the pinball approach the declared hydrodynamic background while preserving a checkable downstream objective?
+
+**Literature roles.** Noack & Morzyński toolkit lineage; Deng et al. (2020) for dynamics; Cornejo Maceda et al. (2021) and Raibaudo et al. (2020) for rotational/ML control; Feng, Wang, Xiang, Jin & Fan (2023) for force extrema/tracking. Group by role.
+
+**Internal project imports.** Rotating-body object; force plus sparse downstream observations; declared references; channel-flow wording.
+
+**Excluded content.** Bare `Re100`; solver/wall details; action signs; drag reduction as cloaking; force match as field match; unverified claims about Dixia Fan.
+
+**Bridge/output.** The force-mediated objective becomes a feedback problem when the reference is nonstationary and the full field is not observed online.
+
+### Paragraph 3 — nonlinear feedback, PPO, and three analytical hooks
+
+**Rhetorical job.** Motivate the learned controller and sparse execution interface, then convert achieved control into three scientific questions.
+
+**Logical moves.**
+1. Acknowledge pinball/control precedent while identifying the harder feature: nonstationary reference following with coupled force and sparse downstream information is less intuitive than steady actuation or scalar optimization.
+2. Motivate DRL for high-dimensional nonlinear closed-loop decisions without a tractable controlled-flow model; identify PPO locally as the selected optimizer, not a contribution by itself. If PPO is explained, say only that clipped policy updates regularize step size relative to the previous policy, supported by the original PPO paper or a suitable review.
+3. Present sparse downstream sensing as application-driven limited remote observation.
+4. Explain that body-integrated force feedback complements the sparse downstream signature. Avoid novelty, optimality, sufficiency, and observability claims.
+5. Pose three hooks: does execution achieve its declared objective; does the flow match the target field beyond observer/reward space; what recurring action and field organizations can be extracted and checked?
+6. End by requiring layered analysis rather than treating reward convergence as the scientific answer.
+
+**Literature roles.** Rabault et al. (2019), Paris et al. (2021), Li & Zhang (2022), Xia et al. (2024), and Wang et al. (2024) establish DRL/partial-observation precedents. Schulman et al. (2017), “Proximal Policy Optimization Algorithms,” arXiv:1707.06347, is the original PPO citation; Garnier et al. (2021) is a verified general review. The Introduction need not tutorialize PPO.
+
+**Internal project imports.** Force plus sparse velocities; independent field evaluation; bounded Kármán-specific interpretation.
+
+**Excluded content.** Any claim that PPO mathematically guarantees gradient descent, monotonic improvement, convergence, or learning success; generic AI claims; sparse sensing as deficiency; priority; optimal sensing/full observability; reward/DTW as field equivalence; training or architecture details; negative-limit list. PPO is motivation, not a result or contribution.
+
+**Bridge/output.** The three hooks determine the realization, independent-evaluation, and interpretation layers.
+
+### Paragraph 4 — layered approach and why each layer exists
+
+**Rhetorical job.** Preview the scientific method as a dependency chain, with a reason for each layer.
+
+**Logical moves.**
+1. **Realize learned control:** establish what the closed-loop policy can do under named task contracts; capability precedes interpretation.
+2. **Evaluate independently:** compare registered downstream fields with target and physical-reference roles outside the online reward; separate execution-space success from field-space evidence.
+3. **Analyze from the action side:** check a reduced policy through executable closed-loop surrogate/deletion tests; ask whether compact organization survives deployment, not merely offline fitting.
+4. **Analyze from the field side:** separate persistent and dynamic contributions, organize the residual through descriptive action correlation, and compare conventional modal structure; connect organization without causality.
+5. Connect prior closed-loop ML control to layer 1, sparse/reduced-order reconstruction to fuller-state interpretation, and modal analysis to layer 4. Literature motivates roles, not internal findings.
+6. End attractively: one compact controlled object, sparse executable feedback, independent field testing, and paired action/field interpretation under explicit contracts.
+
+**Literature roles.** Gautier et al. (2015) and Cornejo Maceda et al. (2021) for deployed ML laws; Loiseau et al. (2018) for sparse sensing to fuller-state representation; Deng et al. (2020) and verified modal/reduced-order records for field interpretation.
+
+**Internal project imports.** Three-level ladder; spatial L2 owner downstream; executable SR; corrected four-role mean comparison; separately centred non-paired residual; `action-correlated correction-field modes`; POD boundary.
+
+**Excluded content.** Packages; formulas/values/windows/masks/panels; independent triangulation; mechanism; causal/unique/superior CCD; OID result; defensive inventory.
+
+**Bridge/output.** The approach prepares the narrowed question and explains why one case receives depth.
+
+### Paragraph 5 — narrowed question, contribution levels, and contracts bridge
+
+**Rhetorical job.** Close calmly with the paper question, three levels, Kármán-depth rationale, and the next definitions.
+
+**Logical moves.**
+1. Ask whether force-plus-sparse-observation rotational control under the separately declared channel-flow contracts can realize hydrodynamic cloak/illusion objectives, and how far the periodic Kármán cloak under the parabolic-inflow/no-slip-wall configuration can be evaluated and interpreted.
+2. Name: learned-control capability across distinct tasks; independent spatial field matching for the named periodic Kármán cloak; bounded Kármán/cloak-only action-and-field control-skeleton interpretation.
+3. Explain Kármán depth positively: it has the richest aligned intersection of policy execution, independent fields, executable action-law tests, corrected mean/dynamic comparisons, and descriptive field/modal organization.
+4. Use one asymmetric-scope sentence: cross-case, steady, and experiment material provides breadth, context, or qualitative feasibility but does not inherit Kármán interpretation.
+5. Mention the author-owned qualitative experiment only if navigation requires it.
+6. End by requiring the next section to define channel plants, boundary/solver configurations, actions, observations, configuration-local Reynolds conventions, roles, comparators, metrics, regions, and windows.
+
+**Literature roles.** Normally light/uncited; contributions come from internal authorities. A capability-figure reference is optional and must not narrate panels.
+
+**Internal project imports.** Full ladder; Kármán-depth rationale; asymmetric scope; contracts ownership.
+
+**Excluded content.** Limitations inventory; negative Illusion SR story; Erase blockers; application disclaimers; experiment caveat list; values; package labels; section-number ending.
+
+**Bridge/output.** Gap/question → explicit plant, boundary, solver, Reynolds, role, comparator, metric, ROI, and window contracts.
+
+## 5. Frozen configuration terminology and Reynolds naming policy
+
+### Reader-facing configuration labels
+
+**Frozen pair:** `parabolic-inflow/no-slip-wall channel configuration` and `uniform-inflow/free-slip-wall channel configuration`. Repository configuration and kernel evidence confirm the first uses a parabolic inlet with bounce-back lateral walls (the no-slip realization) and the second uses a uniform inlet with free-slip lateral walls. These labels describe the principal reader-facing distinction without implying chronology or continuation. Other solver, inlet-scheme, outlet, geometry, body-boundary, action, and observation differences remain explicit in contracts.
+
+Internal chain IDs (`Legacy`, `V5`) are retained only for provenance, ownership, and artifact tracing; they are not manuscript adjectives.
+
+### Reynolds naming policy
+
+- Prefer the reader-facing definition $Re_D=U_{ref}D/\nu$ and define $U_{ref}$ per configuration, together with diameter $D$, viscosity $\nu$, and any relation to a code label.
+- For the uniform-inflow/free-slip-wall configuration, $U_{ref}$ is the declared uniform inlet speed.
+- For the parabolic-inflow/no-slip-wall configuration, $U_{ref}$ is the bulk/mean inlet speed and $U_{max}\simeq1.5U_{ref}$. Source-specific inherited-label conversion remains gated only where historical metadata do not establish whether `U0` meant centreline or bulk/mean speed.
+- Do not use bare `Re100`, `Re=100`, or `Kármán Re100`. Use `the periodic Kármán cloak case` or the exact configuration label until contracts define Reynolds naming.
+- Equal code labels do not imply equal physical Reynolds numbers; never unify them by label coincidence.
+- Exact case and chain IDs remain provenance, not reader-facing physical definitions.
+
+## 6. Candidate citations requiring verification
+
+`full-text web/preprint` means accessible text was inspected for architecture; it is not scientific verification or artifact audit.
+
+### Cloaking and cultural opening
+
+- Pendry, Schurig & Smith, “Controlling Electromagnetic Fields,” *Science* **312** (2006), DOI `10.1126/science.1125907`; Leonhardt, “Optical Conformal Mapping,” *Science* **312** (2006), DOI `10.1126/science.1126493`; Vasquez, Milton & Onofrei, “Active exterior cloaking for the 2D Laplace and Helmholtz equations,” *PRL* **103** (2009), DOI `10.1103/PhysRevLett.103.073901`; Ma et al., “Experiments on active cloaking and illusion for Laplace equation,” *PRL* **111** (2013), DOI `10.1103/PhysRevLett.111.173901`; Urzhumov & Smith, “Flow stabilization with active hydrodynamic cloaks,” *PRE* **86** (2012), DOI `10.1103/PhysRevE.86.056313`. Existing repository state: `full-PDF`; roles limited to genealogy/restricted assumptions.
+- **Mengyao Chen, Xiangying Shen & Lei Xu**, “Realizing the thinnest hydrodynamic cloak in porous medium flow,” *The Innovation* **3**(4) (2022), 100263, DOI `10.1016/j.xinn.2022.100263`. **Identity resolved; full-text open-access web read.** Exact modern porous-medium example; not finite-Re channel evidence.
+- **Harry Potter hook:** unresolved as a suitable manuscript citation. Search found literary theses/secondary cultural studies, but no precise high-authority source selected for the intended claim. Citation-search task: choose primary novel, scholarly history of invisibility, or omit. Safe default: generic fascination, no named hook.
+- **Radar analogy:** not verified. Search only if retained; safe default omission.
+
+### Pinball, force control, and DRL
+
+- Nan Deng, Bernd R. Noack, Marek Morzyński & Luc R. Pastur, “Low-order model for successive bifurcations of the fluidic pinball,” *JFM* **884** (2020), A37, DOI `10.1017/jfm.2019.959`. Existing R4 state: direct `full-PDF` manuscript/preprint read.
+- Guy Y. Cornejo Maceda, Yiqing Li, François Lusseyran, Marek Morzyński & Bernd R. Noack, “Stabilization of the fluidic pinball with gradient-enriched machine learning control,” *JFM* **917** (2021), A42, DOI `10.1017/jfm.2021.301`. Existing state: `full-PDF`.
+- Cédric Raibaudo, Peng Zhong, Bernd R. Noack & Robert J. Martinuzzi, “Machine learning strategies applied to the control of a fluidic pinball,” *Physics of Fluids* **32** (2020), 015108, DOI `10.1063/1.5127202`. Accessible web full-text record inspected; confirm final accenting in bibliography manager.
+- **Haodong Feng, Yue Wang, Hui Xiang, Zhiyang Jin & Dixia Fan**, “How to control hydrodynamic force on fluidic pinball via deep reinforcement learning,” *Physics of Fluids* **35**(4) (2023), 045137, DOI `10.1063/5.0142949`. **Identity/role resolved; full preprint text and final metadata checked.** Paper controls force extrema/reference tracking. Dixia Fan is a coauthor; accessible CRediT lists conceptualization, funding acquisition, and review/editing. Infer no sole-originator or experimental role.
+- **Paul Garnier, Jonathan Viquerat, Jean Rabault, Aurélien Larcher, Alexander Kuhnle & Elie Hachem**, “A review on deep reinforcement learning for fluid mechanics,” *Computers & Fluids* **225** (2021), 104973, DOI `10.1016/j.compfluid.2021.104973`. **Exact identity verified; accessible full text inspected.** Optional general review.
+- Rabault et al. (2019), DOI `10.1017/jfm.2019.62`; Paris et al. (2021), DOI `10.1017/jfm.2020.1170`; Loiseau et al. (2018), DOI `10.1017/jfm.2018.147`; Gautier et al. (2015), DOI `10.1017/jfm.2015.95`; Li & Zhang (2022), DOI `10.1017/jfm.2021.1045`; Xia et al. (2024), DOI `10.1017/jfm.2024.69`; Wang et al. (2024), DOI `10.1017/jfm.2024.333`. Existing style-guide state: `full-PDF`.
+
+**Policy:** citations support transitions, never internal results or priority. Names are retained only if their sentence-level roles survive drafting.
+
+## 7. Figure/equation/table/citation contracts for Introduction
+
+- **Introduction empirical figure:** none; `omit`.
+- **Prospective capability figure:** Capability-owned; optional/dependency-gated; may foreshadow named breadth/Kármán depth; no pooled score, chain equivalence, positive Illusion SR, or experiment; fallback omit.
+- **Conceptual workflow:** optional/currently unnecessary; realization → evaluation → action/field analysis; dependency map, never causal graph; fallback P4 prose architecture.
+- **Equations:** none. Contracts owns Reynolds, action, objective, L2, DTW, role, ROI, and window definitions. No unified Reynolds equation.
+- **Tables:** none in Introduction. Chain/metric contract table belongs downstream; planning ledgers are not manuscript tables.
+- **Citations:** P1 compact cloak cluster; P2 pinball/rotation cluster with Feng et al. isolated for force tracking; P3 DRL/partial observation with optional review; P4 closed-loop ML/sparse ROM/field analysis by role; P5 normally uncited.
+
+## 8. Ownership/import/export and duplication rules
+
+**Introduction owns:** fascination-to-gap funnel; pinball selection; force-target versus drag versus field distinction; three hooks; full high-level ladder; Kármán-depth rationale; contracts bridge.
+
+**Imports:** R2/R4 declared-reference concepts and claim ceiling; M03 separate configuration/action/observation/Reynolds/comparator/metric contracts; M01/M10/final-audit ladder and duplication rules; exact literature precedent only.
+
+**Exports:** background/non-zero target distinction; force-mediated sparse-observation question; realization/evaluation/interpretation separation; configuration-local contract requirement; Kármán as sole deep case without bare Reynolds label.
+
+**Opening dependency:** cultural hook/adjective choices do not block architecture; use safe defaults.
+
+**Closing bridge:** define plants and configuration-local contracts before capability.
+
+**Duplication:** Contracts owns definitions/equations; Capability owns Fig. 1/progression; Performance owns field values/windows; SR owns formulas/deletions; Field/CCD owns corrected means/residual/modes; Cross-case owns extensions/negative routes; Steady owns separate-chain context; Experiment owns provenance/observation; Conclusion restates the ladder once. M04/M06 are different estimands; SR magnitude and steady spin are related contextual diagnostics only—never replication, agreement, validation, or triangulation.
+
+## 9. Author decisions required
+
+1. **Cultural opening — recommendation: generic fascination, no Harry Potter.** Named hook is memorable but lacks a selected JFM-suitable citation. Safe default is one clause and immediate hydrodynamic turn.
+2. **Configuration labels — frozen.** Use `parabolic-inflow/no-slip-wall channel configuration` and `uniform-inflow/free-slip-wall channel configuration`; contracts retain all additional solver/action differences.
+3. **Mengyao Chen example — recommendation: retain only if P1 needs a recent concrete bridge.** Exact but porous-medium scope may slow the funnel. Safe default compact cluster or omit.
+4. **Feng et al./Dixia Fan precedent — recommendation: retain P2 paper collectively.** It directly supports force tracking versus drag optimization. Do not foreground individual authorship.
+5. **Experiment in P5 — recommendation: omit unless navigation is incomplete.** The downstream author-owned qualitative open-loop subsection remains reserved.
+
+## 10. Coverage ledger
+
+`mapped` is not `read`; `read` is not independently verified; neither implies artifact audit.
+
+| State | Source/input | Boundary |
+|---|---|---|
+| `read` | Current Section 1, `00_STYLE_CROSSWALK.md`, `00_GLOBAL_DECISION_LEDGER.md`, `00_MASTER_AGENT_PROMPT.md`; deleted `STYLE_01_INTRODUCTION.md` is historical/superseded provenance only | Full repository reads for style/ownership/claims; not scientific verification or artifact audit |
+| `read` | R2 `05_REVISED_STRUCTURE_FREEZE.md`, `06_SECTION_LEVEL_JFM_STYLE_GUIDE.md` | Full-report reads; architecture and literature-state records |
+| `read` | R3 M01 `01_INTRODUCTION.md`, M03 `03_CONTRACTS_METRICS.md`, M10 `10_LIMITATIONS_CONCLUSION.md`, `11_FINAL_CONSISTENCY_AUDIT.md` | Full-report reads for architecture/contracts/ladder/audit; no recomputation |
+| `read` | Chen et al. (2022) open article; Feng et al. (2023) full accessible preprint; Garnier et al. (2021) full accessible preprint/web text | Targeted identity, DOI, role, framing verification; not reproduction |
+| `mapped` | Exact JFM/cloak/control records in R2/R4, including Pendry, Leonhardt, Vasquez, Ma, Urzhumov, Deng, Cornejo Maceda, Rabault, Paris, Loiseau, Gautier, Li, Xia, Wang | Existing `full-PDF` states accepted as repository records; not reread here |
+| `mapped` | Remaining active R3 owners and artifact authorities | Ownership/boundary navigation only |
+| `not-covered` | Exhaustive cloak history; radar review; Harry Potter source adjudication; full PPO bibliography | Unneeded for architecture; retained as citation/author choices |
+| `not-covered` | Raw fields/models/hashes; CFD/training/SR/CCD/OID recomputation; figures; experiment media | Outside scope; no result, number, or panel inferred |
+| `external-inaccessible` | Undermind workspace listing/search after orientation | Workspace discovery unavailable in this scoped run; targeted public records supplied identity checks |
+| `external-inaccessible` | Hidden/symlink project artifacts and raw experiment material | Authority-led deferral; no artifact-dependent Introduction claim |
+
+## 11. Quality gate and disposition
+
+- [x] Exactly five paragraph jobs; no final manuscript paragraphs.
+- [x] P1 reaches the nonlinear finite-Re gap from fascination.
+- [x] P2 selects the pinball and distinguishes force-target matching from drag.
+- [x] P3 motivates DRL/PPO, application-driven sparse sensing, and three positive hooks.
+- [x] P4 supplies realization → independent evaluation → action/field analysis with literature roles.
+- [x] P5 closes with narrowed question, three levels, Kármán rationale, compact asymmetric scope, and contracts bridge.
+- [x] Internal chain IDs are provenance-only; the exact boundary-condition configuration pair is frozen for readers.
+- [x] $Re_D=U_{ref}D/\nu$ is preferred, $U_{ref}$ is configuration-local, no bare `Re100` is allowed, and the parabolic reference/conversion decision is assigned to Contracts.
+- [x] PPO wording is bounded to clipped-update step-size regularization relative to the previous policy and rejects mathematical guarantees of descent, monotonic improvement, convergence, or success.
+- [x] Mengyao Chen and Dixia Fan identities/roles checked; Harry Potter/radar remain explicit tasks.
+- [x] Fixed SR/Illusion/CCD/OID/steady/experiment cautions preserved.
+- [x] No new internal number, result, priority, mechanism, or panel readiness.
+- [x] Display/equation/table/citation ownership and fallbacks stated.
+- [x] Coverage states explicit.
+
+**Near-collision:** M04 policy field metrics and M06 corrected mean-field accounting remain **different estimands**. Periodic, transient, steady, native/normalized DTW, and field metrics remain non-equivalent. Separate chains are not pooled or called replication, validation, agreement, or triangulation.
+
+**Disposition: `frozen` for Round-4 architecture.** The five-paragraph logic, configuration labels, Reynolds policy, PPO boundary, evidence ladder, and contracts bridge are settled. This is still not manuscript prose; cultural-hook choice, optional Chen example, experiment mention, and sentence-level citation wording remain drafting choices and do not block or reopen the architecture.
+
+## 12. Paragraph-level literature/style enrichment — 2026-08-11
+
+**Boundary.** Post-freeze planning guidance only: no paragraph job, claim, evidence owner, display status or frozen disposition changes. Observations are bounded to the named `full-PDF` sample, not universal JFM rules; literature does not verify DynamisLab science.
+
+### P1 — fascination, definitions and finite-$Re$ boundary
+- **Selected (`full-PDF`, paragraph-specific read):** Pendry, Schurig & Smith (2006), *Science*, DOI `10.1126/science.1125907`; Ma et al. (2013), *PRL*, DOI `10.1103/PhysRevLett.111.173901`; Urzhumov & Smith (2012), *PRE*, DOI `10.1103/PhysRevE.86.056313`; Mirzakhanloo et al. (2020), *JFM*, DOI `10.1017/jfm.2020.665`; Ren et al. (2021), *Physics of Fluids*, DOI `10.1063/5.0060690`.
+- **Observed, not automatically recommended:** concrete field/flow subjects precede methods; background cloaking precedes substitute-target illusion; citations cluster by genealogy/regime; restrictions sit beside the governing regime. **Recommendation:** one fascination clause, then declared background, then non-zero illusion comparator; order nonlinearity → separation/wake → dissipation/BC/forcing → observer/target. **Green/amber/red:** green `approach a declared background`, `retarget`, `does not transfer directly`; amber `cloak` with comparator scope; red `invisible`, `undetectable`, perfect equivalence. Cite at the first technical definition; no owned display, body carries inference. Bridge to a compact force-coupled body. Anti-patterns: literary detour, catalogue, opening equation, regime promotion, applications list.
+
+### P2 — pinball selection and force mediation
+- **Selected (`full-PDF`):** Deng et al. (2020), *JFM*, DOI `10.1017/jfm.2019.959`; Cornejo Maceda et al. (2021), *JFM*, DOI `10.1017/jfm.2021.301`; Raibaudo et al. (2020), *Physics of Fluids*, DOI `10.1063/1.5127202`; Feng et al. (2023), *Physics of Fluids*, DOI `10.1063/5.0142949`.
+- **Observed:** geometry/plant precede objective/optimizer; force-reference tracking differs from extremization by naming the requested quantity first. **Recommendation:** extended-field/compact-boundary mismatch → pinball → scalar drag → time-dependent force target → independent field match; isolate Feng at force tracking. **Green/amber/red:** green `compact nonlinear plant`, `intermediary signal`; amber `control authority` as accessible actuation; red force matching proves cloaking, optimal benchmark, fully controllable. Section 2 captions own geometry/sign; P2 body owns selection. Bridge to nonstationary feedback. Anti-patterns: author catalogue, bare Reynolds label, solver detail, nearby result as present evidence.
+
+### P3 — nonlinear feedback, PPO and analytical hooks
+- **Selected (`full-PDF`):** Rabault et al. (2019), *JFM*, DOI `10.1017/jfm.2019.62`; Paris et al. (2021), *JFM*, DOI `10.1017/jfm.2020.1170`; Li & Zhang (2022), *JFM*, DOI `10.1017/jfm.2021.1045`; Xia et al. (2024), *JFM*, DOI `10.1017/jfm.2024.69`.
+- **Observed:** learning follows a physical nonlinear obstacle; partial-measurement and delay/history limits are local; observables precede reward/algorithm. **Recommendation:** open on nonstationary following; introduce DRL as response, bound PPO in one clause, then sparse sensing plus complementary force. Order hooks execution → field match beyond reward space → checkable action/field organization. **Green/amber/red:** green `learns a policy`, `uses clipped updates`, `complements`; amber `regularizes update size` relative to the previous policy; red convergence/optimality/observability guarantees and generic AI claims. Cite DRL after the obstacle and PPO beside its caveat. No tutorial, hyperparameters or result-like hooks. Bridge to layered evidence.
+
+### P4 — layered evidence chain
+- **Selected (`full-PDF`):** Cornejo Maceda et al. (2021), DOI `10.1017/jfm.2021.301`; Loiseau et al. (2018), *JFM*, DOI `10.1017/jfm.2018.147`; Li et al. (2022), *JFM*, DOI `10.1017/jfm.2022.908`; Li & Zhang (2022), DOI `10.1017/jfm.2021.1045`.
+- **Observed:** layered methods state the gap before stages, preserve dependencies, and separate online objective from post-hoc analysis; captions decode workflow while body explains need. **Recommendation:** `realize` execution → `evaluate` fields → `check` executable reduced policy → `organize` field evidence descriptively; state non-equivalence and avoid `triangulate`. **Green/amber/red:** green `evaluate independently`, `check by deployment`, `organize descriptively`; amber `interpret` only Kármán/cloak; red cross-layer validation, mechanism revelation, internal full-state reconstruction. Caption dependencies without causal arrows; body states inference/limit; bridge to the aligned deep case. Anti-patterns: package names, mini-reviews, decorative lists, formula duplication.
+
+### P5 — narrowed question and contracts bridge
+- **Selected (`full-PDF`, reused):** Deng et al. (2020), DOI `10.1017/jfm.2019.959`; Cornejo Maceda et al. (2021), DOI `10.1017/jfm.2021.301`; Loiseau et al. (2018), DOI `10.1017/jfm.2018.147`.
+- **Observed:** closing paragraphs compress objective, approach and scope before the next definition; strong verbs remain evidence-bound and values absent. **Recommendation:** exact question → three levels in ascending evidence order → Kármán-depth rationale → one asymmetric-scope sentence; keep contributions normally uncited and end on configuration, action/sign, observations, targets, comparators, metrics, ROI and windows. **Green/amber/red:** green `examines`, `evaluates`, `supports a bounded interpretation`; amber `contributes` with immediate scope; red priority, unified proof, generality, robustness, mechanism. No caption/equation; bridge through definitions. Anti-patterns: contribution bullets, negative inventory, experiment caveats, values, roadmap ending.
+
+### Compact coverage/read-state ledger
+| Job | Full-PDF sample | Lower-state boundary |
+|---|---|---|
+| P1 | Pendry; Ma; Urzhumov; Mirzakhanloo; Ren | Chen remains prior web read |
+| P2 | Deng; Cornejo Maceda; Raibaudo; Feng | None promoted |
+| P3 | Rabault; Paris; Li & Zhang; Xia | Schulman remains `mapped/full-text record` |
+| P4 | Cornejo Maceda; Loiseau; Li et al.; Li & Zhang | None promoted |
+| P5 | Deng; Cornejo Maceda; Loiseau | No literature authority for internal contributions |
+
+**Read-state note:** each `full-PDF` item above was read with one paragraph-specific question on 2026-08-11. These are rhetorical samples, not internal-science checks. Mapped/full-report/web states remain distinct; no snippet was promoted, no deep search was launched and no Undermind content was changed.

docs/JFM_WYQ/ROUND4_TOP_DOWN/SECTION_02_PROBLEM_METHOD_EVALUATION.md

@@ -0,0 +1,272 @@
+# Round 4 frozen architecture — Problem formulation, control framework and evaluation
+
+## 1. Status
+
+**Status:** `frozen` for Round-4 architecture; not final manuscript prose, a result dossier, or panel/artifact approval. Run-specific table values, bibliography completion and display production remain dependency-gated without reopening the architecture.
+
+**Scope and claim ceiling.** Section 2 establishes the physical, numerical, control and evaluation setup needed to decode Fig. 1 and all later comparisons. It may define configurations, methods, roles, objectives, metrics and qualification boundaries. It may not narrate performance, imply that current results were regenerated with the improved workflow, merge the two configurations, or use solver qualification to validate control or interpretation.
+
+## 2. Reader-facing title and purpose
+
+**Frozen reader-facing title:** **Problem formulation, control framework and evaluation**.
+
+This is a methodology/problem-formulation section. Its sequence is configuration → numerical realization → closed-loop control → target/objective → independent evaluation → compact numerical qualification. Reader-facing prose uses `configuration`, `formulation`, `definition`, `protocol`, `procedure`, `framework`, `specified cases/conditions`, and precise physical names. `Contract` remains internal governance vocabulary only.
+
+**Section question.** What physical and numerical problem is solved, what information and actuation define the learned controller, how are cloak and illusion targets constructed, and how is performance evaluated independently?
+
+**Bounded answer.** The main text presents one canonical formulation and improved workflow while identifying which existing evidence belongs to the parabolic-inflow/no-slip-wall configuration. It defines signal-level online objectives and field-level independent criteria separately, retains only current CelerisLab solver qualification, and defers implementation history and exhaustive matrices to provenance/supplement material.
+
+## 3. Author feedback and decisions incorporated
+
+- Replaced reader-facing `contracts` and `named task` language with academically natural physical/method terminology.
+- Foregrounded the canonical no-bias, physically scaled, observation-normalized PPO workflow; removed software-development chronology and reader-facing internal chain names.
+- Preserved scientific provenance: core quantitative SR, CCD and field evidence belongs to the parabolic-inflow/no-slip-wall configuration and is not represented as post-improvement retraining.
+- Required one geometry figure showing both parabolic/no-slip and uniform/free-slip configurations, fixed pinball/sensor relative positions, and case-dependent disturbance/target size.
+- Frozen article-wide bulk/mean-based $Re_D$ and exact conversion policy in §7.
+- Retained D2Q9 MRT LBM, boundaries, curved/moving-body treatment, resolution/time-step reporting, and compact selected-observable qualification in the main text.
+- Elevated sensor placement, target construction, DRL flow, DTW, registered spatial L2, and mean/phase semantics to explicit subsection jobs.
+- Preserved all fixed cautions: positive SR is Kármán/cloak-only; Illusion SR is historical/negative; CCD is descriptive/noncausal and uses `action-correlated correction-field modes`; OID is claim-free; steady evidence is contextual and separate; experiment remains qualitative/open-loop/provenance-gated.
+
+## 4. Recommended subsection architecture and logical sequence
+
+### 2.1 Flow configurations and numerical method
+
+**P1 — physical object and coordinates.** Define the three-cylinder fluidic pinball, $D$, the streamwise/transverse coordinates, fixed relative cylinder centres, and independent cylinder rotation. Explain that lateral confinement stabilizes and organizes the wake by bounding cross-stream motion and fixing a reproducible channel environment; do not imply universality or that confinement alone establishes a control mechanism.
+
+**P2 — two configurations and ownership.** Decode the parabolic-inflow/no-slip-wall and uniform-inflow/free-slip-wall configurations side by side. Define inlet, outlet, lateral boundaries, cylinder treatment and case-dependent upstream disturbance or target body. State early that the parabolic/no-slip configuration supplies the principal Kármán field/SR/CCD analysis, while the uniform/free-slip setting supplies improved-workflow capability and separate steady/contextual evidence.
+
+**P3 — nondimensionalization and Reynolds number.** Define $U_{ref}$, $D$, $\nu$, $t^*=tU_{ref}/D$, velocities and forces as needed. Use bulk/mean inlet speed for the parabolic profile, with $U_{max}\simeq1.5U_{ref}$, and uniform inlet speed for the uniform profile. Record exact inherited-label conversion without using bare `Re100`.
+
+**P4 — CFD realization.** State D2Q9 lattice Boltzmann formulation with MRT collision, collision/streaming precision, regularized or parabolic inlet as configuration-specific, outlet treatment, no-slip or free-slip lateral walls, and Bouzidi treatment for curved stationary/moving boundaries. Give grid, diameter resolution, lattice time step or mapping, sampling interval and compact adequacy statement only where current authority supports each value.
+
+**Display cadence.** Place the two-configuration geometry/sensor schematic beside P1–P2. Put a compact configuration/method table after P4 only if it compresses otherwise repeated definitions; historical implementation differences go to supplement/provenance.
+
+**Bridge.** Once the physical plants and numerical realization are decodable, define how body rotation and limited wake information close the control loop.
+
+### 2.2 Rotational actuation, observations and PPO training
+
+**P1 — actuation in physical variables.** Fix the computational body/action order as $(F,U,L)=(\text{front},\text{rear}_{y+},\text{rear}_{y-})$. The current computational convention uses $\omega>0$ in the solver's positive angular direction and the wall law $(U_w,V_w)=(-\omega r_y,\omega r_x)$; thus positive $\omega$ is counter-clockwise in the stated Cartesian coordinates. Define the manuscript command $s_i=\omega_iR/U_{ref}$. Historical SR output is exactly $\alpha_i=\omega_i/U_0$, where $\omega_i$ is the solver angular-velocity command. Thus $\alpha_i$ has inverse-length units in dimensional notation (or inverse-lattice-length in the historical lattice convention), and the physical surface-speed command is $s_i=\alpha_iR$ when $U_0=U_{ref}$. It must not be called dimensionless without also declaring a radius-one nondimensionalization. Present the canonical no-preset-bias mapping and smoothing/persistence only to the degree needed for physical reproduction. This computational map does not validate or determine the physical apparatus command map, which remains Section 9 author-supply-gated.
+
+**P2 — observations and sensor placement.** Define body-force components and both velocity components at three circular sensing regions centred at $x/D=10$ and $y/D=0,\pm2$, each of radius $0.25D$. Explain the rationale: downstream enough to carry a developed wake signature, transversely distributed to capture symmetry/deflection and shedding, and sparse enough to represent limited remote sensing. Raibaudo et al. provide the direct $x/D=10$ experimental precedent; Cornejo Maceda et al. and Li et al. support downstream multi-location wake sensing and phase/state information. These sources do **not** establish optimality, sufficiency or observability for the present arrangement.
+
+**P3 — normalization and policy representation.** Separate solver area/time averaging, fixed physical pre-scaling and online observation whitening. State that the selected policy must travel with its matching frozen normalizer. The canonical formulation has zero preset steady-derived bias; do not expose inherited factors or historical affine maps in main text.
+
+**P4 — PPO principle and project-specific settings.** In one short explanation, PPO updates a stochastic policy using a clipped surrogate objective that limits changes relative to the preceding policy; it does not guarantee gradient descent, monotonic improvement, convergence or control success. Current training authorities bind the recommended/current defaults to SB3 PPO, sinusoidal activation, two hidden layers of 64 units, rollout length 2048, batch 64, ten epochs, discount $0.995$, and default learning rate $3\times10^{-4}$. These are defaults, not universal selected-run metadata: retained run records include learning rates $10^{-4}$ and $3\times10^{-4}$. Main text reports final canonical choices and scientifically relevant project-specific deviations; exact per-result settings and full optimizer metadata go to supplement/provenance.
+
+**Design lesson boundary.** The no-bias canonical design avoids injecting a steady solution before learning. Phrase this as a design rationale, not a tested superiority result, unless a same-protocol authority is supplied; any claim that preset bias prematurely constrains learning remains evidence-gated.
+
+**Bridge.** The controller interface is now fixed; the next subsection defines what signal it is asked to approach and how that request differs from independent field evaluation.
+
+### 2.3 Target construction and control objective
+
+**P1 — role definitions.** Define `target/reference`, `controlled`, `physical zero/uncontrolled`, and, only where used later, `constant` as distinct acquisition roles. Roles are not interchangeable and do not travel between configurations merely because labels match.
+
+**P2 — cloak/background target.** Explain that a cloak objective uses a declared background or zero-body/disturbance reference appropriate to the specified case. Identify the roles needed to create the target, train/control the pinball and evaluate against an independently acquired physical-zero comparator.
+
+**P3 — illusion/non-zero target.** Explain that illusion uses a separately generated non-zero target signature, including the target-body acquisition where applicable. Target-body size and upstream disturbance/target geometry are case-dependent; no target parameter is generalized silently across cases.
+
+**P4 — online objective.** Define the reward conceptually as a weighted combination of (i) body-force-to-target discrepancy and (ii) multichannel sensor-to-target discrepancy. Explain case-dependent force aggregation and conditional/similarity mapping in prose; retain exact implementation weights only if required to reproduce a selected run and bind them to that run in the supplement. State explicitly that the online objective is an intermediary wake-signature objective, not a full-field metric.
+
+**Required schematic.** A DRL flow schematic shows: physical configuration/state → force and sparse velocity observations → normalization → PPO policy → three rotations → CFD evolution, with target signals entering the force and sensor reward terms. Caption ceiling: information/action/reward flow only; no convergence, observability, optimal sensing, field equivalence or causality.
+
+**Bridge.** Because the reward observes only integrated forces and sparse wake signals, evaluation must ask separately whether signals and registered fields approach the declared reference.
+
+### 2.4 Independent signal- and field-level evaluation
+
+**P1 — DTW algorithm and role.** Give DTW meaningful main-text space. Describe a dynamic-programming alignment path through two sampled time series under monotonic ordering and declared endpoint/window constraints. Its advantage is comparative tolerance to local time reparameterization: it can register oscillations that differ in timing and expose amplitude, frequency, phase and waveform discrepancies without privileging a single fixed phase shift. For multiple channels, compute declared per-channel costs/similarities and aggregate with explicit scaling/weights.
+
+**DTW limits.** DTW is a signal-level comparative diagnostic, not physical delay, causality, field equivalence, observability, or a universal cross-case score. Warping can conceal timing errors if unconstrained; therefore orientation (distance versus similarity), channel normalization, window, path constraints and aggregation must accompany every value. Native and normalized DTW are distinct estimands.
+
+**P2 — registered spatial L2.** Define the registered ROI, same-case roles, mask/geometry guard, velocity normalization and area weighting before results. Spatial L2 is independent field-level evaluation and is not a pipeline `Level` or `Stage`. Use the current registered ROI, $-6\leq(x-x_s)/D\leq14$ and $|(y-y_0)/D|\leq5$, a $20D\times10D$ geometry-guarded downstream rectangle registered to the sensor plane and centreline. It supersedes the earlier $[-8D,12D]$ candidate. Current artifacts do not persist solver-exact masks, so do not call it an exact common-fluid-mask metric. Use a complete-cycle mean-field error and a comparator-relative form only where target, controlled and physical-zero roles share the same acquisition definition.
+
+**P3 — phase-resolved and case-specific semantics.** Define periodic phase-resolved error separately from mean-field error. The current eight-slot diagnostic uses independently phased nearest snapshots and is not a repeated-cycle ensemble. Vortex uses event-relative offsets, not phase; Illusion uses its own periodic target contract; Erase is mean-field only if exact roles exist; steady uses late/profile criteria. No pooled efficacy score is licensed.
+
+**P4 — division of labour.** DTW evaluates sparse multichannel temporal signatures close to the controller's information space. Mean and phase spatial L2 evaluate registered fields beyond that space. Agreement is not assumed; their difference is scientifically useful because signal alignment can coexist with spatial mismatch and vice versa.
+
+**Main/supplement split.** Main text retains the DTW idea/limits, spatial definitions, registered ROI semantics and compact case-semantics table. Algorithm pseudocode, path constraints, channel-by-channel formulas, mask construction, registration details and sensitivity checks move to supplement.
+
+**Bridge.** The evaluation definitions license later comparisons but not the numerical solver itself; the selected observables therefore receive a compact configuration-bound qualification.
+
+### 2.5 Compact CFD qualification
+
+**P1 — purpose and ceiling.** State that qualification checks whether selected CelerisLab observables are sufficiently resolved under specified benchmark configurations. It does not validate PPO, SR, CCD, target construction, field interpretation or hydrodynamic cloaking.
+
+**P2 — retained anchors.** Use only the current CelerisLab dossier (`read/full-report`). Kan99b K2 is a `partial pass`: Strouhal number, mean drag and lift/drag fluctuation amplitudes pass their declared bands. The reported mean-lift sign differs because the CFD coordinate/force-direction convention is opposite to the comparison convention; this is not unresolved physics or suspected wrong rotation. Preserve the authority's aggregate `partial pass` label rather than relabelling the dossier. Sah04 S2 is a `pass` for its declared confined-channel Strouhal gate. Only K2 and S2 have retained complete physics runs in the qualification directory; other specified cases are not completed evidence there, and smoke runs are diagnostics only. Report exact values only after direct artifact/manifest binding. Do not state that an older solver was validated.
+
+**Compact table.** Columns: benchmark/configuration; selected observable; comparison authority; resolution/window; status; local caveat. Main text carries a small result/table; full matrices, hashes, sensitivity, precision and smoke-test diagnostics belong to supplement.
+
+**Closing bridge.** With configurations, controller, objectives, independent metrics and selected numerical observables defined, Fig. 1 may compare prescribed cases without pooling their evidence or anticipating the Kármán result.
+
+## 5. Detailed content controls: equations, displays, citations, split and exclusions
+
+### Required/reusable equations
+
+1. **Article-wide Reynolds number** (`required`, owner Section 2):
+   \[
+   Re_D=\frac{U_{ref}D}{\nu},\qquad U_{max}\simeq1.5U_{ref}\ \text{for the parabolic inlet}.
+   \]
+   Definition only; exact conversion note accompanies inherited labels.
+2. **Nondimensional actuation** (`required`): use $(F,U,L)=(\text{front},\text{rear}_{y+},\text{rear}_{y-})$, $\omega>0$ counter-clockwise under $(U_w,V_w)=(-\omega r_y,\omega r_x)$, and $s_i=\omega_iR/U_{ref}$. Historical SR traces/laws use $\alpha_i=\omega_i/U_0$ (inverse-length, or inverse-lattice-length, unless radius has already been nondimensionalized) and convert by $s_i=\alpha_iR$ when $U_0=U_{ref}$. Avoid obsolete affine maps in main text; experiment mapping remains author-supply-gated.
+3. **Online objective** (`required concept; formula optional`): a force discrepancy plus a sensor discrepancy, with case-specific weighting. Prefer one compact symbolic equation if reused; implementation-specific constants belong to supplement.
+4. **Mean-field L2** (`required`): area-weighted, $U_{ref}$-normalised velocity-vector error over the registered ROI, with target/control/zero roles declared.
+5. **Phase-resolved L2** (`required if later reused`): same spatial definition indexed by periodic phase slots; do not imply ensemble averaging.
+6. **Comparator-relative reduction** (`optional`): retain only if later results use exactly the same role/ROI/mask/window definition.
+7. **DTW recurrence/path equation** (`optional but recommended`): one compact recurrence or path-cost definition because DTW is uncommon in fluid mechanics; otherwise define rigorously in prose and supplement.
+
+### Display/table ownership and caption ceilings
+
+| Item | Owner and job | Status/source | Caption ceiling | Fallback |
+|---|---|---|---|---|
+| Two-configuration geometry/sensor schematic | §2.1; decode inlet/walls/outlet, fixed pinball/sensor geometry, case-dependent disturbance/target | `required`; redraw/production and exact geometry-source binding pending | Defines geometry and BCs; does not imply configuration equivalence, optimal sensing or solver-exact scale | Minimal line drawing plus configuration table |
+| DRL information/action/reward schematic | §2.3; distinguish observation, target, reward and independent evaluation | `required`; conceptual production item | No convergence, observability, optimality, causality or field equivalence | Compact flow diagram in prose/table form |
+| Configuration/method table | §2.1; compress physical/numerical ownership without history narrative | `required`; exact rows authority-bound | Defines separate configurations; no cross-configuration transfer | Split main compact table and supplement provenance matrix |
+| Case/metric semantics table | §2.4; separate periodic, event-relative, mean-only and steady estimands | `required`; definition-ready | No pooled score or result | Concise prose list |
+| Compact CFD qualification table | §2.5; qualify selected CelerisLab observables | `required`; current dossier/artifact bound | Solver-observable qualification only | One paragraph plus supplement pointer |
+
+### Main text versus supplement/provenance
+
+**Main text:** two physical configurations; geometry and BCs; article-wide Reynolds convention; essential LBM/MRT and curved-boundary method; physical actuation/observations; canonical normalization/PPO settings; target/reward concept; sensor rationale; DTW and spatial L2 division; case semantics; compact CelerisLab qualification; one minimal provenance statement.
+
+**Supplement/provenance:** internal chain IDs; old versus current action maps, scaling, bias, normalization and DTW variants; software chronology; complete configs/hashes; full PPO tables; exact reward mappings/weights not needed in the argument; qualification matrices and sensitivities; mask/registration implementation; historical solver details; selected policy/normalizer/target binding.
+
+**Explicit exclusions:** no performance values except the compact qualification table; no training diary; no claim that PPO converges; no `named task`; no reader-facing `Legacy`/`V5`; no older-solver validation; no target/control/zero role pooling; no force or DTW claim as field matching; no sensor optimality; no confinement-causes-control mechanism; no positive Illusion SR, causal CCD, OID result, steady transfer, or quantitative experiment claim.
+
+## 6. Canonical method versus evidence provenance policy and rerun gate
+
+### Main-text policy
+
+Present the improved workflow as the **canonical method formulation**: physical action nondimensionalization, zero preset steady-derived bias, explicit normalization layers, matched policy/normalizer, current CelerisLab configuration and concise project-specific PPO settings. Do not write a Legacy-versus-V5 comparison in the scientific narrative.
+
+### Compact provenance statement
+
+Use one short neutral main-text sentence: “The deep analyses reported later use data generated with an earlier implementation under the parabolic-inflow/no-slip-wall physical configuration; implementation differences do not change the physical problem and are documented in the supplement.”
+
+Detailed implementation differences and the fact that no complete canonical rerun was performed belong in supplement/provenance. Do not say that the earlier results were generated by the canonical pipeline or that equivalence, reproduction, or a matched rerun was demonstrated.
+
+## 7. Terminology and notation freeze
+
+- **Section title:** `Problem formulation, control framework and evaluation`.
+- **Reader-facing configurations:** `parabolic-inflow/no-slip-wall channel configuration`; `uniform-inflow/free-slip-wall channel configuration`.
+- **Provenance only:** internal `Legacy` and `V5` identifiers; never reader-facing adjectives.
+- **Cases:** use `periodic Kármán cloak case`, `Vortex case`, `Illusion case`, `Erase case`, or exact physical target/condition; never `named task` or bare `Re100`.
+- **Article-wide Reynolds policy:** $Re_D=U_{ref}D/\nu$. $U_{ref}$ is the uniform inlet speed in the uniform configuration and the bulk/mean inlet speed in the parabolic configuration; $U_{max}\simeq1.5U_{ref}$ for that profile. Where authoritative two-dimensional code labels used $2D$ as reference length, manuscript $Re_D$ is exactly one-half of the inherited code label. Equal labels never imply equal physical Reynolds numbers.
+- **Execution:** `wake-signature matching using forces and sparse downstream velocity observations`.
+- **Evaluation:** `field matching`; spatial L2 primary where supported; DTW complementary, signal-level and case-specific.
+- **Spatial notation:** distinguish ROI, geometry guard/solid mask, area weights, $E_{mean}$, $E_{phase}$ and any relative comparison. Spatial L2 is never `Level-2` or `Stage-2`.
+- **Roles:** target/reference, controlled, physical zero/uncontrolled, constant (only where acquired).
+- **Strength:** `defines`, `uses`, `compares`, `qualifies the selected observable`; not `fully validates`, `proves`, `guarantees`, `optimal`, `robust`, `equivalent`, or `mechanism` without dedicated authority.
+
+## 8. Result, display and equation ownership inventory
+
+| Item | Section 2 ownership | Later ownership / prohibition |
+|---|---|---|
+| Geometry, BCs, nondimensionalization, solver | Define once here | Later captions identify configuration without redefining method |
+| Actuation, observations, normalization, PPO settings | Canonical definitions here | SR owns its exact historical executable law; no action-map transfer |
+| Target/reference and online objective | Define roles and objective here | Capability/cross-case own outcomes under exact cases |
+| Sensor placement | Coordinates/radius/rationale and literature boundary here | No later optimality or observability inference |
+| DTW | Algorithm, aggregation, orientation and limits here | Performance/cross-case own values/windows |
+| Spatial L2/ROI | Reusable definitions here | Kármán performance owns detailed values; Field/CCD uses a different estimand and must not claim replication/agreement |
+| Mean/phase/event/steady semantics | Definition table here | Each result owner renews changed roles/window only |
+| CFD qualification | Compact selected-observable table here | No result section may promote it to controller/interpretation validation |
+| SR/CCD/OID/steady/experiment | Definitions only when needed for boundaries | Owned downstream under fixed scientific cautions |
+
+**Near-collision freeze.** The standardized Kármán field metric and corrected CCD mean-field accounting remain **different estimands** because acquisitions, role sets, masks, weighting/normalization and windows differ. Periodic mean/phase, event-relative, steady profile, and native/normalized DTW also remain different estimands. SR action magnitude and steady spin are related contextual diagnostics only. No arithmetic pooling, replication, validation, agreement, closure or triangulation wording.
+
+## 9. Exact literature/style identities and read levels
+
+### Retained from `STYLE_02_CONTRACTS_METRICS.md`
+
+1. G. Y. Cornejo Maceda, Y. Li, F. Lusseyran, M. Morzyński & B. R. Noack (2021), “Stabilization of the fluidic pinball with gradient-enriched machine learning control”, *Journal of Fluid Mechanics* **917**, A42. DOI `10.1017/jfm.2021.301`. `full-PDF`, targeted section-architecture read; metadata and relevant setup/sensor passages checked, not reproduced.
+2. J. Rabault, M. Kuchta, A. Jensen, U. Réglade & N. Cerardi (2019), “Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control”, *Journal of Fluid Mechanics* **865**, 281–302. DOI `10.1017/jfm.2019.62`. `full-PDF`, targeted section-architecture read; not equation-by-equation verified.
+3. R. Paris, S. Beneddine & J. Dandois (2021), “Robust flow control and optimal sensor placement using deep reinforcement learning”, *Journal of Fluid Mechanics* **913**, A25. DOI `10.1017/jfm.2020.1170`. `full-PDF`, targeted section-architecture read; not equation-by-equation verified.
+4. X. Mao, H. M. Blackburn & S. J. Sherwin (2015), “Nonlinear optimal suppression of vortex shedding from a circular cylinder”, *Journal of Fluid Mechanics* **775**, 241–265. DOI `10.1017/jfm.2015.304`. `full-PDF`, targeted section-architecture read; not equation-by-equation verified.
+
+**Observed practice retained:** these papers vary in headings and equation density, but share dependency order from physical plant/baseline through objective/method and numerical qualification to results. Reusable definitions earn equations; geometry/sensors appear early; qualification caveats sit beside the observable they limit; tables compress settings or convergence rather than dumping provenance. This observation supports the architecture but establishes no project value or validity.
+
+### Sensor-placement and DRL authorities
+
+5. C. Raibaudo, P. Zhong, B. R. Noack & R. J. Martinuzzi (2020), “Machine learning strategies applied to the control of a fluidic pinball”, *Physics of Fluids* **32**, 015108. DOI `10.1063/1.5127202`. `full-text web/accepted-manuscript` targeted read. Direct precedent for three downstream hot-wire sensors at $x/D=10$ used to monitor/evaluate the wake; it does not establish the present transverse coordinates, radius, two-component measurement or optimality.
+6. S. Li, W. Li & B. R. Noack (2022), “Machine-learned control-oriented flow estimation for multi-actuator multi-sensor systems exemplified for the fluidic pinball”, *Journal of Fluid Mechanics* **952**, A36. DOI `10.1017/jfm.2022.908`. `full-report/full accessible manuscript` targeted sensor-placement read. Supports downstream multi-location velocity sensing for phase/state information and explicitly builds on Cornejo Maceda et al.; its nine-probe grid is not the present three-disk geometry.
+7. L. Marra et al. (2024), “Actuation manifold from snapshot data”, *Journal of Fluid Mechanics*. DOI `10.1017/jfm.2024.593`. Repository state `mapped/full record`; the full PDF was not read in this audit. Use only for low-dimensional controlled-flow plausibility and compact-sensing rationale, not exact placement, optimality or observability.
+8. J. Schulman, F. Wolski, P. Dhariwal, A. Radford & O. Klimov (2017), “Proximal Policy Optimization Algorithms”, arXiv:1707.06347. Read state inherited as exact original citation, `mapped/full-text record`; use only for the clipped-update principle unless directly reread during bibliography finalization.
+9. P. Garnier, J. Viquerat, J. Rabault, A. Larcher, A. Kuhnle & E. Hachem (2021), “A review on deep reinforcement learning for fluid mechanics”, *Computers & Fluids* **225**, 104973. DOI `10.1016/j.compfluid.2021.104973`. `full-text` state inherited from the Introduction freeze; optional broad DRL precedent.
+
+## 10. Dependency-gated execution items
+
+No unresolved author choice blocks the architecture. Before numerical drafting or production, bind exact per-result PPO settings, grid/time/precision values, selected policy/normalizer/target, qualification rows, and every displayed inherited Reynolds label. These are execution dependencies, not architecture decisions. Zero preset bias remains a design rationale rather than a tested superiority claim.
+
+The article-wide convention uses bulk/mean inlet speed for the parabolic profile, with the ideal planar relation $U_{max}=1.5U_{ref}$. The current training authority verifies $Re_D=Re_{code}/2$ for its uniform-inflow code labels. An inspected historical parabolic repository note instead labels `U0` as inlet centreline velocity, so parabolic inherited labels must be converted only after source metadata identify centreline versus bulk/mean speed; no unsupported half-label conversion is inferred there.
+
+## 11. Coverage ledger
+
+`mapped` is not `read`; `read` is not independently verified; neither implies artifact audit.
+
+| State | Source/input | Why in scope | Evidence boundary |
+|---|---|---|---|
+| `read` | Round-4 `README.md`, `00_MASTER_AGENT_PROMPT.md`, `00_GLOBAL_DECISION_LEDGER.md`, `00_DISPLAY_EQUATION_INVENTORY.md`, `00_QUALITY_GATE.md`, `00_STYLE_CROSSWALK.md`, current `SECTION_02_PROBLEM_METHOD_EVALUATION.md`, and `SECTION_01_INTRODUCTION_FREEZE.md`; deleted Style 01/02 files are historical/superseded provenance only | Workflow, author decisions, style evidence, interfaces, deletion check | Full repository reads; not independent scientific verification or artifact audit |
+| `read` | Round-2 `05_REVISED_STRUCTURE_FREEZE.md`, `06_SECTION_LEVEL_JFM_STYLE_GUIDE.md`; Round-3 `03_CONTRACTS_METRICS.md`, `00_MODULE_INTERFACE_LEDGER.md`, `11_FINAL_CONSISTENCY_AUDIT.md` | Controlling architecture, method dossier, ownership and consistency | `full-report`; internal values/artifacts not recomputed |
+| `read` | `src/drl_pinball/train/TRAIN_PIPELINE.md`, `DESIGN_DECISIONS.md` | Current canonical training method, settings and explicit no-post-fix-rerun boundary | Current authority read directly; selected historical policies/results not independently rerun or artifact-audited |
+| `read` | Cornejo Maceda et al., Rabault et al., Paris et al., Mao et al. records retained from Style 02 | JFM organization and method-section practice | Declared `full-PDF` targeted reads; literature is style/context only |
+| `read` | Raibaudo et al. accessible accepted-manuscript/web text; Li et al. accessible full manuscript/repository copy | Sensor precedent and rationale | Targeted full-text reads; do not establish project geometry or optimality |
+| `read/full-report` | `CELERISLAB_VALIDATION_DOSSIER.md` | Current CelerisLab qualification architecture | Kan99b K2 is `partial pass` (selected frequency/drag/fluctuation observables pass; the mean-lift sign differs under the understood CFD coordinate/force-direction convention, not unresolved physics or wrong rotation; retain the dossier's aggregate label) and Sah04 S2 is `pass`; exact publication values remain artifact/manifest-bound; no older-solver validation |
+| `mapped` | Current solver/config/environment/evaluation code and manifests named by M03/training authorities | Exact implementation and artifact provenance | Located through read authorities; not exhaustively read or rerun |
+| `not-covered` | Raw policies/normalizers/targets, external fields, masks, hashes, full qualification matrices, new CFD/training/SR/CCD | Outside section-freeze scope | No rerun, equivalence, result or panel inferred |
+| `not-covered` | Full DTW sensitivity study, exhaustive PPO tutorial, software history and old implementation comparison | Supplement/provenance or unnecessary | Main architecture remains compact and physical |
+| `read` | `WAKE_L2_AND_SENSOR_PLACEMENT_RESEARCH_NOTES.md` | Exact sensor geometry/rationale, registered ROI, metric/phase semantics and bibliography boundaries | Working full-report authority reconciled with code on 2026-08-09; not an artifact audit |
+| `mapped/full record` | Marra et al. (2024), DOI `10.1017/jfm.2024.593` | Low-dimensional controlled-flow plausibility | Full PDF not read here; no exact placement, optimality or observability claim |
+| `external-inaccessible` | Hidden/external evidence payloads named by manifests | Artifact support outside the visible checkout | Authority-led deferral; no result inferred |
+
+## 12. Quality gate and disposition
+
+- [x] Frozen integrated section/style architecture, not manuscript prose.
+- [x] Reader-facing title and language are physical and JFM-compatible; `contracts`, `named task`, and internal chain labels are removed from reader-facing recommendations.
+- [x] Logical sequence makes Fig. 1 and later results decodable.
+- [x] Both inlet/wall configurations, confinement rationale, fixed sensor relationship and case-dependent disturbance/target geometry are represented.
+- [x] Bulk/mean-based article-wide $Re_D$ policy and centreline relation are explicit; bare `Re100` is prohibited.
+- [x] Main-text CFD, curved/moving boundary treatment, compact selected-observable qualification and older-solver exclusion are specified.
+- [x] Canonical no-bias/normalization/PPO method and historical-evidence provenance are separated; rerun completion/equivalence is not asserted.
+- [x] Sensor coordinates, radius, both velocity components, rationale, current ROI/metric authority, exact citation roles and non-optimality ceiling are retained.
+- [x] Target construction, reward roles and DRL flow schematic have distinct ownership.
+- [x] DTW receives algorithmic space and exact limitations; L2/ROI/mean/phase/case semantics remain independent.
+- [x] Displays/equations have owner, job, readiness, ceiling and fallback; inventory does not imply artifacts exist.
+- [x] Exact Style 02 identities/DOIs/read levels and observed practice are preserved.
+- [x] Canonical-main-text policy, compact earlier-implementation provenance, supplement detail, coverage states and fixed scientific cautions are explicit.
+- [x] `STYLE_02_CONTRACTS_METRICS.md` may be deleted because all unique necessary style evidence, identities/read levels, observed practice, recommendations, exclusions and ownership have been integrated here.
+- [x] `STYLE_01_INTRODUCTION.md` may be deleted because its unique necessary Introduction style evidence and exact identities/read states are preserved in `SECTION_01_INTRODUCTION_FREEZE.md`; no Section 1 architecture or caution depends solely on the style memo.
+
+**Disposition:** `frozen` for Round-4 Section 2 architecture. No unresolved author choice remains. Run-specific numerical rows, source-specific parabolic Reynolds conversions, bibliography completion and display production are dependency-gated execution tasks and do not block the freeze.
+
+## 13. Paragraph-level literature/style enrichment — 2026-08-11
+
+**Boundary.** Post-freeze planning guidance only: no method, value, provenance statement, claim ceiling, display readiness, ownership or disposition changes. Observations are bounded to the named `full-PDF` sample; literature cannot qualify DynamisLab numerics or promote an artifact.
+
+### §2.1 — configuration and numerics dependency order
+- **Selected (`full-PDF`, paragraph-specific read):** Cornejo Maceda et al. (2021), *JFM*, DOI `10.1017/jfm.2021.301`; Rabault et al. (2019), *JFM*, DOI `10.1017/jfm.2019.62`; Mao et al. (2015), *JFM*, DOI `10.1017/jfm.2015.304`; Li & Zhang (2022), *JFM*, DOI `10.1017/jfm.2021.1045`.
+- **Observed, not automatically recommended:** geometry/normalization precede numerical realization/control; reusable equations follow physical meaning; qualification follows the discretization/observable it limits. Definitions use present tense; completed checks use past tense. **Recommendation:** object/coordinates → separate configurations/ownership → nondimensionalization → CFD; define $Re_D$ after inlet references and keep confinement rationale beside its non-mechanism caveat. **Green/amber/red:** green `defines`, `uses`, `is resolved with`; amber `adequate` only for a named observable/condition; red `validated solver`, unsupported `grid independent`, equivalent configurations, canonical rerun. Cite exact methods locally; captions decode boundaries/grids, body states consequence. Bridge to action and wake information. Anti-patterns: solver-first order, action sign in geometry, implementation chronology, bare `Re100`, footnote-only configuration split.
+
+### §2.2 — action/sign, observations and PPO
+- **Selected (`full-PDF`):** Raibaudo et al. (2020), *Physics of Fluids*, DOI `10.1063/1.5127202`; Cornejo Maceda et al. (2021), DOI `10.1017/jfm.2021.301`; Li et al. (2022), *JFM*, DOI `10.1017/jfm.2022.908`; Xia et al. (2024), *JFM*, DOI `10.1017/jfm.2024.69`; Rabault et al. (2019), DOI `10.1017/jfm.2019.62`.
+- **Observed:** plant/actuator precede observations, objective and optimizer; sensor coordinates/components, rationale and limited-observation caveats are local; scaling precedes algorithms. **Recommendation:** action order/sign/wall law/command → forces and velocity disks → averaging/scaling/whitening with matched normalizer → PPO. Derive sign from wall law before $s_i$; keep historical $\alpha_i$ units adjacent. Cite Raibaudo for $x/D=10$ and Cornejo Maceda/Li for phase/state precedent, with non-optimality/non-observability local. **Green/amber/red:** green `commands`, `measures`, `area-averages`, `normalizes`, `clipped surrogate`; amber `captures asymmetry/phase information` as rationale; red ungrounded sign, dimensionless $\alpha_i$, optimal/observable sensors, PPO convergence, zero-bias superiority. Computational authority owns sign; PPO citation/caveat stay together. Geometry caption carries coordinates/sign; body carries role/limit; flow caption decodes arrows. Bridge to target definition. Anti-patterns: PPO before interface, merged normalization layers, detached normalizer, apparatus-map transfer.
+
+### §2.3 — target roles and online objective
+- **Selected (`full-PDF`):** Ma et al. (2013), *PRL*, DOI `10.1103/PhysRevLett.111.173901`; Feng et al. (2023), *Physics of Fluids*, DOI `10.1063/5.0142949`; Rabault et al. (2019), DOI `10.1017/jfm.2019.62`; Li & Zhang (2022), DOI `10.1017/jfm.2021.1045`.
+- **Observed:** comparator roles precede methods; background cloak precedes non-zero illusion; force reference is named before tracking differs from minimization; online reward follows the interface. **Recommendation:** roles → background → non-zero target → force-plus-sensor discrepancy; separate target generation from physical-zero acquisition and declare case-dependent geometry at first use. **Green/amber/red:** green exact role names and `compares signals`; amber `similarity` only with orientation/scaling; red unsupported `ground truth`, ambiguous `zero`, reward-as-field-match, role transfer. Ma supports target order; Feng supports reference versus minimization. Put full-field non-equivalence after reward. Caption places independent evaluation outside the loop; body explains the boundary. Bridge to independent evaluation. Anti-patterns: undefined slash roles, caption-only target construction, merged weights, code narration.
+
+### §2.4 — DTW and spatial L2 independent evaluation
+- **Selected (`full-PDF`):** Ishar et al. (2019), *JFM*, DOI `10.1017/jfm.2019.447`; Li & Zhang (2022), DOI `10.1017/jfm.2021.1045`; Loiseau et al. (2018), *JFM*, DOI `10.1017/jfm.2018.147`; Li et al. (2022), DOI `10.1017/jfm.2022.908`.
+- **Observed:** compared objects precede recurrence/norm; elementary distance precedes aggregation; sensitivities follow affected definitions; online objective and post-hoc analysis remain distinct. **Recommendation:** DTW object/path/orientation → local DTW limits → registered ROI/mask/weights/roles → mean/phase/event/steady semantics → division of labour. Define constraints before tolerance and comparators/guard before L2. `Independent` means outside reward for same-case roles, not triangulation. **Green/amber/red:** green `aligns under declared constraints`, `registered spatial error`, `complementary`; amber `independent` under that limit; red `corrects phase`, physical delay, field equivalence, common score, exact common-fluid mask. Put algorithm sources at path definition and warping caveat locally; Ishar marks the non-periodic boundary only. Semantics captions decode roles/windows; body explains distinct questions. Bridge to selected-observable qualification. Anti-patterns: DTW afterthought, L2 before ROI, nearest-snapshot phase average, pooling, M04/M06 collision.
+
+### §2.5 — compact qualification and provenance close
+- **Selected (`full-PDF`):** Rabault et al. (2019), DOI `10.1017/jfm.2019.62`; Cornejo Maceda et al. (2021), DOI `10.1017/jfm.2021.301`; Mao et al. (2015), DOI `10.1017/jfm.2015.304`; Paris et al. (2021), DOI `10.1017/jfm.2020.1170`.
+- **Observed:** checks follow what they qualify and precede results; observable-specific rows replace provenance dumps; caveats sit in affected rows; tables carry conditions/values and body states one bounded inference. **Recommendation:** purpose/ceiling → K2/S2 exact labels → local incomplete/smoke boundary → frozen neutral provenance sentence. Preserve K2 `partial pass` and put coordinate/force-direction explanation in or beside its row. Keep implementation differences/no-rerun supplementary. **Green/amber/red:** green `qualifies selected observables`, `passes the declared gate`, `retains partial pass`; amber `sufficiently resolved` only with named condition; red solver/control validation, reproduction, equivalent implementation, canonical rerun, unresolved-physics sign framing. Benchmark citations belong in exact rows; internal dossier/artifacts own status. Table decodes conditions; body states inference. Bridge to non-pooled case capability. Anti-patterns: full matrices, smoke promotion, old-solver validation, detached provenance.
+
+### Compact coverage/read-state ledger
+| Job | Full-PDF sample | Lower-state boundary |
+|---|---|---|
+| §2.1 | Cornejo Maceda; Rabault; Mao; Li & Zhang | CelerisLab dossier remains internal `read/full-report` |
+| §2.2 | Raibaudo; Cornejo Maceda; Li et al.; Xia; Rabault | Schulman remains `mapped/full-text record` |
+| §2.3 | Ma; Feng; Rabault; Li & Zhang | None promoted |
+| §2.4 | Ishar; Li & Zhang; Loiseau; Li et al. | DTW foundations remain mapped unless reread |
+| §2.5 | Rabault; Cornejo Maceda; Mao; Paris | Qualification status remains internal-only |
+
+**Read-state note:** each `full-PDF` item above was read with one self-contained paragraph-specific question on 2026-08-11. These are style samples only. Mapped/full-report/repository states remain distinct; no mapped paper was promoted, no deep search was launched and no Undermind content was written.

docs/JFM_WYQ/ROUND4_TOP_DOWN/SECTION_03_CAPABILITY_PROGRESSION.md

@@ -0,0 +1,114 @@
+# Section 3 — From steady cloaking to wake retargeting
+
+**Status:** R5 author-choice gate closed at the two-figure safe default; integrated Round-4 section/style packet and R5 drafting authority. This is not final manuscript prose or a claim that either composition has passed source, provenance, caption, artwork or publication-size review.
+
+## Physical question and concise answer
+
+**Question.** What should the reader see first to grasp the range of flows produced by learned rotational control, and why does periodic Kármán cloaking become the paper’s central case?
+
+**Answer.** Preview the historical steady profile discovery, then make the periodic Kármán cloak the dominant representative comparison. Add one source-bound Taylor-vortex event view and one source-bound 0.75-radius Illusion target → controlled → physical-zero view to establish breadth. Detailed event sequences, changed-target triplets and efficacy assessment belong to Section 7. The distinct configurations and temporal coordinates make this a progression, not a pooled test.
+
+## Six-paragraph flow and figure cadence
+
+### 1. Open with the observable transformation
+
+Open with the physical inference: learned rotation can approach declared steady, periodic, event-relative and non-zero references under distinct definitions, while only the periodic Kármán cloak supports deep evaluation. Use the figures as evidence for that inference. Keep crop, scale, panel geometry and publication-size composition in the Figure Plans; body jobs do not narrate layout.
+
+### 2. Recover the steady background in measured components
+
+In the uniform-inflow/free-slip-wall configuration, the historical controlled endpoint follows the uniform target in both velocity components more closely than the stationary pinball profile. Use $U_\infty\equiv U_{ref}$ only for this uniform-inflow configuration. Section 8 owns the exact spin ratio, endpoint metric, settling qualification and theory context.
+
+### 3. Make periodic Kármán cloaking the visual centre
+
+The large target → controlled → physical-zero vorticity triplet should be read before any scalar annotation. In the parabolic-inflow/no-slip-wall reference case, the controlled wake preserves the alternating street and periodic organization far more closely than the passive pinball. A rounded annotation may state that the independent complete-cycle mean and eight-slot phase velocity errors are each about 82% lower than physical zero. Section 4 owns the exact values, ROI, phase construction and DTW; repeating those details here would weaken the reveal. This block explains the choice of Kármán for depth without implying that reward or visual resemblance is itself a field metric.
+
+### 4. Let the isolated vortex unfold as an event
+
+Use one representative source-bound Taylor-vortex event view in target → controlled → physical-zero order. It previews a named event-relative role-field response only; Section 7 owns offset selection, event evolution, detailed role fields and the unresolved quantitative evaluator binding.
+
+### 5. End with a changed target, not a footnote
+
+Use one representative source-bound 0.75-radius Illusion view in target → controlled → physical-zero order. It previews positive PPO retargeting to a declared non-zero reference under the uniform-inflow/free-slip-wall configuration; Section 7 owns the detailed triplet, acquisition binding and quantitative gap. No positive Illusion SR is implied.
+
+### 6. Select the case that can sustain deeper questions
+
+Close by stating what changes across the progression: a steady profile becomes a periodic street, periodic phase becomes an isolated event coordinate, and the desired field changes from a background to a non-zero wake. Kármán is selected for detailed analysis because it uniquely combines repeated learning trajectories, retained condition variation, sparse signal histories, registered target/control/physical-zero fields, executable policy reduction and corrected field analysis. The next question is consequently quantitative and case-specific: over which tested Kármán conditions is high-performance control learned, and in the reference wake how closely do fields and measured trajectories approach the target?
+
+## Recommended figures and responsibilities
+
+1. **S3-1 — Steady downstream profile recovery** (`FIGURE_PLANS/SECTION_03_FIGURE_01_STEADY_DISCOVERY_PROFILE.md`). Two primary profile panels at $x/D=10$, with optional compact streamline context. It owns the historical first result and preserves actual component information.
+2. **S3-2 — Wake cloaking and retargeting** (`FIGURE_PLANS/SECTION_03_FIGURE_02_WAKE_CLOAKING_AND_RETARGETING.md`). One Kármán-dominant composition with a periodic target/control/zero triplet, one source-bound representative Taylor-vortex view/comparison and a 0.75-radius Illusion target/control/zero triplet.
+
+This two-figure arrangement is preferred over forcing all material into one very tall plate: the profiles require graph axes and fine component differences, whereas the wake composition relies on spatial fields and rapid visual comparison. If page layout demands one figure, S3-1 can become a narrow top strip above S3-2; it must not be reduced to a scalar badge.
+
+## Result ownership and interfaces
+
+Section 3 owns the visual reveal, steady-first chronology, the rounded Kármán field-improvement preview, one representative Vortex preview and one representative positive 0.75-radius Illusion PPO preview. Section 2 supplies configuration, role, Reynolds-number and metric definitions. Section 4 owns the Kármán condition range, five learning trajectories, exact mean/phase L2, DTW, actions and sensor portrait. Section 7 owns detailed Vortex and Illusion interpretation and their quantitative gaps; Section 8 owns steady kinematics, endpoint physics and theory limits. No SR law, deletion, corrected four-role mean field, CCD mode or experiment panel enters either Section-3 figure.
+
+## Local inference limits
+
+- The four displays do not share a temporal averaging rule, physical configuration or efficacy metric.
+- The Kármán field preview is an independent spatial comparison; reward and DTW remain separate quantities.
+- The single Taylor-vortex preview identifies source-bound event roles and morphology only; no evolution or quantitative event efficacy is inferred.
+- The Illusion block is positive PPO capability only. Target has no action trajectory, and no target-action proxy is constructed.
+- Visual similarity does not imply complete field identity, arbitrary-flow control, transfer, stability, optimality or a common mechanism.
+
+## Compact source, production and reading appendix
+
+### Existing material and production decision
+
+- **Steady profiles — redraw from data.** `src/steady_pinball_theory/figures/profiles_long.csv`, `figures/summary/profile_Einf_vector.csv`, `figures/manifest.json`, and `plot_results.py`. The current CSVs are direct redraw sources. Optional streamlines `figures/fields/streamlines/cfd-accepted-s3p55.png` and `cfd-stationary.png` are source views to crop/recompose, not final panels.
+- **Periodic Kármán — crop/recombine or redraw from roles.** `src/drl_pinball/data/reproduction/legacy/karman_re100/{target,controlled,zero}/phase_fields.npz`, metadata and summaries; current diagnostic `data/reproduction_plots/legacy/karman_re100.png`; renderers `plot_reproduction_summary.py` and `plot_flow_fields.py`. The diagnostic already supplies the role triplet but includes sensor portraits and full-domain framing that should not be inherited here.
+- **Taylor vortex — select one representative comparison from existing event fields.** `data/reproduction/legacy/vortex_taylor_y000/{target,controlled,zero}/event_fields.npz` contains offsets $[-10,-5,0,5,10]$; use one source-bound offset/view only in Section 3. `plot_flow_fields.py` renders every offset, and `docs/My_Confirmation/Graphs/cloak_tylor.pdf` is provenance/design precedent only; Section 7 owns any detailed event sequence.
+- **Illusion — crop/recombine or redraw from roles.** `data/reproduction/v5/ill_075L/target`, `ill_075L_seed43/controlled`, `ill_075L/zero`; current diagnostic `data/reproduction_plots/v5/ill_075L_seed43.png`; manifest and plotting scripts above. Use phase slot 0 unless a later production audit records another literal phase.
+
+### Literature identities and read states
+
+External literature informs presentation only and supplies no project result. Metadata and DOI were checked in Undermind on 2026-08-11; all seven records below were read at `full-PDF` level with paragraph-specific questions, without numerical reproduction or phrase copying.
+
+1. Jean Rabault, Miroslav Kuchta, Atle Jensen, Ulysse Réglade & Nicolas Cerardi, “Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control,” *Journal of Fluid Mechanics* **865** (2019), 281–302, DOI `10.1017/jfm.2019.62`.
+2. Romain Paris, Samir Beneddine & Julien Dandois, “Robust flow control and optimal sensor placement using deep reinforcement learning,” *Journal of Fluid Mechanics* **913** (2021), A25, DOI `10.1017/jfm.2020.1170`.
+3. Chao Xia, Jiaqi Zhang, Eric C. Kerrigan & Georgios Rigas, “Active flow control for bluff body drag reduction using reinforcement learning with partial measurements,” *Journal of Fluid Mechanics* **981** (2024), A17, DOI `10.1017/jfm.2024.69`.
+4. Qiulei Wang, Dong Yan, Guodong Hu, Gang Chen, Jean Rabault & Bernd R. Noack, “Dynamic feature-based deep reinforcement learning for flow control of circular cylinder with sparse surface pressure sensing,” *Journal of Fluid Mechanics* **988** (2024), A4, DOI `10.1017/jfm.2024.333`.
+5. Xuerui Mao, Hugh M. Blackburn & Spencer J. Sherwin, “Nonlinear optimal suppression of vortex shedding from a circular cylinder,” *Journal of Fluid Mechanics* **775** (2015), 241–265, DOI `10.1017/jfm.2015.304`.
+6. Guy Y. Cornejo Maceda, Yiqing Li, François Lusseyran, Marek Morzyński & Bernd R. Noack, “Stabilization of the fluidic pinball with gradient-enriched machine learning control,” *Journal of Fluid Mechanics* **917** (2021), A42, DOI `10.1017/jfm.2021.301`.
+7. Ji-chao Li & Mengqi Zhang, “Reinforcement-learning-based control of confined cylinder wakes with stability analyses,” *Journal of Fluid Mechanics* **932** (2022), A44, DOI `10.1017/jfm.2021.1045`.
+
+### Observed practice versus Section-3 recommendation
+
+Across this seven-paper sample, the recurring observed practice is to define a baseline, target or reference before interpreting a controlled state; establish the principal result before adding sensor, action or mechanistic detail; use captions to decode variables, conditions, symbology and representative-state selection; and place a limitation where it changes the adjacent inference. The papers differ in whether scalar performance or fields appear first, so that ordering is not a universal JFM convention. **Recommendation for this section:** the viewing contract, not a scalar score, should lead because Section 3 is a capability reveal. Preserve target → controlled → physical-zero order, allocate visual area by argumentative importance, and let each paragraph state one inference and one local ceiling. Use present tense for what a displayed comparison shows and past tense only for source-bound acquisition or selection actions.
+
+### Paragraph-level style enrichment
+
+The support sets below identify 3–5 close papers for each paragraph job. They support rhetorical decisions only; citations, if retained in Round 5, should sit after the external methodological or presentation context they actually support, never after a sentence reporting a DynamisLab result.
+
+1. **Viewing contract — support: Rabault et al. (2019), Xia et al. (2024), Mao et al. (2015), Cornejo Maceda et al. (2021).** Observed: these papers make the comparator and displayed variable legible before physical interpretation, then move from a representative outcome toward deeper evidence. Recommendation: open with the declared-reference comparison and explicitly tell the reader that the four blocks are non-equivalent demonstrations. **Green:** `shows`, `compares`, `previews`, `under its declared reference`. **Amber:** `approaches` or `recovers`, only with the named component/comparator beside it. **Red:** `validates`, `generalizes`, `across flows`, `common performance`, or any sentence that invites a pooled ranking. Put the non-equivalence caveat in this opening paragraph, not only in the appendix.
+
+2. **Qualitative steady seed — support: Cornejo Maceda et al. (2021), Mao et al. (2015), Rabault et al. (2019).** Observed: a representative controlled state is useful when it is tied to a specific baseline and measured observable; detailed mechanism and parametric interpretation follow later. Recommendation: name the two velocity components and the stationary comparator, keep the endpoint qualitative, and bridge by asking what changes when the reference itself is periodic. **Green:** `follows more closely in both displayed components`, `historical controlled endpoint`. **Amber:** `recovery`, only if immediately restricted to the downstream profiles. **Red:** `steady cloaking is established`, `mechanism`, `stability`, or importing the Section-8 spin ratio/metric.
+
+3. **Dominant Kármán preview — support: Rabault et al. (2019), Paris et al. (2021), Xia et al. (2024), Wang et al. (2024), Mao et al. (2015).** Observed: comparator fields become persuasive when the baseline/reference is decoded before scalar interpretation, while exact metrics remain distinct from morphology. Recommendation: let the target → controlled → physical-zero triplet carry the first reading; place the rounded 82% mean/phase preview only after the independent field estimands are named, and cite no external paper after that internal number. **Green:** `closer under both declared estimands`, `preserves the displayed alternating organization more closely`. **Amber:** `cloaking`, only as declared-reference shorthand. **Red:** `reward confirms the field`, `reproduces the target`, `identical wake`, or collapsing mean and phase into replicated evidence. The local caveat is that visual resemblance, reward and field L2 answer different questions.
+
+4. **Event-relative Vortex preview — support: Mao et al. (2015), Li & Zhang (2022), Wang et al. (2024), Cornejo Maceda et al. (2021).** Observed: representative snapshots are normally bounded by the selected condition or coordinate; time evolution requires a dedicated sequence or temporal diagnostic. Recommendation: identify the source-bound event offset in the caption and make the body inference morphological only. **Green:** `representative event-relative view`, `role-field morphology`. **Amber:** `response`, only when grammatically tied to the displayed offset rather than an evolution. **Red:** `suppression`, `tracking performance`, `event evolution`, or an arbitrary frame presented as typical without the source-bound selection rule. Bridge to the changed-reference preview, not to event efficacy.
+
+5. **Non-zero-target Illusion preview — support: Xia et al. (2024), Wang et al. (2024), Li & Zhang (2022), Rabault et al. (2019).** Observed: changed control objectives remain intelligible when target and baseline roles precede the controlled state and when the test condition is kept local. Recommendation: retain target → controlled → physical-zero order, name the 0.75-radius target and positive PPO scope, and attach the no-positive-Illusion-SR caveat to this paragraph. **Green:** `retargets toward the declared non-zero reference`, `positive PPO capability`. **Amber:** `Illusion`, only with the declared-reference meaning nearby. **Red:** `transfer`, `generalization`, `SR Illusion`, `target action`, or cross-task metric comparison.
+
+6. **Selection of Kármán for depth — support: Paris et al. (2021), Xia et al. (2024), Wang et al. (2024), Li & Zhang (2022), Cornejo Maceda et al. (2021).** Observed: close papers establish a reference case before extensions and let the next analysis answer a sharper physical question rather than merely announce another section. Recommendation: select Kármán because of the intersection of repeated learning records, registered fields and action-side evidence, not because it is visually strongest or supposedly representative of the other tasks. **Green:** `selected for detailed analysis because`, `uniquely combines the retained evidence needed to ask`. **Amber:** `representative`, only for the named periodic case, never for all tasks. **Red:** `canonical`, `general`, `transferable`, or a package inventory as the bridge. End with the tested-range and independent-field question already stated above.
+
+### Citation and caveat placement
+
+- External citations belong after statements about prior DRL/wake-control presentation or methodology; internal visual, metric and status statements point to project authorities, not literature.
+- The opening paragraph carries the cross-task non-equivalence ceiling; the Vortex and Illusion paragraphs repeat only their dangerous local caveats. Do not collect all repairs at the section end.
+- Captions decode configuration, role, variable, phase/event coordinate, source-bound representative selection, normalization and crop/scale. The body states the comparison, its bounded inference and why Kármán is selected.
+- If a caption must mention that a panel is representative, it must also name the selection coordinate or rule; `representative` alone is not a provenance statement.
+
+### Caption practice and anti-patterns
+
+Captions should decode configuration, role, variable, phase/event coordinate, normalization, crop/scale and the single local limit. They should not carry the whole selection argument. Avoid an equal-weight gallery, a common colour scale that suggests cross-configuration equivalence, comparator labels added only after viewing, exact Kármán result duplication, an arbitrary one-frame Vortex panel, tiny Illusion thumbnails, mechanism pre-announcement, or a bridge that merely names the next section.
+
+### Readiness
+
+`read`: current Round-4 Sections 1–8 and controls; Round-2 freeze/style; Round-3 Sections 2–8 and final audit; Confirmation Result chapter and named graphics; current DRL claims/results/data dictionary, manifests, plotting/evaluation code and retained tables; former Style-03 memo. `external full-PDF, checked 2026-08-11`: seven JFM papers listed above, each queried at paragraph level for comparator order, result/caption division, realization language, metric hierarchy, caveats, tense and transitions; this is a style read, not independent verification of DynamisLab evidence. `mapped`: symlink-backed role payloads and existing derived plot families through current manifests. `not-covered`: new CFD, training, artifact generation, publication typography rendering or a new event evaluator. External payload portability remains a production dependency, not a text blocker.
+
+## Author decision
+
+Author accepted the two-figure safe default for R5: S3-1 retains the steady profile figure and S3-2 is the Kármán-dominant wake composition with one source-bound Vortex and one 0.75-radius Illusion preview; detailed sequences/triplets remain with Section 7. This closes only the figure-composition author choice. Exact panel selection, source/provenance, crop/scale, caption, artwork, rendering and publication readiness remain gated.

docs/JFM_WYQ/ROUND4_TOP_DOWN/SECTION_04_KARMAN_RESULTS.md

@@ -0,0 +1,131 @@
+# Section 4 — Kármán-wake cloaking
+
+**Status:** `frozen` text contract; Round-5 eligible under recorded artifact gates. Field redraw and caption binding remain dependency-gated, and no figure has been generated or approved. This is an author-facing argument and production design, not final prose or panel approval.
+
+## Physical question and concise answer
+
+**Question.** Across the tested periodic Kármán conditions, how repeatably is high-performance control learned, and in the deeply documented reference wake does the sparse-signal objective coincide with a closer downstream field?
+
+**Answer.** The retained uniform-inflow/free-slip-wall series covers $Re_D=30,50,100,200$ and disturbance-radius ratios 0.75, 1.5 and 2.0. At $Re_D=50$, all five stochastic learning trajectories rise from low reward toward high-performing checkpoints, with best rewards 0.916–0.941, showing repeatable high-performance policy discovery across the five tested realizations. In the separate reference periodic Kármán cloak case in the parabolic-inflow/no-slip-wall channel configuration, controlled complete-cycle mean and eight-slot phase velocity errors are about 82% lower than physical zero, while sensor trajectories, DTW, spectrum and applied cylinder rotations show the accompanying periodic organization. Reward records learning under the online objective; registered spatial L2 supplies the independent field comparison.
+
+## Argument and figure cadence
+
+### 4.1 Learning across Kármán conditions
+
+#### 1. Begin with the tested range
+
+Lead with the tested-condition inference, then use S4-1 to map retained reward and six-sensor DTW similarity over the uniform-inflow/free-slip-wall conditions, using points rather than fitted trends. The Reynolds series is $Re_D=30,50,100,200$; the disturbance-radius series is 0.75, 1.0, 1.5 and 2.0, with the 1.0 reference supplied by the $Re_D=50$ case. Only the reference condition has five training realizations; every other condition is a single retained demonstration. The 0.75 and 2.0 disturbance cases used learning rate $10^{-4}$ rather than the common $3\times10^{-4}$, so their differences cannot be read as a pure geometry effect. These facts belong in labels and caption, without displacing the positive range result.
+
+#### 2. Let the five learning curves carry repeatability
+
+Repeated high-performance discovery across the five tested stochastic realizations is the comparator-led inference, not a claim based on one selected success. All five $Re_D=50$ reward trajectories rise from low initial values into the high-reward region and attain authoritative best rewards of 0.9266, 0.9302, 0.9160, 0.9222 and 0.9412 at iterations 471, 315, 360, 465 and 447. Interpret the complete 500-iteration histories and their high-performing checkpoints together; do not recast the result through late-window summaries. Figure production details belong to S4-1.
+
+#### 3. Separate the learning objective from field evidence once
+
+Reward combines force and sparse downstream-signal terms. The steadily improving curves show that high-performing policies were repeatedly found under that online objective, but they do not measure the registered velocity field. This distinction is stated here once and turns the argument to the reference wake, rather than becoming a repeated disclaimer.
+
+### 4.2 Fields and trajectories in the reference wake
+
+#### 4. Define the comparison and lead with the flow
+
+Ask first whether the frozen-PPO controlled field is closer to the declared target than physical zero under the independent field estimand. The target, controlled and physical-zero roles belong to the reference periodic Kármán cloak case in the parabolic-inflow/no-slip-wall channel configuration. The controlled field follows the target street while physical zero substantially distorts it. These roles are independent acquisitions within this physical configuration; the internal source label `karman_re100` remains provenance only, historical `U0` may mean centreline or bulk/mean speed, and no reader-facing $Re_D$ is assigned until source-specific conversion closes. Crop, scale and panel treatment belong to S4-2.
+
+#### 5. Quantify persistent and phase-resolved field proximity
+
+The complete-cycle mean velocity error is 0.085771 for controlled and 0.475206 for physical zero, a relative reduction of 0.819508. The separately constructed eight-slot phase diagnostic gives 0.112680 and 0.630878, a reduction of 0.821392. Display the two pairs separately so mean and phase are not mistaken for repeated measurements or uncertainty estimates. Both use the registered $20D\times10D$ downstream region; it is geometry-guarded because a solver-exact common-fluid mask was not persisted. The quantitative statement is therefore that the controlled field is closer to the target under both declared estimands in this region, not that every flow state is identical.
+
+#### 6. Connect sparse observations, frequency content and applied motion
+
+Ask whether the sparse observations and applied motion exhibit periodic organization consistent with the independent field comparison. Over samples 96–145 ($t_D=38.4$–$58.0$, about three target cycles), the centre-sensor trajectory, its frequency content and the three applied PPO rotations provide complementary signal/action evidence. Target has no action trajectory, and no zero-action surrogate is constructed. Target-normalized six-channel DTW similarity is 0.932899 for controlled and 0.625127 for physical zero; native DTW remains supplementary. The current sources do not license the old Confirmation claim that harmonics are essential under a same-protocol test. Production choices belong to S4-2.
+
+#### 7. Hand the achieved result to action-law compression
+
+The condition map and five learning trajectories show where high-performance policies were found and that discovery repeats across the five tested realizations at the reference uniform/free-slip condition. The separate deep comparison shows that, in the parabolic/no-slip reference wake, the learned trajectory is accompanied by much smaller registered field errors and periodic signal organization. What remains unresolved is not whether a high reward was reached, but whether the applied periodic action requires the full neural policy or can be represented by a compact executable law.
+
+## Why two figures are sufficient
+
+1. **S4-1 — Kármán conditions and repeated learning** (`FIGURE_PLANS/SECTION_04_FIGURE_01_KARMAN_CONDITIONS_AND_LEARNING.md`). It combines the retained Reynolds/disturbance map with all five learning trajectories. This is one scientific question: where was control learned, and how repeatably at the only five-realization condition?
+2. **S4-2 — Field matching and periodic control in the reference wake** (`FIGURE_PLANS/SECTION_04_FIGURE_02_REFERENCE_FIELD_SIGNAL_ACTION.md`). It combines role fields, mean/phase L2, one sensor portrait, a compact spectrum, DTW and applied PPO rotations. This is one deep-case question: what field and trajectory accompany the learned result?
+
+A third main figure is not recommended. If S4-1 cannot remain legible at 178 mm, move the disturbance-size mini-axis to the supplement before splitting the five learning curves. If S4-2 is too tall, move the spectrum first, then DTW as prose; retain the field triplet, L2, action traces and one portrait.
+
+## Result ownership and section interfaces
+
+Section 4 owns the retained periodic Kármán condition range, the positive five-realization learning result, the exact reference-case PPO mean/phase L2 values, complementary DTW, representative applied PPO rotations, one sensor portrait and the optional compact spectrum. Section 3 carries only the rounded visual preview. Section 5 owns the symbolic law, imposed reflection map, SR deployment and term deletion. Section 6 owns its different four-role mean-field errors, constant-reference accounting, residual and action-correlated correction-field modes. No SR or CCD panel enters Section 4, and the numerically close Section-4 and Section-6 errors are never described as agreement or replication.
+
+## Local inference limits
+
+- The five curves show repeated discovery in the tested realizations; they do not provide broad hyperparameter, Reynolds-number or geometry robustness, statistical uncertainty, asymptotic convergence or stability.
+- Non-reference condition points have $n=1$; the two outer disturbance-size cases also change learning rate.
+- Reward, DTW and spatial L2 are different quantities. DTW is a higher-is-better signal similarity, not spatial error or physical delay.
+- The phase diagnostic uses one nearest saved snapshot per independently phased slot, not a repeated-cycle ensemble.
+- The field ROI has a geometry guard but no persisted solver-exact common mask.
+- The representative action/sensor/spectrum window describes one controlled trajectory. It does not prove harmonic necessity, mechanism, optimality or a universal control law.
+
+## Compact source, production and reading appendix
+
+### Existing material and production decision
+
+- **Condition outcomes — redraw from tables; reuse plots only as design references.** `src/drl_pinball/train/results/latest/tables/latest_eval_summary.csv`, `latest_results_summary.json`, `latest/HOW_TO_READ.md`, and `train/plot_latest_results.py`. Existing `fig07_eval_cross_scene_metrics`, `fig12_vorticity_cross_re` and `fig13_vorticity_disturbance` designs are source views if present in a local generated package; the manuscript figure should redraw selected reward/DTW points and not copy the nine-axis dashboard.
+- **Five learning curves — reuse existing exported trajectories and regenerate a focused panel.** `train/results/training_csv/training_iterations.csv` plus its manifest, `train/results/figures/scratch_training_summary.csv`, `train/output/kar_re100_seed41–45/meta.json`, and the `plot_multiseed` logic in `plot_latest_results.py`/`plot_scratch_results.py`. No training is needed. Use the complete exported trajectories and authoritative best checkpoints; leave late-window diagnostic products outside the main narrative.
+- **Reference fields and L2 — crop/recombine or redraw from roles.** `data/reproduction/legacy/karman_re100/{target,controlled,zero}/{metadata.json,phase_fields.npz,timeseries.csv,dtw_summary.json}`; `eval/wake_l2.py`; current diagnostic `data/reproduction_plots/legacy/karman_re100.png`; `plot_reproduction_summary.py` and `plot_flow_fields.py`. Exact scalar values are ready; field production still needs source/hash/scale/crop audit.
+- **Actions and sensors — reuse exact CSV window, redraw.** `src/SR_analysis/results/runs/article2-timeseries-csv-20260720/karman_re100/{target_wide.csv,ppo_wide.csv,ppo_manifest.json}` and `article2-plotting-package-20260721/{README.md,phase_alignment.json}`. Use target and PPO sensor records and PPO applied action columns only. The package explicitly fixes indices 96–145 and records that target actions do not exist.
+- **Spectrum — recompute offline from the same declared window or crop an exact source only after window audit.** Confirmation `Graphs/freq_cloak_vortex.pdf` is narrative precedent, not numerical authority. The production script should use the same centre-sensor signal/window shown in the portrait; otherwise omit the spectrum.
+- **Confirmation figures — visual precedent only.** `Graphs/vort_cloak_vortex.pdf`, `error_cloak_vortex.pdf`, `sensor_cloak_vortex.pdf`, `freq_cloak_vortex.pdf`, `action_cloak_vortex.pdf`, `train_cloak.pdf`, and `re_cloak_vortex.pdf` show the prior result ordering and useful role layouts. Their old percentages, inherited Reynolds labels and `harmonics essential` narrative are not imported as current claims.
+
+### Literature identities and read states
+
+External papers establish presentation practice only. Metadata and DOI were checked in Undermind on 2026-08-11; all seven records below were read at `full-PDF` level with paragraph-specific questions, without numerical reproduction or phrase copying.
+
+1. Jean Rabault, Miroslav Kuchta, Atle Jensen, Ulysse Réglade & Nicolas Cerardi, “Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control,” *Journal of Fluid Mechanics* **865** (2019), 281–302, DOI `10.1017/jfm.2019.62`.
+2. Romain Paris, Samir Beneddine & Julien Dandois, “Robust flow control and optimal sensor placement using deep reinforcement learning,” *Journal of Fluid Mechanics* **913** (2021), A25, DOI `10.1017/jfm.2020.1170`.
+3. Chao Xia, Jiaqi Zhang, Eric C. Kerrigan & Georgios Rigas, “Active flow control for bluff body drag reduction using reinforcement learning with partial measurements,” *Journal of Fluid Mechanics* **981** (2024), A17, DOI `10.1017/jfm.2024.69`.
+4. Qiulei Wang, Dong Yan, Guodong Hu, Gang Chen, Jean Rabault & Bernd R. Noack, “Dynamic feature-based deep reinforcement learning for flow control of circular cylinder with sparse surface pressure sensing,” *Journal of Fluid Mechanics* **988** (2024), A4, DOI `10.1017/jfm.2024.333`.
+5. Xuerui Mao, Hugh M. Blackburn & Spencer J. Sherwin, “Nonlinear optimal suppression of vortex shedding from a circular cylinder,” *Journal of Fluid Mechanics* **775** (2015), 241–265, DOI `10.1017/jfm.2015.304`.
+6. Guy Y. Cornejo Maceda, Yiqing Li, François Lusseyran, Marek Morzyński & Bernd R. Noack, “Stabilization of the fluidic pinball with gradient-enriched machine learning control,” *Journal of Fluid Mechanics* **917** (2021), A42, DOI `10.1017/jfm.2021.301`.
+7. Ji-chao Li & Mengqi Zhang, “Reinforcement-learning-based control of confined cylinder wakes with stability analyses,” *Journal of Fluid Mechanics* **932** (2022), A44, DOI `10.1017/jfm.2021.1045`.
+
+### Observed practice versus Section-4 recommendation
+
+Across this sample, close JFM papers commonly establish a named baseline or reference before interpretation, bind parametric claims to the tested envelope, distinguish aggregated learning histories from representative deployments, and progress from a principal performance comparison to fields and then subordinate signal/action diagnostics. Captions carry conditions, encodings, windows and realization summaries; body text carries inference and physical consequence. The papers do not share a universal reward-first order, and several use metrics unlike this study’s field estimands. **Recommendation for this section:** retain the frozen sequence `tested range → five complete learning trajectories → reward/field separation → target/controlled/physical-zero fields → separate mean and phase L2 → subordinate DTW/sensor/action evidence → compact-law question`. Use present tense for displayed results and enduring comparisons; use past tense for completed training, acquisition and processing choices.
+
+### Paragraph-level style enrichment
+
+The support sets below give each paragraph job 3–5 close papers, with deliberate reuse. They support rhetoric only. Round-5 citations should follow external context, not DynamisLab values or realization claims.
+
+1. **Tested-condition range — support: Paris et al. (2021), Wang et al. (2024), Li & Zhang (2022), Cornejo Maceda et al. (2021).** Observed: parametric results are introduced with explicit tested envelopes and local changes in protocol; points or cases do not automatically imply a fitted law. Recommendation: state the Reynolds and disturbance-radius sets before interpretation, then put `n=1` and the changed learning rate beside the affected points/caption. **Green:** `across the tested conditions`, `retained demonstrations`, `the points show`. **Amber:** `range`, only with exact endpoints and sampling. **Red:** `trend`, `scaling`, `robust across Reynolds number/geometry`, or treating the outer-radius difference as geometric when learning rate also changes. The boundary sentence should not eclipse the positive fact that high-performing retained cases exist across the declared set.
+
+2. **Five-realization discovery — support: Rabault et al. (2019), Paris et al. (2021), Wang et al. (2024), Xia et al. (2024).** Observed: papers that support repeatability expose the realization count and either all histories or an explicit aggregate; representative snapshots are identified separately. Recommendation: show all five complete trajectories and name them `five tested stochastic realizations`; do not convert them into a mean band or use one best seed as the result. **Green:** `repeated high-performance discovery across the five tested realizations`, `all five trajectories attain`. **Amber:** `repeatable`, only with `within these five realizations` local or immediately preceding. **Red:** `robust`, `statistically significant`, `converged`, `reproducible`, or any uncertainty claim. Exact best rewards/checkpoints may remain in body because they define the range and prevent selected-success narration; iteration-by-iteration detail stays in caption/data.
+
+3. **Online reward versus independent field question — support: Rabault et al. (2019), Paris et al. (2021), Xia et al. (2024), Li & Zhang (2022).** Observed: reward or force objectives establish optimization success, while flow fields and stability/physical diagnostics answer a distinct question. Recommendation: state the separation once, immediately after the learning curves, and use it as the hinge into the reference field comparison. **Green:** `high-performing under the online objective`, `does not measure the registered velocity field`, `independent field comparison`. **Amber:** `performance`, only with the metric named. **Red:** `reward validates field matching`, `physical accuracy from reward`, or repeating the disclaimer in every later paragraph. No external citation should trail the internal reward values.
+
+4. **Exact target–controlled–physical-zero field comparison — support: Mao et al. (2015), Xia et al. (2024), Rabault et al. (2019), Li & Zhang (2022), Cornejo Maceda et al. (2021).** Observed: controlled fields are interpreted only after the baseline/reference roles and displayed variable are defined; body prose states the physical difference while captions decode conditions and snapshots. Recommendation: ask the field question before giving scalar errors and preserve target → controlled → physical-zero order. **Green:** `closer to the declared target than physical zero`, `follows the displayed street more closely`. **Amber:** `matches` or `cloaks`, only with `under the declared field estimand` nearby. **Red:** `reproduces`, `identical`, `validation`, or a reader-facing inherited Reynolds label before the source-specific conversion gate closes. Put independent-acquisition and Reynolds-label boundaries in this paragraph/caption, not in a remote provenance note.
+
+5. **Separate mean and phase L2 — support: Mao et al. (2015), Li & Zhang (2022), Cornejo Maceda et al. (2021), Paris et al. (2021).** Observed: temporal/global and spatial or stability measures are shown as different representations, with exact values promoted only when they change the inference. Recommendation: present complete-cycle mean and eight-slot phase L2 as two named estimands in separate rows or clauses; never graph them as replicate points or an uncertainty pair. **Green:** `under both declared estimands`, `mean error`, `phase error`, `relative reduction`. **Amber:** `consistent`, only in the weak sense that both comparisons have the same ordering, and preferably avoid it. **Red:** `replicated`, `confirmed`, `agreement`, `uncertainty`, or `about 82%` without naming both estimands. Attach the geometry-guard/no-common-mask caveat directly to the ROI claim; attach the nearest-snapshot caveat directly to phase construction.
+
+6. **Subordinate DTW, sensor, spectrum and action evidence — support: Rabault et al. (2019), Paris et al. (2021), Xia et al. (2024), Wang et al. (2024).** Observed: signal and actuation diagnostics follow the principal result and explain the realized trajectory; captions carry windows, channels, normalization and line encodings. Recommendation: lead this paragraph with the question of accompanying periodic organization, keep target-normalized DTW explicitly complementary, and let one centre-sensor portrait plus PPO actions carry the body inference. **Green:** `complementary signal/action evidence`, `periodic organization`, `applied PPO rotations`. **Amber:** `aligns` or `tracks`, only with the signal and metric named. **Red:** `field validation`, `mechanism`, `harmonic necessity`, `target action`, or a zero-action surrogate. Move the spectrum to supplement if its window is not exactly shared with the portrait; DTW is the next item to demote if figure height fails.
+
+7. **Compact boundary and law bridge — support: Cornejo Maceda et al. (2021), Mao et al. (2015), Li & Zhang (2022), Xia et al. (2024).** Observed: strong transitions convert an achieved result into the unresolved question that motivates a deeper control or mechanism analysis. Recommendation: summarize the two evidence chains without merging their configurations, then ask whether the applied periodic action needs the full neural policy or admits a compact executable representation. **Green:** `in the separate reference wake`, `accompanied by`, `can the applied action be represented`. **Amber:** `strategy`, only for the observed policy behavior, not a mechanism. **Red:** `therefore the law is`, `same case across configurations`, `mechanism established`, or a bridge that merely announces Section 5.
+
+### Citation, caption and caveat placement
+
+- Cite external literature only for prior methods or genuinely comparative context. Internal ranges, seed counts, rewards, L2, DTW, windows and role identities bind to project authorities and should not borrow credibility from an adjacent literature citation.
+- S4-1 caption carries exact tested points, `n=5` only at the reference uniform/free-slip condition, `n=1` elsewhere, and the two changed-learning-rate points. The body carries the bounded range inference and the five-realization result.
+- S4-2 caption carries configuration, target/controlled/physical-zero roles, displayed variable/phase, ROI/mask guard, mean/phase construction, sensor/action window and DTW orientation. The body carries field ordering, metric hierarchy, local limits and the law question.
+- Place each dangerous caveat locally: protocol confounding with the outer-radius points; realization ceiling with the five curves; reward/field separation at the hinge; mask and phase construction with L2; absent target actions and no harmonic-necessity claim with signal/action evidence.
+
+### Caption practice and anti-patterns
+
+Captions carry configuration, condition, realization count, roles, variable, phase/window, normalization, crop/mask and metric orientation. Body text states why the comparison matters. Avoid a training diary, mean-only five-run band, late-retention graphics, reward-first self-validation, dual-axis reward/DTW plots, fitted Reynolds laws, unannounced comparator changes, eight-phase montages, target actions, action spectra used to assert harmonic necessity, SR/CCD imports, or package/path language in the reader-facing argument.
+
+### Readiness
+
+`read`: current Round-4 Sections 1–8 and controls; Round-2 freeze/style; Round-3 Sections 2–8 and final audit; Confirmation Result chapter and named graphics; current training/evaluation tables, five-run CSV/metadata, role manifest, plotting scripts, wake-L2 evaluator, SR PPO time-series package and former Style-04 memo. `external full-PDF, checked 2026-08-11`: seven JFM papers listed above, each queried at paragraph level for comparator order, tested-range and realization language, reward/field separation, metric hierarchy, result/caption division, caveats, tense and transitions; this is a style read, not independent verification of DynamisLab evidence. `mapped`: symlink-backed raw field roles through their metadata/manifests and locally generated figure families that may not exist in every checkout. `not-covered`: new CFD/training, new artifact generation, or publication-size rendering. Nonportable field payloads remain a production dependency; scalar and section architecture are not blocked.
+
+## Coordinator production defaults and recommendations
+
+1. Retain both Reynolds and disturbance-size mini-axes in S4-1 if readable; otherwise move disturbance size to the supplement. Keep the five curves.
+2. Use phase vorticity for the S4-2 field triplet. Add mean velocity-error maps only if a final redraw shows distinct argumentative value; default omit.
+3. Keep one centre-sensor portrait. Retain the compact spectrum only if it uses the exact displayed signal/window; otherwise move it to the supplement.
+
+These are coordinator defaults/recommendations, not unresolved scientific author choices. Final reconciliation froze the text contract; production remains under the recorded artifact gates.

docs/JFM_WYQ/ROUND4_TOP_DOWN/SECTION_05_KARMAN_SR_LAW.md

@@ -0,0 +1,200 @@
+# Round 4 integrated architecture — A compact control law for Kármán cloaking
+
+## 1. Status and scope
+
+**Status:** `frozen` text contract; Round-5 eligible under recorded artifact gates after the final narrow Figure-Plan reconciliation; **not final manuscript prose**, a new scientific audit, or panel/artifact approval. No figure has been generated or approved. **Body-text share:** approximately 17%, implemented as five compact paragraphs.
+
+**Scope.** Section 5 compresses the Kármán-cloaking PPO action into one executable symbolic surrogate, discloses its imposed reflection-equivariant deployment structure, lets closed-loop execution arbitrate it, and ranks tested parent-relative term deletions. It preserves the current parabolic-inflow/no-slip-wall evidence contract and does not reopen research, infer mechanism, or promote diagnostic drafts to publication readiness.
+
+**Reader-facing title:** **A compact control law for Kármán cloaking**.
+
+## 2. Reader question and bounded answer
+
+**Question.** Can the learned Kármán-cloaking action be compressed into a compact executable law, and what action-side structure survives closed-loop and deletion tests?
+
+**Bounded answer.** In the reference periodic Kármán cloak case in the parabolic-inflow/no-slip-wall channel configuration (internal source label `karman_re100` in provenance only), an imposed reflection-equivariant symbolic surrogate executes in closed loop. The law is organized by dominant persistent rear counter-rotation, smaller rear-lift feedback and a weak tested front correction. Parent-relative deletion under one fixed 40-control-step native-DTW similarity contract ranks those tested contributions; it does not establish necessity, uniqueness, mechanism, robustness, universality or positive Illusion transfer.
+
+## 3. Accepted author feedback implemented
+
+- Adopt the concise reader-facing title above; retain the long scientific question only as an internal planning object.
+- Keep recovery/deployment logic to one compact paragraph. Search stages, causal-pairing details, topology selection, refits, gates, hashes and implementation diagnostics move to the supplement.
+- Explain $G$ as an **imposed design constraint** for the symmetric pinball/cloaking objective. It organizes the surrogate by tying mirrored observations to mirrored actions and reducing separate action heads. The PPO policy was neither trained nor constrained to obey this equivariance; $G$ neither proves PPO symmetry nor establishes plant symmetry.
+- Use no repeated geometry/action schematic and no decorative formula decomposition. Section 2 owns geometry and action signs; $G$ and the law are equation-plus-prose objects.
+- Use the main display for closed-loop SR behavior/direct supported flow result and parent-relative deletion. Same-protocol PPO/SR/Zero is used only where exact standardized roles are bound; no comparison is invented.
+
+## 4. Five-paragraph architecture
+
+1. **Question and estimand.** Move from Section 4's independent PPO field matching to action compression: SR is an executable observation-to-action surrogate under the same physical configuration, not a flow-field model or causal mechanism. Offline action fit is diagnostic; closed-loop behavior is the arbiter.
+2. **Recovery logic and imposed reflection map.** State only that causally aligned PPO trajectories were used to discover/refit a compact law and then deploy it in CFD. Define reflection across the channel centreline: upper/lower sensor locations exchange; transverse velocity and lift change sign; drag is even; upper/lower cylinder forces exchange with the corresponding parity; actions transform as $G_\alpha(\alpha_F,\alpha_U,\alpha_L)=(-\alpha_F,-\alpha_L,-\alpha_U)$. This is useful because the objective and pinball arrangement admit a natural mirrored organization: front action is forced odd and the two rear heads share one generator, reducing and structuring the surrogate. State that equivariance was imposed on SR after PPO data collection, not learned or enforced in PPO training, and does not prove exact plant symmetry.
+3. **Canonical executable law and physical translation.** Display the canonical law once, define $C_{d,\mathrm{rear},a}=(C_{d,U}-C_{d,L})/2$ and $C_{l,\mathrm{rear},s}=(C_{l,U}+C_{l,L})/2$, and translate the historical SR output $\alpha=\omega/U_0$: opposite rear-cylinder rotation has a persistent magnitude set by $3.431209$, modulated by rear-mean lift; the front receives a smaller drag-asymmetry correction. The executable authority states the historical output exactly as $\alpha=\omega/U_0$ and Stage 3 applies $\omega=\alpha U_0$ as the solver angular-velocity command. Therefore $\alpha$ has inverse-length units in dimensional notation (inverse-lattice-length in the historical lattice convention), despite stale metadata calling it dimensionless. The general physical surface-speed command is $s=\omega R/U_{ref}=\alpha R(U_0/U_{ref})$. It reduces to $s=\alpha R$ only if source authority establishes $U_0=U_{ref}$; the historical parabolic source is not retroactively assigned that equality. Trace labels use historical $\alpha=\omega/U_0$ and state the lattice/nondimensional length convention; they do not call $\alpha$ dimensionless or use it interchangeably with $s$. Avoid software-stage narration and governing-law/mechanism wording.
+4. **Closed-loop deployment and direct result.** Lead with formula-parent SR deployment and the standardized reference-case SR visual/field result. The standardized role has 480 warm-up control intervals and 160 retained boundaries, eight complete SR cycles, native DTW mean $0.942357$, $E_{mean}=0.125404$ versus physical-zero $0.475206$ ($\eta_{mean}=0.736106$), and $E_{phase8}=0.156040$ versus $0.630878$ ($\eta_{phase8}=0.752662$). Treat this as SR-versus-target with a physical-zero comparator under the standardized SR acquisition; do not merge it with Section 4's separately acquired PPO estimand. The existing four-role Target/PPO/SR/Zero image is a derived diagnostic, not publication-ready.
+5. **Deletion, ceiling and bridge.** Under the fixed 40-step parent-relative native-DTW contract across internal `Re_code` 50/100/200/400 (provenance labels only in this parabolic-source campaign), deleting the rear constant causes the largest listed degradation; deleting rear-lift feedback is smaller but non-zero; deleting the tested front term is weak/mixed. Show exact deltas in the main display or compact caption. Call these term deletions, not causal ablations. End by asking how persistent and dynamic action components relate to the separately defined field estimands in Section 6.
+
+## 5. Detailed content and style controls
+
+### Equation introduction, notation and physical translation
+
+- Import physical variables, action order, sign, normalization, objective and comparator from Section 2 before displaying the law. Define the rear drag/lift aggregates directly beside the law so the reader need not reconstruct subscripts.
+- Keep the executable map, parent PPO policy, physical angular velocity and normalized surrogate action distinct. Present tense defines symbols and equation meaning; past tense reports recovery, deployment and measured outcomes.
+- Display the exact law once, then translate only its lower-dimensional action organization: persistent rear counter-rotation, smaller rear-lift modulation and weak front drag-asymmetry correction. Do not narrate every coefficient or call the expression a governing physical law.
+- Keep recovery history subordinate. Causal pairing, libraries, candidate genealogy, refits, package paths, hashes and gate bookkeeping belong in the supplement unless a failed candidate is needed to explain the selected topology.
+
+### Imposed symmetry and topology lesson
+
+- State the reflection map before interpreting the canonical law. The odd front head and shared mirrored rear generator are imposed deployment choices that reduce and organize the surrogate; they are not properties discovered in PPO.
+- Topological compactness supports readability and model selection only. It does not establish exact plant symmetry, PPO equivariance, physical truth, necessity or causality.
+- The useful JFM pattern is to disclose restricted actuation spaces, sparse libraries or equality constraints as choices, then distinguish what was imposed from what was identified within that structure. Here no new unconstrained-versus-constrained result is invented.
+
+### Closed-loop arbitration and deletion language
+
+- Closed-loop execution under a named plant, role set, metric and window is the primary arbiter. Offline action fit, symbolic complexity, convergence and static/schema gates are subordinate diagnostics answering different questions.
+- Do not assemble a synthetic PPO/SR/Zero comparison from labels alone. Formula-parent deployments and standardized Target/SR/physical-zero acquisition remain different evidence contracts unless exact common-target reconciliation is completed.
+- Use **parent-relative term deletion** or **fixed-window deletion test**, not unqualified `ablation`. State parent, deleted/replaced term, held-fixed case, metric, 40-step window and tested set.
+- A deletion launches a new closed-loop trajectory and changes its visited state distribution; it is not a local coefficient derivative. Rank only the tested variants and do not infer causal contribution, necessity, sufficiency, unique decomposition, mechanism or universal sensitivity.
+
+### Equation/figure cadence and density
+
+Use the compact sequence `imposed deployment map → canonical executable law → matched closed-loop evidence → parent-relative deletion → limitation/bridge`. Keep each equation close to the evidence that depends on it. One display carries one principal inferential job; implementation detail goes in captions or supplement. A summary appears only after the law hierarchy is established, not as an opening gallery. No law graph or symmetry schematic is needed because the equations resolve order and sign; no field panel is added merely to strengthen symbolic interpretation.
+
+## 6. Display and equation plan
+
+### Main text
+
+1. **Reflection/deployment equation (required; no panel).**
+
+   $$G_\alpha(\alpha_F,\alpha_U,\alpha_L)=(-\alpha_F,-\alpha_L,-\alpha_U),\qquad
+   \alpha_F(x)=\tfrac12[h_F(x)-h_F(Gx)],\quad
+   \alpha_U(x)=h_R(x),\quad \alpha_L(x)=-h_R(Gx).$$
+
+   Prose gives the observation parity and the imposed-not-learned boundary.
+2. **Canonical law (required; no panel).**
+
+   $$\alpha_F=\operatorname{odd}(-0.381391 C_{d,\mathrm{rear},a}),\qquad
+   \alpha_U=1.307782 C_{l,\mathrm{rear},s}-3.431209,\qquad
+   \alpha_L=-\alpha_U(Gx).$$
+
+3. **One Section-5 main figure (dependency-gated production), specified in `docs/JFM_WYQ/FIGURE_PLANS/SECTION_05_KARMAN_SR_CONTROL_LAW.md`:** (a) standardized reference-case closed-loop SR behavior/action traces; (b) direct target/SR/physical-zero vorticity result at phase slot 0 with scalar L2 annotations; (c) optional compact across-case formula-parent deployment summary; (d) fixed-window parent-relative term deletion. If space is tight, omit (c) first. No geometry panel, $G$ schematic or formula decomposition.
+
+### Supplement
+
+Recovery/search library, causal pairing and split details; topology/refit history; 40/200/400-step gate bookkeeping and hashes; rejected candidates; offline RMSE; full re-code and unseen-condition matrices; additional phase slots; full diagnostic sensor portraits; complete deletion table and provenance. Historical/negative Illusion SR stays supplementary or in its owning cross-case boundary, never as a positive Section-5 result.
+
+## 7. Exact data and readiness summary
+
+- **Existing standardized raw/source role data:** `src/drl_pinball/data/reproduction/legacy/karman_re100/{target,controlled,sr,zero}/` resolves to machine-local external storage. Each role binds `metadata.json`, `timeseries.csv`, `phase_fields.npz`; periodic arrays contain eight `(ny,nx)` `ux`,`uy` snapshots and complete-cycle means. SR metadata binds the canonical formula pair, 480 warm-up intervals, 160 retained boundaries, eight accepted complete cycles and parabolic/no-slip plant provenance. Status: `available-data`, but nonportable/symlink-backed and still requires production extraction/audit.
+- **Existing standardized SR summary:** `src/drl_pinball/data/reproduction/sr_wake_l2_summary.{json,csv}`. Only the standardized reference Kármán SR row is present (plus negative Illusion 1L); it supplies exact mean/phase8 SR-versus-target and zero-versus-target values. Status: `available-data`; geometry-guarded ROI, no persisted solver-exact mask.
+- **Existing draft images:** `src/drl_pinball/data/reproduction_plots_sr/legacy/karman_re100.png` and `.../variants/karman_re100_k_{front0,rear0,rear1}.png`. They show slot-0 vorticity plus sensor phase portraits. Status: `draft-only`; diagnostic labels, full-domain framing and layout are not JFM-ready.
+- **Existing processed/render manifest:** `src/drl_pinball/data/reproduction_plots_sr/manifest.json` binds source paths, hashes, slot 0 and fixed vorticity range $[-10^{-3},10^{-3}]$. It reports 11/23 plots with `allow_partial: true`: four training Kármán four-role plots, four Kármán Target/SR/Zero unseen-condition plots, and three reference-case deletion plots. Status: `partial-derived-package`, not completeness or panel readiness.
+- **Rendering source:** `src/drl_pinball/plot_reproduction_summary.py` validates role files, computes $\omega_z=\partial_xu_y-\partial_yu_x$ from slot 0 and renders vorticity/sensor portraits. Status: `available-script`, diagnostic rather than publication renderer.
+- **Deletion numbers/source:** `src/SR_analysis/results/runs/article2-plotting-package-20260721/tables/term_deletion.{csv,md}` binds canonical parent and variants over the same 40-step native-DTW windows. Status: `available-data`; standardized variant images are longer-window field acquisitions and are not the source of the 40-step deletion deltas.
+- **Formula-parent deployment:** `src/SR_analysis/results/runs/article-L3-karman-20260718/`, `article2-long-karman-20260720/` and current claims/results authorities bind 200/400-step SR deployments. Status: `available-data`; not automatically a same-protocol PPO/SR/Zero comparison.
+- **Same-protocol PPO/SR/Zero in the reference case:** role data and one four-role diagnostic image exist, but PPO and SR were independently acquired and have distinct run-generated target-array hashes. A publication comparison requires explicit target-role reconciliation and one common metric extraction. Status: `dependency-gated`; fallback is formula-parent SR deployment plus the directly supported standardized SR/target/zero field result.
+
+## 8. Dated paragraph-level literature/style enrichment (2026-08-11)
+
+Literature establishes sampled presentation practice only: it does not prove an internal SR result, causality, validity, priority or artifact readiness. DOI/PDF states were checked in the DynamisLab Undermind workspace on 2026-08-11. All records below are `full-PDF` targeted reads of the cited sections, equations, figures and captions, not equation-by-equation verification or independent reproduction.
+
+### Exact papers, DOIs and read states
+
+1. Gautier et al. (2015), “Closed-loop separation control using machine learning”, *JFM* **770**, 442–457, DOI `10.1017/jfm.2015.95`, `full-PDF`: sensor/action definitions §§2.3–2.4; objective §3.2; executed-law reading §4.2; comparator §4.3; robustness §4.4; non-optimality §5.
+2. Cornejo Maceda et al. (2021), “Stabilization of the fluidic pinball with gradient-enriched machine learning control”, *JFM* **917**, A42, DOI `10.1017/jfm.2021.301`, `full-PDF`: nested control spaces §4; exact law (4.3); controlled snapshots/mean figure 15; averaged-control intervention §5.1/figure 16; appendices for algorithm and forcing comparisons.
+3. Loiseau & Brunton (2018), “Constrained sparse Galerkin regression”, *JFM* **838**, 42–67, DOI `10.1017/jfm.2017.823`, `full-PDF`: library §2.1; imposed constraints §§2.2–3; results §5; coefficient tables Appendix A; sparsity/accuracy selection Appendix B.
+4. Loiseau, Noack & Brunton (2018), “Sparse reduced-order modelling: sensor-based dynamics to full-state estimation”, *JFM* **844**, 459–490, DOI `10.1017/jfm.2018.147`, `full-PDF`: sensor/features §3.1; library/equations §§3–4; reconstruction §3.3; modes/interpretation §4; selection Appendix A.
+5. Rabault et al. (2019), “Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control”, *JFM* **865**, 281–302, DOI `10.1017/jfm.2019.62`, `full-PDF`: geometry/observations/actions §2.1; smoothing/reward §2.2; controlled-flow results §3; implementation/sensitivity Appendices B–E.
+
+### Paragraph jobs: close examples, observed practice and recommendation
+
+- **P1 — surrogate boundary** (Gautier; Loiseau & Brunton; Loiseau et al.; Rabault): sampled papers order `physical input/output and objective → admissible learned object → validation task` and distinguish searched laws, sparse plant models and black-box policies. **Recommend:** define SR as an executable observation-to-action surrogate; contrast offline parent-visited action fit with surrogate-visited closed-loop execution. Put method citations after this distinction, never after an internal number as apparent verification.
+- **P2 — recovery/deployment/symmetry disclosure** (Loiseau & Brunton; Loiseau et al.; Cornejo Maceda; Rabault): libraries, equality constraints, control spaces, smoothing and sensors are disclosed before interpretation; coefficient genealogy and optimizer checks can move to appendices. **Recommend:** state causal alignment, post-PPO imposition of $G$, parity and the no-PPO/no-plant-symmetry ceiling together; move libraries, refits, hashes and gate chronology to the supplement.
+- **P3 — exact law and physical translation** (Gautier; Cornejo Maceda; Loiseau & Brunton; Loiseau et al.): symbols/search space precede equations; interpretation shifts from coefficients to actuation organization, mean forcing or phase dependence. **Recommend:** put aggregates and historical $\alpha=\omega/U_0$ beside the law; translate only persistent rear counter-rotation, rear-lift modulation and smaller front correction. Keep the dimensional caveat beside $\alpha$ and $s$.
+- **P4 — closed-loop arbitration** (Gautier; Cornejo Maceda; Rabault; Loiseau et al.): executed trajectories, fields or reconstruction error arbitrate under named comparators; admissibility, compactness and training diagnostics remain distinct. **Recommend:** formula-parent deployment first, then standardized Target/SR/physical-zero under its own contract; exclude independently acquired PPO evidence until target reconciliation.
+- **P5 — deletion ranking, ceiling and field bridge** (Cornejo Maceda; Loiseau & Brunton; Gautier; Rabault): interventions name what was replaced/suppressed and report executed performance; sparse threshold paths are model selection, not plant causality. **Recommend:** name parent, deleted term, 40-step window, native-DTW similarity direction and tested set before ranking; put the changed-visited-state/noncausal ceiling in that sentence. Bridge by asking how this action organization relates to Section 6's independently defined field estimands.
+
+**Sample-bounded synthesis.** Across these five close examples, definitions precede learned equations, imposed restrictions precede interpretation, execution follows the law, implementation detail can move to appendices, captions decode conditions/traces, and strong claims are locally bounded. Cornejo Maceda et al.'s averaged-control replacement is a close rhetorical example for naming an intervention but is not equivalent to the present term deletions. Recommended cadence: `surrogate boundary → imposed map → exact law/units → closed-loop arbitration → parent-relative deletion → field question`; this is not a universal JFM rule.
+
+### Green/amber/red lexicon
+
+- **Green:** `executable surrogate`, `imposed reflection-equivariant deployment`, `closed-loop execution`, `parent-relative term deletion`, `largest degradation among tested variants`, `persistent rear counter-rotation`, `smaller modulation`, `weak/mixed tested correction`, `compatible with`.
+- **Amber—only with named test and adjacent limit:** `compact law`, `interpretable`, `feedback contribution`, `symmetry`, `robust`, `generalizes`, `ablation`. Prefer `organized/readable`, `imposed map`, `tested deployment envelope`, `term deletion`.
+- **Red:** `governing law`, `discovered PPO symmetry`, `plant symmetry`, `causal ablation`, `necessary`, `sufficient`, `unique`, `mechanism`, `optimal`, `universal`, `validated by literature`, or a common PPO/SR acquisition before reconciliation.
+
+### Citation/caveat, caption/body, bridges and anti-patterns
+
+Cite Gautier/Cornejo Maceda for explicit learned-law positioning; Loiseau & Brunton/Loiseau et al. for imposed structure versus identified terms/model selection; Rabault for policy implementation versus controlled-flow interpretation. Never cite literature as verification of a project coefficient, error, DTW value or deletion delta. Place imposed-not-learned in P2, units in P3, acquisition separation in P4 and deletion/noncausal ceiling in P5.
+
+**Caption:** configuration, roles, trace definition, window, metric direction, parent/variant and panel decoding. **Body:** what execution arbitrates, bounded physical organization, tested ranking, limit and consequence. End with a live field-estimand question, not a section announcement. Avoid a literature catalogue in results, coefficient count as physical truth, transfer of “need for feedback” to this non-equivalent deletion, optimizer chronology before the executable object, caption-level causality, or citations that appear to certify internal evidence.
+
+### Coverage
+
+- `read/full-PDF`: five papers above at the named locations; no independent reproduction.
+- `workspace-inspected`: existing folders/files and completed search `Symbolic regression for ML flow control`; existing library sufficed, so no new semantic/citation expansion or deep search was launched.
+- `recall/read`: narrow Nowledge recall and current frozen packet; memory was navigation only, packet remained controlling.
+- `not-covered`: exhaustive sentence coding, new science, CFD/training, artifacts, rendering or figure approval.
+
+## 9. Ownership, interfaces and near-collision controls
+
+**Imports:** Section 2 owns physical configuration, $Re_D$, body/action/sign and metric definitions. Section 4 owns PPO target/controlled/physical-zero L2 and DTW performance; its values are a different acquisition/estimand and are not repeated as SR evidence.
+
+**Owns:** imposed reflection map; canonical executable Kármán/cloak law; concise recovery/deployment logic; formula-parent deployment; standardized SR result where exact roles are available; fixed-window parent-relative deletion ranking.
+
+**Exports:** to Section 6, the bounded action-side organization—persistent rear counter-rotation, smaller rear-lift feedback, weak tested front correction—and the question of how it relates to independently defined field estimands. To later steady synthesis, only contextual recurrence/magnitude, not configuration equivalence or mechanism.
+
+### Near-collision controls
+
+- **Offline parent-action fit vs SR execution:** different estimands (fixed parent-visited states/action error versus surrogate-generated trajectory/plant outcome); fit is not control success.
+- **Static or short gate vs standard closed-loop comparison:** related diagnostic, not full-result validation.
+- **Formula-parent deployment vs standardized role acquisition:** different estimands unless exact sameness is authority-bound; no synthetic panel or arithmetic.
+- **Parent-relative deletion vs causal ablation:** different estimands; use tested-deletion ranking only.
+- **Persistent SR action vs steady-chain spin:** related contextual diagnostics under different chain, solver/geometry, role, metric and temporal contract; no agreement or validation.
+- **SR action organization vs Section-6 field/CCD organization:** different estimands; no independent triangulation, causality or identification of action terms with field modes.
+
+## 10. Exclusions, anti-patterns and non-claims
+
+No repeated geometry/action schematic; decorative equation panel; long software pipeline; training diary; positive Illusion SR; cross-configuration transfer; PPO equivariance; proof of plant symmetry; DTW as field equivalence or physical delay; deletion as causality; necessity, sufficiency, uniqueness, mechanism, stability, robustness, optimality, universality or continuous Reynolds law. Reader-facing prose does not use `Legacy`/`V5`; those IDs remain only in provenance.
+
+### Relevant anti-patterns
+
+- Opening with the formula before defining input, output, normalization, action order and objective.
+- Calling imposed oddness, shared topology or reflection a discovered PPO symmetry.
+- Narrating every coefficient or treating sparsity as self-interpreting physical truth.
+- Presenting offline RMSE, Pareto rank, convergence or a short gate as closed-loop evidence.
+- Combining different acquisitions because Target/PPO/SR/Zero labels appear compatible.
+- Using `ablation` without naming whether control, sensors, a model class or one term changed.
+- Inferring causality, necessity or mechanism from deletion, or treating DTW, spatial L2 and gate checks as interchangeable validation.
+- Importing CCD modes, field panels or steady results to manufacture independent confirmation.
+- Building a figure gallery or repeating Section-2 equations, package stages, hashes and artifact inventories in narrative.
+- Borrowing `robust`, `stable`, `optimal` or `governing law` from external precedents without dedicated internal tests.
+
+
+## 11. Author choices and execution dependencies
+
+1. **Panel (c):** retain a compact 200/400-step formula-parent across-$Re_D$ summary only if it helps establish deployment envelope; otherwise omit to keep the reference-case physics central.
+2. **Four-role comparison:** include PPO in the final field panel only after target-role reconciliation and common metric extraction. Safe default: Target/SR/physical-zero field triptych, with PPO performance remaining owned by Section 4.
+3. **Action trace:** production must choose effective smoothed angular velocity or commanded target angular velocity; do not mix them. Recommended: effective smoothed historical $\alpha_i=\omega_i/U_0$ trace, labelled in inverse-lattice-length units, together with the general conversion $s_i=\alpha_iR(U_0/U_{ref})$; use $s_i=\alpha_iR$ only after source authority establishes $U_0=U_{ref}$.
+
+## 12. Coverage ledger
+
+- `read`: targeted Nowledge memory recall for R2/R3/R4 history; Round-4 global ledger/display inventory; R2 freeze/style; R3 Sections 4–6 interfaces; current SR claims/results, $G$ implementation, deletion table; current `drl_pinball` metadata, SR L2 summary, plot manifest, draft images and renderer.
+- `mapped`: formula-parent run directories and external raw role payloads through manifests/metadata; mapping is not independent artifact verification.
+- `not-covered`: unrelated sections, new literature, CFD/training, new rendering, exhaustive binary/hash recomputation.
+- `external-inaccessible/nonportable`: reproduction payload resolves to `/home/frank14f/optane/...`; repository metadata/manifests bind it, but publication production must verify access and hashes.
+- **Coordinator audit issue:** standardized PPO and SR role target arrays are independently generated and have different hashes. Do not call the four-role draft a common-target quantitative comparison until reconciled. Keep standardized SR L2 (`0.125404`) distinct from Section 4 PPO L2 (`0.085771`) and Section 6 corrected four-role mean error (`0.085779`): these are non-equivalent estimands.
+
+## 13. Quality gate and disposition
+
+- [x] One integrated Section-5 architecture contains paragraph, style, display, literature, ownership and coverage controls; it is not manuscript prose.
+- [x] All five paragraph jobs and all current scientific facts, values, readiness states and accepted author decisions are retained.
+- [x] Exact four-paper JFM identities, DOIs and `full-PDF` read levels are retained; observed practice is separated from project recommendation.
+- [x] Imposed symmetry/topology is disclosed before interpretation and cannot be read as PPO discovery or plant-symmetry proof.
+- [x] Closed-loop execution remains the arbiter; offline fit and gates remain subordinate diagnostics.
+- [x] Deletion wording names the parent-relative fixed-window intervention and rejects causal/necessary/unique interpretation.
+- [x] Equation/figure cadence, caption ceilings, fallbacks and the canonical `FIGURE_PLANS` dependency are explicit.
+- [x] Ownership, bridges, near-collisions, anti-patterns, exclusions and fixed Kármán/cloak-only claim ceiling are explicit.
+- [x] Coverage states preserve `read`, `mapped`, `not-covered` and `external-inaccessible/nonportable` distinctions.
+- [x] No new research, number, result, mechanism, evidence, comparison or artifact readiness was introduced.
+- [x] The former style memo's unique necessary identities, observations, recommendations, cadence, caveats, anti-patterns, ownership controls and coverage boundaries have been integrated here without dossier-scale repetition; no parallel style file is retained.
+
+**Disposition:** `frozen` text contract and Round-5 eligible under recorded artifact gates. The optional multi-role panel remains dependency-gated; display production and drafting-level choices do not authorize a scientific claim beyond the bounded answer, and no figure has been generated.
+
+### DTW orientation and deletion-sign freeze
+The listed deletion metric is the native Legacy DTW **similarity** `legacy_dtw_v1_abs_n_unclipped`, six sensor channels scored separately after one shared historical circular-lag alignment and then averaged, on the same 40-control-step window for parent and deletion variant. Higher is better. Each row uses $\Delta=\mathrm{similarity}_{deletion}-\mathrm{similarity}_{parent}$; therefore a negative delta is a degradation. Aggregate means/minima summarize the listed cases only and are not uncertainty estimates. Illusion rows remain historical/negative and are omitted from the manuscript-facing Kármán comparison.

docs/JFM_WYQ/ROUND4_TOP_DOWN/SECTION_06_KARMAN_FIELD_CCD.md

@@ -0,0 +1,225 @@
+# Round 4 integrated section architecture — Kármán mean fields and action-correlated modes
+
+## 1. Status and scope
+
+**Status:** `frozen` text contract; Round-5 eligible under recorded artifact gates. Main composites remain dependency-gated, POD remains supplemental, and no figure has been generated or approved. This is an integrated author-facing Round-4 Section-6 architecture, not manuscript prose, a result dossier, or panel/artifact approval. It supersedes the separate Section-6 style memo while retaining its unique necessary style evidence and controls.
+
+**Scope and claim ceiling.** This section owns the corrected four-role Kármán mean-field comparison, scalar error-difference accounting, the separately centred non-paired residual, descriptive CCD organization and the same-object POD boundary. It may interpret only the periodic Kármán cloak in the parabolic-inflow/no-slip-wall channel configuration under the exact local estimands below. It may not reopen research, transfer findings across chains/cases, promote diagnostics without authority, or turn correlation into mechanism.
+
+Targeted Nowledge recall was used for prior Round-2/3/4 decisions; every operative fact below was checked against current repository authorities or is explicitly labelled as a new read-only diagnostic.
+
+## 2. Reader-facing title, question and bounded answer
+
+**Recommended title:** **Mean-field correction and action-correlated organization**.
+
+**Question.** For the periodic Kármán cloak in the parabolic-inflow/no-slip-wall channel configuration, how does the controlled mean streamwise field alter the passive pinball deficit, how much of the declared vector mean-field error reduction is measured by the constant-reference step, and what additional information does CCD provide about the separately centred residual?
+
+**Bounded answer.** On the exact four-role common-fluid wake ROI, the passive field has a predominantly negative `zero - target` mean-$u_x$ difference and the controlled field adds a largely opposite-signed `DRL - zero` mean-$u_x$ difference. This is compatible with adding downstream streamwise velocity where the pinball produces a deficit; it is not a closed momentum budget, exact cancellation or causal mechanism. Under the separate coordinate-weighted vector-error estimand, the constant reference accounts for `93.42%` of the measured zero-to-DRL scalar mean-field error reduction and the constant-to-DRL increment accounts for the remaining `6.58%`. CCD then associates a rank-3 representation of a separately centred, non-paired DRL-minus-constant-template residual with action coordinates. POD organizes field variance/energy under its metric without action information; for this same residual its tested rank-3 field subspace is essentially the same, so CCD's bounded added information is action-coordinate association, not superior or unique modes.
+
+## 3. Accepted author feedback implemented
+
+- Foreground `zero - target` and `DRL - zero` mean-$u_x$ fields before the scalar accounting and CCD.
+- Retain target, physical zero, constant reference and DRL as four distinct roles.
+- Use only **measured contribution to the scalar mean-field error reduction** (or exact equivalent) for `93.42/6.58`.
+- Introduce CCD relative to POD without superiority, causality, response or control-authority language.
+- Keep detailed POD/CCD principal-cosine comparison in the supplement by default.
+- Use mean-$u_x$ fields for physical reading and explicitly connect them to the two-component coordinate-weighted L2-style scalar ledger.
+- Keep the exact term **action-correlated correction-field modes** and reader-facing physical configuration names.
+
+## 4. Six-paragraph architecture (~17% body-text share)
+
+1. **Four-role estimand and viewing rule.** Introduce target, physical zero, constant reference and DRL means for the reference periodic Kármán cloak case in the parabolic-inflow/no-slip-wall channel configuration; the inherited code label remains provenance only and supplies no reader-facing Reynolds value. Bind all quantitative Section-6 accounting to the inclusive `34<=x/D<=54`, `|y/D|<=5` ROI, exact four-role common solver-fluid mask, coordinate quadrature and retained mean window. State immediately that the Section-4 mean/phase L2 values are a different estimand.
+2. **Physical mean-$u_x$ observation and four means.** Read the four mean streamwise fields, then the signed differences `zero - target` and `DRL - zero`. The former identifies the passive deficit relative to the target; the latter shows where controlled flow adds/removes streamwise velocity relative to physical zero. Their near-opposition supports only the bounded compatibility statement above. The two signed scalar fields are selected views of the same two-component means used by the vector ledger, not the vector error itself.
+3. **Exact scalar accounting.** Give `E_zero=0.4694352273`, `E_constant=0.1110140739`, `E_DRL=0.0857787860`, followed by `(E_0-E_C)/(E_0-E_D)=93.42%` and `(E_C-E_D)/(E_0-E_D)=6.58%`. Say the constant reference **accounts for 93.42% of the measured zero-to-DRL scalar mean-field error reduction** and the constant-to-DRL increment **accounts for the remaining 6.58%**; prohibit energy, explained variance, physical contribution, SR-term share and causal attribution.
+4. **Separately centred non-paired residual.** Define `U_(c,b)=(q^D_(c,b)-bar(q_D))-(hat(q)^C_b-bar(q_C))` from 19 DRL cycles, a separate 19-cycle constant template and 10 phase bins. No cross-role cycle pairing, lag, wrap, interpolation, whitening, POD preprojection or counterfactual claim.
+5. **CCD relative to POD.** Explain conceptually: POD selects a low-rank field basis by variance/energy under its declared weighted metric and does not use actions; CCD relates the same centred residual representation to centred same-boundary DRL effective-action coordinates under the exact cross-correlation operator. Name the outputs **action-correlated correction-field modes**. The preferred pinball-inclusive sibling gives a rank-3 action-coordinate/phase view; detailed principal cosines move to supplement because they establish a boundary rather than the physical main line.
+6. **Boundary and synthesis bridge.** Conclude that the Kármán evidence supports a mean-dominant scalar error-reduction ledger plus a smaller, descriptively action-associated residual organization. It does not establish a momentum budget, mechanism, response time, unique basis, universality or independent triangulation with SR. Export only this bounded organization to cross-case and later steady synthesis.
+
+**Compression rule.** If space is tight, combine paragraphs 2–3 but retain the local scalar caveat and the Section-4 different-estimand warning. Never combine the mean and modal estimands in the opening paragraph.
+
+## 5. Detailed content and style controls
+
+### Observed JFM practice versus Section-6 recommendation
+
+**Observed practice.** Loiseau et al. move from sensor dynamics and predictive performance to full-state reconstruction and then feature-correlated modes as interpretation. Deng et al. establish phenomenology before mean/fluctuation decomposition and choose representations according to regime. Cornejo Maceda et al. couple scalar objective, explicit law and corresponding fields through a progressively richer control hierarchy. Across these sampled papers, figures advance an inference rather than forming mode galleries, and equations appear when the object they define becomes necessary. One paragraph carries one principal figure claim; captions carry role, component, normalization/orientation, domain, arbitrary sign, lineage and the local noncausal ceiling.
+
+**Recommendation.** Use the narrowing cadence `four-role contract → mean fields → scalar error-difference accounting → local caveat → separately centred residual → action-correlated modal organization → same-object POD boundary → non-transfer bridge`. Means precede modes; the scalar estimand and comparator precede CCD. This ordering prevents attractive modal views from being mistaken for the primary performance estimand or a causal decomposition. Literature supplies style precedent only and establishes no project result, priority or validity.
+
+### Scalar-versus-modal separation and local caveats
+
+Keep three propositions distinct: means are compared under one role/ROI/mask/window contract; scalar error differences provide arithmetic accounting; CCD organizes a separately defined residual. Never use scalar shares as modal energy, explained variance, modal contribution, actuator attribution or dynamic-reconstruction validation.
+
+Place each limit beside the object it limits:
+- scalar shares are arithmetic accounting, not energy or causal contribution;
+- the residual is separately centred and non-paired, not a matched counterfactual or response;
+- CCD is descriptive action correlation, not lag, direction, response time, authority or mechanism;
+- mode sign is globally arbitrary, and rotation/parity views do not measure wall vorticity;
+- POD comparison uses one parent lineage and is not independent validation or triangulation;
+- synthesis is restricted to the named Kármán cloak and does not transfer to the other configuration, steady, Vortex, Illusion or OID.
+
+A final disclaimer cannot repair causal verbs used earlier. Prefer `defines`, `compares`, `separates`, `uses separate centring`, `is non-paired`, `is associated with`, and `spans the tested rank`. Use `accounts for`, `organizes the residual`, `is action-correlated`, `is phase-organized`, or `adds action-coordinate association` only with the exact contract and adjacent caveat.
+
+### Equation placement and POD comparison rationale
+
+1. **Scalar-share identity — required if the percentages are retained.** Place after the three mean errors and before interpretation; define numerator and denominator and say locally that the shares are not energy or causality.
+2. **Separately centred residual — required.** Place immediately before residual/mode evidence and state non-pairing in the same paragraph.
+3. **CCD operator — optional.** Retain only if its field side, action side, weighting or rank is needed for the adjacent interpretation; otherwise use prose and place the derivation in the supplement.
+4. Do not duplicate reusable L2/action definitions owned by Section 2, the SR law owned by Section 5, a general POD derivation, or OID equations.
+5. Compare CCD and POD only on identical residual snapshots, rank, weighting and domain. POD is introduced after CCD to test whether the action-informed construction supplies a distinct tested field subspace. Because the subspaces are near-equivalent here, POD detail bounds the claim: CCD adds action-coordinate association/phase organization, not a superior, unique or more physical field basis.
+
+### Bridges and anti-patterns
+
+The bridge sequence is: SR term ranking motivates the field-level constant-reference question but does not predict it; the scalar ledger leaves a residual organization question; fixed centring/non-pairing permits a descriptive CCD question; action association motivates the same-object POD boundary; the close states that changed cross-case contracts cannot inherit the Kármán interpretation.
+
+Avoid:
+- linear algebra or a mode gallery before roles, means and scalar estimand;
+- four fields without comparator, mask, window and acquisition;
+- percentages without denominator and the local non-energy caveat;
+- calling the residual a response or paired counterfactual, or a lag a response time/direction;
+- calling CCD strengths energy/explained variance, or inferring superiority from clean modes/action labels;
+- equating SR terms with CCD coordinates, using POD as a straw comparator, or cataloguing methods without a live boundary question;
+- calling Section-4 and Section-6 values agreement because they are close, or one lineage triangulation;
+- delaying caveats until Discussion, narrating panels mechanically, treating inventory as artifact existence, or closing with a section-number announcement instead of the non-transfer bridge.
+
+## 6. Displays and figure sequence
+
+### Main S6-1 — Mean streamwise correction and vector-error ledger
+
+A two-tier figure. Top: four aligned mean-$u_x$ fields in the sole intentional accounting-order exception: target → physical zero → constant reference → DRL. Bottom-left: signed `zero - target` and `DRL - zero` mean-$u_x$ fields with one shared symmetric scale centred at zero. Bottom-right: compact vector-error dumbbell/bar ledger for `E_zero`, `E_constant`, `E_DRL`, with the two ordered scalar differences and `93.42/6.58` brackets. The caption must say the fields show only the streamwise component, while `E_r` uses both velocity components, coordinate weights and the exact four-role common mask. This order is required because the additive ledger reads zero → constant → DRL; it does not redefine the standard three-role reader order target → controlled → physical zero used elsewhere. Figure specification: `docs/JFM_WYQ/FIGURE_PLANS/SECTION_06_FIGURE_SPEC.md`.
+
+### Main S6-2 — Residual definition, action coordinates and phase organization
+
+A compact inference-led sequence: define the residual needed to test whether the smaller post-constant remainder has action-associated organization; three action-coordinate columns (`front`, `rear-symmetric`, `rear-antisymmetric`) or an equally compact action-coordinate matrix; then two or four registered phase-bin rank-3 residual reconstructions. Avoid a mode gallery. Use the preferred pinball-inclusive derived sibling only, label mode sign as globally arbitrary, and keep body-local rotation out unless needed as a small diagnostic. This figure establishes descriptive action association and phase organization, not response or causality.
+
+### Supplement S6-S1 — POD boundary and diagnostics
+
+Place the detailed raw weighted POD versus CCD comparison, exact principal cosines `[0.9999616609, 0.9998971005, 0.9976439804]`, rank-3 stability (`minimum declared cosine 0.9974651867`), operator/centering detail, and optional mode parity/rotation diagnostics here. Caption: same residual object and essentially same tested rank-3 field subspace; CCD adds action-coordinate association only. If S6-2 becomes too dense, move individual modes and all rotation diagnostics here as well.
+
+### Supplement S6-S2 — Cross-cloak mean-$u_x$ audit, only if promoted after coordinator review
+
+A small multiples audit may show `zero - target` and `controlled - zero` means for selected additional periodic cloak cases, each with its own configuration, case, role/window and scale. It is qualitative support for recurrence of the signed pattern only. It must not inherit the Section-6 ROI, exact mask, coordinate weighting, `93.42/6.58`, constant-reference arm or CCD interpretation. Default is omit until cases and scales are selected and their role/window comparability is audited.
+
+The sole production authority for all Section-6 layouts, variables, subtraction signs, sources, readiness, color conventions, captions, fallbacks and non-claims is `docs/JFM_WYQ/FIGURE_PLANS/SECTION_06_FIGURE_SPEC.md`. No architecture inventory entry implies an existing publication-ready artifact.
+
+### Display and equation readiness controls
+
+- **S6-1 mean comparison + scalar ledger:** arrays and values are authority-bound; manuscript composite requires redraw. Fallback is exact scalar ledger plus available signed-field material, never a relabelled `DRL - constant` panel.
+- **Scalar-share equation:** required with retained percentages; fallback is a verbal definition or compact ledger with no unexplained percentage.
+- **Residual equation:** required and definition-ready; retained panels remain dependency-gated. Fallback is the equation plus explicit non-pairing.
+- **S6-2 residual/action/phase display:** source arrays and draft views exist; manuscript recomposition and final orientation/lineage checks remain dependency-gated. Drop redundant modes before dropping the residual definition or local caveats.
+- **POD boundary:** numerical comparison and draft view are ready for supplementary recomposition, but it is valid only for identical residual/rank/weight/domain. Fallback is one bounded body sentence.
+- **Cross-cloak audit:** source inventory only, not panel-ready; omit by default.
+
+## 7. Data and readiness summary
+
+- **Exact Kármán four-role processed arrays — available.** External authority `.../canonical/results/roi-mean/arrays.npz` contains `mean_target`, `mean_zero`, `mean_constant_mean`, `mean_drl`, common/ROI masks and weights, plus stored `constant - zero` and `DRL - constant` vector increments. It does **not** store named `zero - target` or `DRL - zero` arrays, but both are exact direct subtractions of the stored means. No CFD is needed.
+- **Existing Kármán drafts — available, not publication-ready.** `01_roi_mean_performance` is the vector scalar ledger; `02_roi_mean_target_error_fields` contains `zero - target`, `constant - target`, `DRL - target` for both components; `03_roi_mean_corrections` contains `constant - zero` and `DRL - constant`. No existing draft shows the requested `DRL - zero` panel.
+- **Read-only sign check — independently derived, not current claim authority.** On the exact Kármán ROI, direct array inspection gives weighted mean-$u_x$ cosine `-0.9961` between `zero - target` and `DRL - zero`, with relative residual norm `0.1662`. This verifies subtraction order and near-complementarity for figure design; the coordinator must decide whether to authority-bind/report either diagnostic. The main claim does not require these new numbers.
+- **CCD processed arrays/drafts — available.** The preferred repository reader publication includes `plot_data.npz/json`, three mode velocity fields, action coordinates and phase reconstructions. It is redraw-capable, but the current four figure groups are drafts/reader views rather than manuscript panels.
+- **Cross-cloak mean arrays — source data available, no signed-difference publication panels.** `src/drl_pinball/data/reproduction` links to role artifacts whose periodic `phase_fields.npz` normally contain complete-cycle `mean_ux/mean_uy`. Direct inspection found means for parabolic/no-slip inherited Kármán labels 50/100/200/400; uniform/free-slip Kármán Reynolds cases (`kar_re60/100/200/400`, five controlled realizations at `kar_re100`); and disturbance-size cases (`kar_d075/d15/d2`). Existing `plot_reproduction_summary.py` renders phase-0 vorticity, while `plot_flow_fields.py` can render target-error magnitude and streamlines; neither renders signed complementary mean-$u_x$ pairs. These arrays make a supplementary audit feasible, not ready.
+- **Cross-case caution.** A read-only unweighted full-grid diagnostic showed opposite-signed tendency in most inspected cases but clear weakening for harder/high-Re or large-disturbance cases. Because masks, configurations, seeds, scales and estimands differ, no broad quantitative complementarity claim is frozen and no Kármán exact estimand transfers.
+
+## 8. Dated paragraph-level literature/style enrichment (2026-08-11)
+
+Literature establishes sampled presentation practice only: it does not prove an internal CCD result, causality, validity, priority or artifact readiness. DOI/PDF states were checked in the DynamisLab Undermind workspace on 2026-08-11. These are `full-PDF` targeted reads, not equation-by-equation verification or independent reproduction.
+
+### Exact papers, DOIs and read states
+
+1. Deng et al. (2020), “Low-order model for successive bifurcations of the fluidic pinball”, *JFM* **884**, A37, DOI `10.1017/jfm.2019.959`, `full-PDF`: phenomenology §2.3; distorted mean/fluctuation §3.2; POD/modal hierarchy §§4–5.
+2. Schlegel et al. (2012), “On least-order flow representations for aerodynamics and aeroacoustics”, *JFM* **697**, 367–398, DOI `10.1017/jfm.2012.70`, `full-PDF` (workspace-uploaded): centred fluctuations/POD §2.1; observable objective/assumption §2.3; centring/algorithm §2.5; comparisons §4; boundaries Appendix A.
+3. Borée (2003), “Extended proper orthogonal decomposition: a tool to analyse correlated events in turbulent flows”, *Experiments in Fluids* **35**, 188–192, DOI `10.1007/S00348-003-0656-3`, `full-PDF` (workspace-uploaded): POD §2; extended modes and correlated/residual split §3; LSE relation §4.
+4. Tadmor et al. (2010), “Mean field representation of the natural and actuated cylinder wake”, *Physics of Fluids* **22**, 034102, DOI `10.1063/1.3298960`, `full-PDF` (workspace-uploaded): roles §II.B; base-before-fluctuation §II.C; shift-mode comparisons §§III–V; actuation Appendix.
+5. Loiseau, Noack & Brunton (2018), “Sparse reduced-order modelling: sensor-based dynamics to full-state estimation”, *JFM* **844**, 459–490, DOI `10.1017/jfm.2018.147`, `full-PDF`: state §3.1; reconstruction §3.3; feature/generalized modes and POD comparison §4.
+6. Lyu (2023), “Principal correlation decomposition of experimental and numerical data for observable diagnosis”, arXiv `2312.14858`, no DOI found; `https://arxiv.org/abs/2312.14858`, `full-PDF`: paired matrices/decomposition §2.1; cross-correlation §2.2; sampling/observable limits §§2.3–2.4; POD comparison §2.5; reconstruction §4. This preprint informs CCD assumptions, not JFM style authority.
+7. Reused mean-field comparators from the same batch: Cornejo Maceda et al. (2021), DOI `10.1017/jfm.2021.301`, `full-PDF`; Rabault et al. (2019), DOI `10.1017/jfm.2019.62`, `full-PDF`.
+
+### Paragraph jobs: close examples, observed practice and recommendation
+
+- **P1 — four roles/estimand** (Tadmor; Schlegel; Deng; Loiseau): sampled papers define baseline/actuated objects, domain/norm and optimized quantity before modes. **Recommend:** name target, physical zero, constant reference and DRL; bind ROI, mask, quadrature and mean window; put the Section-4 different-estimand warning in this opening contract.
+- **P2 — mean fields then scalar ledger** (Tadmor; Deng; Cornejo Maceda; Rabault): physical state/mean fields precede richer decomposition; captions identify cases/components while body interprets comparators. **Recommend:** four means and signed `zero - target`/`DRL - zero` mean-$u_x$ views before the two-component vector-error ledger; state locally that the scalar views are not the vector error.
+- **P3 — 93.42% local caveat** (Deng; Tadmor; Schlegel; Lyu): percentages are tied to explicit denominators—energy, objective resolution, correlation or reconstruction. **Recommend:** put the share identity after the three errors; repeat `measured zero-to-DRL scalar mean-field error reduction`; place `not energy, variance, modal share, physical contribution or causality` in the same paragraph.
+- **P4 — non-paired residual definition** (Schlegel; Borée; Deng; Lyu): mean subtraction, paired objects and correlated/residual parts precede modes; these pairing assumptions sharpen the present boundary. **Recommend:** display $U_{(c,b)}$ immediately before CCD evidence; identify separate 19-cycle DRL/constant ensembles and ten bins; state no cross-role cycle pairing, lag, wrap, interpolation, whitening, POD preprojection or counterfactual.
+- **P5 — action-correlated correction-field modes** (Schlegel; Borée; Lyu; Loiseau): `mode` denotes a basis vector and `subspace` its span; energy ranking differs from observable-correlation ranking. **Recommend:** reserve the exact term for CCD basis vectors; say they span a tested rank-3 field subspace and associate the centred residual with centred same-boundary effective-action coordinates. Use `associated/correlated`, never `response`, `authority`, `caused`, `driven` or `mechanism`.
+- **P6 — POD boundary/non-transfer bridge** (Schlegel; Borée; Tadmor; Deng): useful comparisons hold object, rank, metric and purpose fixed. **Recommend:** identical residual snapshots/rank/weight/domain only; use the near-equivalent field subspace to bound distinctness. Ask what remains legible under changed role/temporal/configuration contracts; export neither percentages nor modes.
+
+**Sample-bounded synthesis.** The six-paper core defines means before fluctuations/modes, places centring/correlation operators immediately before dependent evidence, distinguishes modes from their span, binds POD percentages to energy/reconstruction, and compares observable-oriented decompositions by matched purpose. Captions decode field/component/domain/sign; body states inference. Recommended cadence: `four-role contract → means → scalar ledger/local caveat → separately centred non-paired residual → action-correlated modes → identical-object POD boundary → changed-contract question`. This is not a universal convention, and no paper verifies the internal ledger, residual, modes or POD comparison.
+
+### Green/amber/red lexicon
+
+- **Green:** `four-role mean-field comparison`, `coordinate-weighted vector error`, `accounts for ... of the measured scalar mean-field error reduction`, `separately centred`, `non-paired residual`, `action-correlated correction-field modes`, `field basis`, `rank-3 field subspace`, `same-object POD boundary`, `associated with action coordinates`, `descriptive`, `compatible with`.
+- **Amber—exact object and adjacent caveat required:** `correction`, `captures`, `explains`, `contribution`, `dominant`, `phase-organized`, `reconstruction`, `POD comparison`, `percentage`. Prefer `signed difference`, `accounts for`, `organizes`, `measured scalar share`, `tested subspace`; state the percentage denominator.
+- **Red:** `momentum restoration`, `exact cancellation`, `closed budget`, `response mode`, `actuation-caused mode`, `control authority`, `mechanism`, `paired counterfactual`, `explained variance` for 93.42%, `CCD energy`, `CCD superiority`, `unique modes`, `independent validation/triangulation`, or cross-chain transfer.
+
+### Citation/caveat, caption/body, bridges and anti-patterns
+
+Cite Tadmor/Deng for mean-first decomposition; Schlegel/Borée for POD versus correlated objects and basis/subspace vocabulary; Lyu only for CCD assumptions; Loiseau for feature-linked reconstruction/purpose-matched comparison. Never cite literature as verification of internal percentages, cosines or fields. Place different-estimand in P1; streamwise-view/vector-ledger distinction P2; denominator/non-energy ceiling P3; centring/non-pairing P4; correlation/nonresponse and arbitrary sign P5; identical-object/non-transfer P6.
+
+**S6-1 caption:** configuration, roles, component, subtraction signs, ROI/mask/window, scale and two-component weighted $E_r$. **S6-1 body:** bounded physical compatibility, scalar accounting, semantic ceiling, consequence. **S6-2 caption:** residual lineage, action centring/boundary, rank, weight/domain, phase registration, arbitrary sign, descriptive ceiling. **S6-2 body:** added action association and noncausal limit. **POD supplement caption:** identical residual/rank/weight/domain and “essentially the same tested rank-3 field subspace”; body keeps only the boundary sentence.
+
+End with a changed-contract question, not a section announcement or SR/CCD mutual confirmation. Avoid opening with modes/singular values; interchanging basis/mode/subspace; importing paired EPOD/OID interpretation into this non-paired residual; comparing CCD correlation with POD energy as one denominator; unexplained 93.42%; caption-level causal propagation; clean modes as “more physical”; or near-identical subspaces as validation.
+
+### Coverage
+
+- `read/full-PDF`: six core papers plus two reused mean-field papers at named locations; no independent reproduction.
+- `workspace-inspected`: folders/files and completed searches `Observable correlated decomposition lineage` and `JFM rigor for correction-field and PCD analysis`; existing library sufficed, so no new search was launched.
+- `recall/read`: narrow Nowledge recall and frozen packet; memory was navigation only, packet remained controlling.
+- `not-covered`: exhaustive sentence coding, new science, CFD/training, artifacts, panel approval or cross-case audit.
+
+## 9. Ownership, interfaces and near-collision controls
+
+**Owns:** exact corrected four-role means and scalar shares; requested mean-$u_x$ physical reading; residual definition; CCD/POD conceptual boundary; action-correlated correction-field modes; S6-1/S6-2 and supplementary POD detail.
+
+**Imports:** Section 2 configuration/Re/action/role/metric definitions; Section 4 named Kármán performance only as context and explicit different-estimand warning; Section 5 the action-law question and bounded persistent/smaller-correction bridge, not its formula or deletion numbers.
+
+**Exports:** the exact Kármán scalar error-reduction semantics, the bounded mean-$u_x$ compatibility observation, and descriptive action-coordinate residual organization. Section 7 may import only that the deep Kármán interpretation is case-specific; Section 8 may use compatibility language, not validation or mechanism.
+
+**Near-collision controls:**
+- Section-4 standardized PPO fields and Section-6 corrected means differ in role sets, masks, weighting/normalization and retained windows. No arithmetic, replication, validation, agreement, closure or triangulation.
+- Section-6 scalar shares and modal quantities are different estimands: mean target-error differences versus weighted centred residual/action decomposition. Shares are not energy, variance or modal contribution.
+- Section-5 SR ranking and Section-6 action coordinates are related diagnostics under different observation/action and field/action constructions. Use a brief descriptive bridge only; no derivation, equivalence or validation.
+- CCD versus POD is a same-field-object boundary only when residual, samples, rank, weighting and domain match. It can bound distinctness but cannot confer causality or independence.
+- Other configurations and steady/cross-cases have different chains, roles, fields, metrics and temporal contracts; context/non-transfer only.
+
+## 10. Exclusions and non-claims
+
+The generic manuscript metric dictionary remains owned by Section 2; this section preserves its non-equivalent coordinate-weighted four-role mean-vector errors and separately centred residual metrics. No closed momentum or energy budget; no `momentum restoration`; no exact cancellation; no causal/unique/universal cloak mechanism; no response time; no paired counterfactual; no physical contribution percentage; no energy or explained-variance share; no SR-term share; no CCD superiority; no unique CCD modes; no POD inferiority; no uncertainty/generalization claim; no OID result; no reader-facing internal chain labels; no M04/M06 replication, agreement, validation or triangulation.
+
+## 11. Author choices, dependencies and deferrals
+
+1. Approve the title above or shorten to **Kármán mean-field correction and correlated modes**.
+2. Approve S6-1's four means + two signed mean-$u_x$ differences + compact vector ledger. Recommended: yes; redraw required from existing arrays.
+3. Keep the independently checked complementarity numbers internal/caption-free (safe default); promotion requires a later explicit authority extension.
+4. Select two versus four phase bins for S6-2 after final-size testing; default two if the action-coordinate matrix remains legible.
+5. Keep detailed POD comparison supplementary (safe default); promote only one prose sentence in the body.
+6. Omit cross-cloak small multiples (current safe default). Any future reinstatement requires a separately authorized role/window/mask/scale audit.
+
+**Execution dependencies and safe defaults.** S6-1 needs a new render from existing means; S6-2 needs manuscript recomposition; retained captions require exact orientation, lineage and role/component labels. The POD comparison stays supplementary unless the identical-object contract remains legible. The CCD operator is omitted unless it materially clarifies adjacent interpretation. Missing optional panels do not block the bounded text architecture.
+
+**Deferred, not reopened:** the mapped fourth modal paper; artifact audit of hidden/external payloads; optional complementarity-number promotion; cross-cloak case/scale selection; publication-size typography. None changes the current scientific answer.
+
+## 12. Coverage ledger
+
+- `read`: Round-4 global/display/quality/style-crosswalk authorities, the former Section-6 style memo, and integrated Section 1/2 pattern files; Round-2 freeze/style; Round-3 Sections 4–6 at required boundaries; current CCD results, final results, derivation, corrected contract, figures/data, code that creates exact mean arrays and draft panels; DRL data dictionary, role manifest and relevant render/evaluator scripts. Repository reports were read for architecture/evidence boundaries; this is not recomputation or artifact audit.
+- `read/full-PDF`: Loiseau et al. (2018), Deng et al. (2020), and Cornejo Maceda et al. (2021), targeted as style precedents at the levels stated in §8; literature establishes no internal result.
+- `mapped`: Noack et al. (2016) exact record/abstract, with no section-practice inference; preferred external corrected and pinball-inclusive immutable results/publications; linked periodic role payload families and existing vorticity composites. `mapped` is not `read`.
+- `not-covered`: raw-array recomputation beyond the already recorded read-only checks; CFD/training; new rendering; exhaustive historical archives or modal-literature sentence coding; publication-scale typography testing; independently audited common masks for cross-cloak cases.
+- `external-inaccessible`: Noack et al. (2016) full text during the style pass and hidden/external payloads not needed for architecture. External CCD and DRL payloads already recorded as readable remain authority-led, not newly artifact-audited here.
+- `gap`: requested Kármán differences are derivable but not both stored/named as arrays or jointly plotted; S6-1 needs a new render. Cross-cloak signed mean-$u_x$ panels and a common audited qualitative comparison contract do not exist. S6-2 needs manuscript recomposition from existing arrays; POD detail is ready for supplement recomposition.
+
+
+## 13. Quality gate and disposition
+
+- [x] One integrated Section-6 architecture retains the useful style controls without reproducing a dossier.
+- [x] Status/scope, title/question/answer, author feedback, six paragraph jobs, detailed content/style controls, displays/equations, exact literature identities, ownership/interfaces, choices/dependencies, coverage and disposition are explicit.
+- [x] Mean fields precede scalar accounting; scalar and modal estimands remain separate; local caveats accompany each object.
+- [x] Exact `zero - target` and `DRL - zero` subtraction order is preserved.
+- [x] The constant reference `accounts for 93.42%` and the constant-to-DRL increment `accounts for` the remaining `6.58%` of the measured scalar mean-field error reduction; neither is energy, causality or physical contribution.
+- [x] The residual remains separately centred and non-paired; CCD remains descriptive/noncausal and uses `action-correlated correction-field modes`.
+- [x] POD detail remains supplementary by default and is compared only on the identical residual/rank/weight/domain.
+- [x] Complementarity diagnostics `-0.9961` and `0.1662` remain internal-only unless current authority is extended.
+- [x] Every Section-6 figure reference points to `docs/JFM_WYQ/FIGURE_PLANS/SECTION_06_FIGURE_SPEC.md`; inventory does not imply readiness.
+- [x] Exact selected JFM identities, DOIs, read levels, observed practice and recommendation are retained.
+- [x] Ownership, near-collision rules, anti-patterns, readiness/fallbacks and coverage states are explicit.
+- [x] The former style memo's unique necessary content has been integrated here; no parallel style file is retained.
+
+**Disposition:** `frozen` text contract and Round-5 eligible under recorded artifact gates. Current author decisions and scientific boundaries are preserved; optional title/display production choices remain open without reopening the text architecture, and no figure has been generated.

docs/JFM_WYQ/ROUND4_TOP_DOWN/SECTION_07_CROSS_CASE_EXTENSIONS.md

@@ -0,0 +1,134 @@
+# Round 4 integrated section architecture — cross-case extensions
+
+## 1. Status, title and scope
+
+**Status:** `frozen` text contract; Round-5 eligible for text under the recorded gates. Quantitative Vortex/Erase displays remain blocked, role-field visuals remain dependency-gated, and no figure has been generated or approved. This is an integration-corrected author-facing Round-4 Section-7 architecture, not manuscript prose or publication-panel approval. It combines the former Section/Style functions in one file and implements the accepted 2026-08-10 author feedback.
+
+**Recommended title:** **Extensions under changed temporal and target contracts**.
+
+**Question.** What role-field material is available when the periodic Kármán cloak is replaced by an incoming-vortex event, and when background matching is replaced by a declared non-zero illusion target or an erased-flow target?
+
+**Bounded answer.** The named PPO acquisitions exercise force-plus-sparse-observation state feedback under changed temporal and target definitions. For Vortex and Erase, current target, controlled and physical-zero role fields support bounded visual/role-field material only. Quantitative event or Erase efficacy is blocked until a versioned evaluator binds the exact target, roles, ROI/mask, normalization, window and provenance. Illusion retains a named positive PPO non-zero-target capability. None transfers the Kármán SR/CCD interpretation, supplies one cross-case score, or establishes observability, memorylessness, robustness or mechanism.
+
+Targeted Nowledge `memory_search` was used for prior Section-7 decisions. Current repository authorities, scripts, manifests, linked payloads and draft renders were then checked; current evidence supersedes the older Round-3 conservative artifact statements.
+
+## 2. Accepted feedback implemented
+
+- Keep a changed-contract argument, not a case gallery.
+- Use two displays only: the Vortex event response and the Illusion-plus-Erase target extension. Detailed production instructions live solely in the two Figure Plans referenced below.
+- Prefer the central Taylor-vortex scene at event offset `+10` for the main result; offsets are event-relative acquisition indices, never periodic phase.
+- Read Vortex through role identity and morphology only; reactive/local-information interpretations remain hypotheses, with a strict arbitrary-flow nonclaim.
+- Connect actions to the current observation-state interface only: state-feedback/Markov execution motivates reactive/local-information potential, but does not prove physical-flow memorylessness, complete observability or a small spatial support for cloaking.
+- Restore Erase as a brief bounded result/design from its actual target, controlled and physical-zero field roles. Separate the historical rolling-reward caveat from the independent field comparison.
+- Use physical configuration names in manuscript-facing prose; internal chain labels occur only in provenance.
+
+## 3. Five paragraph jobs (~10% body-text share)
+
+1. **Change the contract, not the case label.** Begin from the periodic Kármán reference but import none of its numerical results or control-skeleton interpretation. Separate three changes: periodic phase becomes event-relative time for Vortex; the desired signature becomes non-zero for Illusion; and Erase uses a clean-flow target in the presence of an upstream disturbance. State that action and observation interfaces are related execution devices, while role acquisitions, clocks and estimands remain local.
+2. **Vortex potential through state feedback.** Explain that each action is evaluated from the current observation-state interface and that the central Taylor-vortex rollout contains event-relative actuation around the incoming event. **Available role material:** named target, controlled and physical-zero event fields exist for the tested Taylor scene and offsets. **Hypothesis only:** these fields motivate possible reactive/local-information control of comparable-scale nonperiodic or arbitrary-inflow-like disturbances. **Not demonstrated:** arbitrary family/amplitude/offset control, physical-flow Markovianity, observability, a memoryless plant, or dependence on only a small wake region.
+3. **Vortex field result at `+10`.** Lead with the target/controlled/physical-zero field triplet at central incidence and offset `+10`, where existing draft vorticity, streamlines and target-error images are available. State that no versioned event evaluator output currently binds target, roles, ROI/mask, normalization, window and provenance. Therefore print no scalar, reduction, confirmation or controlled-below-zero efficacy statement. Retain the triplet as bounded visual/role-field material and classify quantitative event efficacy as blocked.
+4. **Illusion PPO retargeting.** Define Illusion locally as matching a declared non-zero target signature, with target, controlled and physical-zero roles separated. Use the retained `0.75` target-size PPO demonstration as the principal example. Its role fields, phase/error draft material and evaluation/training summaries support a named PPO capability statement; exact scalar prose remains conditional on binding the selected acquisition and metric output.
+
+5. **Erase role material, synthesis and strict non-transfer.** The Erase acquisition contains named clean-target, controlled and physical-zero role fields plus draft views. The historical reward used a same-rollout evolving target construction and is not an independent clean-field efficacy metric. No current versioned evaluator binds the exact target, roles, ROI/mask, normalization, window and provenance, so quantitative Erase efficacy remains blocked and no scalar, reduction, confirmation or controlled-below-zero claim is retained. Close by separating **available role-field material** from **hypotheses** about changed-target control, then prohibit transfer, robustness, universality, common efficacy and Kármán mechanism claims.
+
+## 4. Figure references
+
+- **S7-1 — Vortex event response:** central Taylor-vortex target/controlled/physical-zero evidence at offset `+10`, with compact event/action context and an explicit quantitative-efficacy-blocked label. Production authority: `docs/JFM_WYQ/FIGURE_PLANS/SECTION_07_FIGURE_01_VORTEX_EVENT_RESPONSE.md`.
+- **S7-2 — Non-zero-target and erase extensions:** retained `0.75` Illusion PPO target/controlled/zero evidence plus bounded Erase target/controlled/physical-zero role material; no efficacy annotation. Production authority: `docs/JFM_WYQ/FIGURE_PLANS/SECTION_07_FIGURE_02_ILLUSION_ERASE_EXTENSIONS.md`.
+
+The author supplies/redraws any geometry or event schematic. Existing outputs are source/draft material, not publication-ready panels.
+
+## 5. Actual data and readiness
+
+### Vortex
+
+- `src/drl_pinball/data/reproduction/legacy/vortex_taylor_y000/{target,controlled,zero}/event_fields.npz` exists for all three roles with `ux`, `uy`, `field_indices`, `relative_offsets=[-10,-5,0,5,10]` and `selected_event_score`.
+- `src/drl_pinball/data/reproduction_flow_plots/all/legacy/vortex_taylor/vortex_taylor_y000/` contains vorticity and streamline renders for every role/offset and target-error renders for controlled and zero. The offset-`+10` triplet is directly available as draft material.
+- `src/drl_pinball/eval/wake_l2.py` explicitly excludes Legacy Vortex and has no event-field loader/output. No stored event-L2 summary was found.
+- Existing snapshots may support a future evaluator, but they are not quantitative field-efficacy authority. A versioned output must bind all quantitative fields before any efficacy statement; no CFD rerun is implied by this planning note.
+
+### Illusion
+
+- Retained PPO source roles exist under both current and historical provenance trees. The selected manuscript example is the uniform-inflow/free-slip-wall `0.75` target-size PPO run (retained seed 43); do not merge it numerically with historical parabolic/no-slip files.
+- Current role navigation is bound by `data/reproduction_plots/manifest.json`; detailed flow-field drafts and error/mean-error materials exist under `data/reproduction_flow_plots/all/` for corresponding retained roles.
+- A final scalar annotation requires choosing one acquisition and binding its existing wake-L2/evaluation summary, role window, mask/geometry guard and phase semantics. The figure remains redraw-ready without an unsourced scalar.
+
+### Erase
+
+- `data/reproduction/legacy/erase/target` contains `late_field.npz`, metadata, timeseries and a clean-target summary. `controlled` and `zero` each contain eight `phase_fields.npz` snapshots, phase-cycle payloads, timeseries, metadata and `erase_summary.json`.
+- `data/reproduction_flow_plots/all/legacy/erase/erase/` contains target late-field vorticity/streamlines and controlled/zero slot-0 vorticity, streamlines and target-error images.
+- Stored signal summaries exist, but they are not field-efficacy authority and are omitted from the section and figure.
+- No stored versioned field-efficacy output binds an exact Erase target/role/ROI-mask/normalization/window/provenance contract. Quantification remains blocked; the role fields remain bounded visual material.
+
+## 6. Imports, exports and exclusions
+
+**Imports.** Section 2 supplies configuration, action/observation, role and metric definitions. Sections 4–6 supply only the periodic Kármán reference and the statement that its deep interpretation is local. Section 6 exports a non-transfer boundary, not a mechanism for these cases.
+
+**Exports.** Section 8 may import only that changing temporal and target contracts exercises named PPO capabilities while leaving transient transfer and objective-transfer questions open. Discussion may compress this into demonstrated cross-task capability plus asymmetric evidence depth.
+
+**Exclusions/nonclaims.** no pooled cross-case score; no Kármán SR/CCD transfer; no arbitrary-flow universality; no unseen-family/amplitude/offset generalization; no robustness, optimality, stability or causal force-to-field claim; no physical-flow memorylessness or full observability; no claim that a small spatial region is sufficient; no reward/field equivalence; no Legacy/current numerical continuation; no publication-ready figure claim.
+
+## 7. Exact production choices
+
+1. Select `vortex_taylor_y000`, offset `+10`, for S7-1; retain other offsets/vertical incidences as supplementary source inventory only.
+2. Omit all Vortex scalars now. A later versioned evaluator must close the complete quantitative contract before reconsideration.
+3. Select the retained `0.75` non-zero-target PPO demonstration for the Illusion half of S7-2; scalar annotation is optional and acquisition-bound.
+4. Retain Erase only as target, controlled and physical-zero role-field material; omit scalars and efficacy wording. Do not use native historical reward as field efficacy.
+5. Use case-local scales where physics differs; a common scale is allowed only within a role triplet for the same variable/metric. Adjacency never implies a common estimand.
+6. Keep all layouts, crops, scales, metric semantics, rendering work, caption language and readiness decisions in the Figure Plans.
+
+## 8. Compact read/gap ledger
+
+- `read`: Round-4 global/display controls, former Section-7 style memo, Round-3 Section 7 and final audit, current Section-6 interface; current Vortex/Erase acquisition authorities, manifests, evaluators/renderers, metadata/summaries and linked field schemas.
+- `memory-recalled`: Round-2/3 Section-7 changed-contract, event-relative and strict non-transfer decisions; repository evidence controlled every operative update.
+- `read`: existence/schema records for central Taylor Vortex and Erase role payloads and relevant draft-render manifest entries; this is not quantitative field-efficacy authority or verification.
+- `mapped`: selected Illusion retained-run summaries and complete current flow-plot family; final scalar binding remains production work.
+- `gap`: no versioned Vortex or Erase evaluator binds exact target, roles, ROI/mask, normalization, window and provenance; quantitative efficacy remains blocked.
+- `not-covered`: new CFD, training, policy evaluation, exhaustive case matrix, publication rendering, or transfer tests.
+
+**Disposition:** `frozen` text contract and Round-5 text eligible after final reconciliation. Vortex/Erase quantitative displays remain blocked, role-field visuals remain dependency-gated, and no figure has been generated.
+
+## 9. Paragraph-level JFM literature/style enrichment — 2026-08-11
+
+**Scope and authority.** This dated style area enriches the frozen five-paragraph text contract; it does not reopen its science, statuses, displays, controller interpretation or quantitative gates. The observations below come from targeted full-PDF paragraph reads. They are rhetorical evidence only: project roles, clocks, estimands and blockers remain controlled by the packet and current repository authorities.
+
+### Exact literature identities and read states
+
+- Deng et al., *Galerkin force model for transient and post-transient dynamics of the fluidic pinball*, **JFM 918, A4** (2021), DOI `10.1017/jfm.2021.299` — `full-PDF read`; used for explicit transient/post-transient separation, local regime clocks and model-failure placement.
+- Deng et al., *Cluster-based hierarchical network model of the fluidic pinball – cartographing transient and post-transient, multi-frequency, multi-attractor behaviour*, **JFM 934, A24** (2022), DOI `10.1017/jfm.2021.1105` — `full-PDF read`; used for regime-local comparison and for showing a failed flat route before the hierarchical remedy.
+- Pino et al., *Comparative analysis of machine learning methods for active flow control*, **JFM 951, A39** (2023), DOI `10.1017/jfm.2023.76` — `full-PDF read`; used for renewing objective, baseline and metric in every test case and refusing a universal algorithm ranking.
+- Cornejo Maceda et al., *Stabilization of the fluidic pinball with gradient-enriched machine learning control*, **JFM 917, A42** (2021), DOI `10.1017/jfm.2021.301` — `full-PDF read`; used for separating visually similar mean states from residual dynamics and for keeping distinct actuation branches local.
+- Beckers & Eldredge, *Deep reinforcement learning of airfoil pitch control in a highly disturbed environment using partial observations*, **Physical Review Fluids 9, 093902** (2024), DOI `10.1103/physrevfluids.9.093902` — `full-PDF read`, non-JFM close comparator; used for event/pulse disturbance language, explicit held-out disturbance tests and the distinction between partial observations and restored Markovian information.
+
+### Observed practice and manuscript recommendation by paragraph job
+
+1. **Changed-contract framing.** **Observed:** Pino et al. reset the uncontrolled baseline, objective and reward/metric for each problem before comparing methods; Deng et al. define transient and post-transient regimes before modelling either. **Recommend:** open with the periodic Kármán comparator, then change one contract coordinate at a time: clock, target and role definition. Cite Pino et al. after the sentence explaining why each extension needs a local comparator, not after the project result. Keep `event-relative`, `non-zero target` and `clean-flow target` in the opening paragraph; do not delay these distinctions to captions.
+
+2. **Vortex event role fields and bounded reactive potential.** **Observed:** Beckers & Eldredge name pulsed disturbances and test unseen pulse combinations explicitly; their memory result is phrased as helping to *partially* restore a Markovian property, not proving the plant Markovian. Deng et al. distinguish the path to an attractor from its post-transient state. **Recommend:** define the incoming event, central incidence and offset convention before interpreting action timing. Attach Beckers & Eldredge to the general precedent that sparse-feedback control under disturbances needs explicit disturbance and information tests. Put the project caveat in the same paragraph: present-state execution **motivates** reactive/local-information potential, whereas arbitrary disturbance control, observability and physical-flow memorylessness remain untested.
+
+3. **Vortex `+10` field result and blocked efficacy.** **Observed:** Deng et al. place the failure of a lower-order transient model immediately beside the comparison that reveals it; captions identify trajectories/regimes while body text states what the mismatch requires. **Recommend:** body order is role triplet → event-relative offset → morphology-only inference → missing evaluator contract → consequence. The no-efficacy sentence must directly follow the attractive `+10` visualization reference. Caption decodes target/controlled/physical-zero, incidence, offset, variable and case-local scale; body states only role identity, morphology, the quantitative block and its consequence. Do not let `response`, `attenuation`, `recovery` or `improvement` appear in the Vortex caption.
+
+4. **Illusion positive PPO / negative SR boundary.** **Observed:** Pino et al. retain successful and unsuccessful algorithm routes within the local problem and use qualifications such as `for this specific problem`; they explicitly reject a general ranking across cases. Cornejo Maceda et al. distinguish a mean field that resembles a target from residual dynamics hidden by that mean. **Recommend:** renew target/controlled/physical-zero roles before naming the retained `0.75` PPO capability. Place the historical negative SR route in the same paragraph, after the positive PPO statement, as a method-specific boundary rather than a contradiction of PPO. A citation can follow the sentence explaining why success of one controller class does not transfer to another; it must not be used to support the internal PPO result.
+
+5. **Erase role fields and strict non-transfer synthesis.** **Observed:** the close papers make negative routes local and conclude with case-specific scope rather than averaging unlike tests. Deng et al.'s hierarchical treatment also keeps global-regime identity separate from local dynamics. **Recommend:** state clean-target, controlled and physical-zero roles before the historical rolling-reward caveat; then put the absent evaluator binding beside the no-efficacy consequence. Close with a two-level synthesis: available changed-target role material, followed by explicitly untested transfer. The bridge should ask what physical context survives across non-equivalent control settings; it should not announce Section 8 or imply a common efficacy axis.
+
+### Wording ladder and local placement
+
+- **Green — licensed by this packet:** `under a changed temporal contract`; `event-relative role fields`; `the fields motivate a reactive/local-information hypothesis`; `a named non-zero-target PPO capability`; `available role-field material`; `case-local`; `compatible with`; `has not been evaluated`; `quantitative efficacy remains blocked`.
+- **Amber — use only with the qualifier adjacent:** `response` only as event organization, never efficacy; `retargeting` only after the declared target is defined; `comparable-scale disturbance` only for the tested scene; `partial-observation state feedback` only as the execution interface; `suggests` or `motivates` only when followed by the exact missing test.
+- **Red — delete from S7 body and captions:** `generalizes`, `transfers`, `robust`, `memoryless`, `observable`, `local control is sufficient`, `suppresses the incoming vortex`, `restores the target`, `erase succeeds`, `controlled below zero`, `common performance`, `same mechanism`, or any Kármán SR/CCD explanation applied to these cases.
+
+### Citation, caveat, caption and bridge controls
+
+- Use one compact citation cluster at the changed-contract claim and at most one local comparator citation in each Vortex/Illusion paragraph. Never append an external citation to an internal field/result sentence in a way that makes the paper appear to verify DynamisLab evidence.
+- Place the Vortex and Erase evaluator caveat immediately after each role-field observation, not in a terminal limitations paragraph. Place the negative Illusion SR route directly after the positive PPO capability.
+- Captions own configuration, roles, event offset/target identity, variables, scales and `visual/role-field only` status. Body owns the inferential ceiling, blocked route and cross-case consequence. Captions must not narrate efficacy.
+- Preferred S7→S8 bridge job: ask whether a separately acquired steady endpoint can provide physical context without transferring the periodic Kármán interpretation or any cross-case metric.
+
+### Anti-patterns and coverage
+
+- Avoid a three-case gallery, package chronology, repeated role inventories, one generic `baseline`, detached caveat paragraphs, a literature catalogue, or a final sentence claiming breadth from adjacency alone.
+- `read`: the five full PDFs listed above, with paragraph-, section- and figure-specific questions; DOI and PDF states confirmed in the DynamisLab Undermind workspace on 2026-08-11.
+- `observed practice`: comparator renewal, event/regime naming, local negative routes, conditional vocabulary, and caption/body division in those PDFs.
+- `recommended practice`: the S7 paragraph order, wording ladder, citation/caveat locations and bridge above; recommendation is synthesis, not a universal JFM rule.
+- `not covered`: no new Vortex/Erase evaluator, scalar, artifact audit, transfer test, controller analysis, figure approval or manuscript prose.

docs/JFM_WYQ/ROUND4_TOP_DOWN/SECTION_08_STEADY_THEORY.md

@@ -0,0 +1,130 @@
+# Section 8 — Steady control as physical context and a theory boundary
+
+**Status:** R5 author-choice gate closed at the recommended equation/composition; integrated combined SECTION+STYLE packet and R5 drafting authority. This is not manuscript prose, equation-source re-verification or figure/artifact approval.  
+**Target share:** about 6%; five argumentative paragraphs.  
+**Question:** What can the separately defined steady result add to the physical interpretation of the Kármán control skeleton, and where does that contextual link fail?  
+**Bounded answer:** In the uniform-inflow/free-slip-wall configuration at $Re_D=50$, the accepted prescribed-rear-wall-motion case is a settling-qualified low-deficit endpoint and exhibits reorganized viscous near-wake fields. This is compatible with the recurrent persistent rear-rotation component established in Sections 5–6, but it is neither its continuation nor its validation. Potential flow gives a useful outer organization when circulations are prescribed, yet leaves those circulations independent of wall spin and selects the opposite sign from the descriptive finite-Re projection. The surviving physical explanation is therefore a testable viscous hypothesis—wall motion changes vorticity and shear-layer production, inter-cylinder transport, separation/recirculation and downstream mean transport—not a closed mechanism.
+
+## Accepted author direction implemented
+
+- Keep the accepted five-part late-return logic and use the section as the manuscript-wide physical/contextual connector, not a second climax or detached theory chapter.
+- Replace the old label-led `gap-facing/base-bleed/gap-jet-family` account with a kinematics-first physical reading. `Gap jet` may describe an observed transport feature only after its direction and field evidence are stated; it does not explain itself. No boat-tail analogy.
+- Use no action-direction schematic in the main figure. Rotation and circulation are made legible by streamline direction/arrow treatment and concise panel labels.
+- Display a compact multi-cylinder outer-solution form, define prescribed circulations and their boundary role, then retain the one-circle incompatibility as the shortest no-slip boundary.
+- Integrate actual CFD endpoint evidence, outer-theory streamlines and the sign failure in one composition. Existing fields and plots are source material, not publication-ready artwork. Detailed production design is owned by `docs/JFM_WYQ/FIGURE_PLANS/SECTION_08_STEADY_CONTEXT_AND_THEORY.md`.
+
+## Five paragraph jobs
+
+1. **Contextual return and narrow question.** Recall in one sentence that steady control was the historical discovery seed. Import from Section 5 only the ranked observation that persistent rear counter-rotation is the dominant tested SR term, and from Section 6 only the bounded organization `persistent/constant reference + smaller dynamic correction`. State immediately that those results belong to the parabolic-inflow/no-slip-wall configuration, whereas this section concerns the separate uniform-inflow/free-slip-wall configuration. The paragraph asks whether the steady branch supplies physical context; it must not imply continuation, amplitude agreement, replication or validation.
+
+2. **Finite-Re endpoint before interpretation.** Define the local actuation only as needed: body order `(front, rear_y_plus, rear_y_minus)`, accepted action $[0,+\Omega,-\Omega]$, and $s=\Omega R/U_\infty$. State the settling-qualified $s=3.55$ endpoint and the direct full-cross-section two-component value $E_{\infty,\mathrm{vector}}(q_{ctl},q_{in})=0.0231672176752314$ at $x/D=10$. Use stationary and reverse states visually to establish that the accepted field is a distinct low-deficit endpoint; stationary is an unsteady reference, and reverse $s=5.2$ is not amplitude matched. Call $s=3.55$ an endpoint, never an optimum, stable branch or universal law.
+
+3. **Physical reading from wall motion to observed fields.** Begin with the no-slip cylinder-wall kinematics $(U_w,V_w)=(-\omega r_y,\omega r_x)$. On the accepted branch, both rear gap-facing cardinals move downstream and both outer cardinals move upstream. That motion directly prescribes tangential wall velocity. The accepted endpoint then shows a reorganized inter-cylinder transport pattern, shear/vorticity distribution, separation/recirculation region and near-wake mean flow relative to stationary/reverse source fields. The compatible but unclosed hypothesis is: imposed wall motion $\rightarrow$ altered wall shear/vorticity production and shear-layer development $\rightarrow$ reorganized inter-cylinder jets and near-body separation/recirculation $\rightarrow$ changed near-wake mean transport $\rightarrow$ low downstream deficit. Only prescribed tangential wall velocity is kinematic fact and the fields are endpoint observations; the arrows between them remain a hypothesis because there is no amplitude-matched intervention, complete same-control-volume momentum/mechanical-energy budget, established force ownership or stability analysis. Do not let `gap jet`, `base bleed`, or a borrowed branch name replace this explanation.
+
+4. **Outer solution, visual organization and failed closure.** Introduce the irrotational impermeability family
+
+$$
+\boldsymbol{u}(\boldsymbol{x})=\boldsymbol{u}_{N}(\boldsymbol{x})+\sum_{j=1}^{3}\Gamma_j\boldsymbol{h}_j(\boldsymbol{x}),
+\qquad \nabla\!\cdot\boldsymbol{h}_j=\nabla\!\times\boldsymbol{h}_j=0,
+\quad \boldsymbol{h}_j\!\cdot\boldsymbol{n}=0,
+\quad \oint_{B_k}\boldsymbol{h}_j\!\cdot d\boldsymbol{\ell}=\delta_{jk}.
+$$
+
+Here $\boldsymbol{u}_N$ is the zero-period Neumann solution and each geometric counter-clockwise circulation $\Gamma_j$ is prescribed independently; the circulations span the finite-dimensional harmonic freedom left by impermeability and are not selected by cylinder spin. The theory streamlines for zero circulation, favorable rear-opposite circulation and the finite-Re-fitted circulation should make that outer organization visible. The concise boundary is
+
+$$
+u_\theta(R,\theta)=-2U_\infty\sin\theta+\frac{\Gamma}{2\pi R}\neq R\Omega\quad\text{for all }\theta\text{ when }U_\infty\ne0,$$
+
+so the outer family is not a rotating no-slip solution. Close the paragraph with the decisive sign failure: the strip outer objective selects $g_{opt}=+6.147944$, while the descriptive projection of the accepted finite-Re endpoint gives $g=-12.200554$. The prescribed-circulation mathematics and bounded strip streamline visualization survive; the wall-spin-to-circulation cancellation explanation does not. Do not call the plotted MFS field finite-Re physics or a validated mechanism.
+
+5. **Synthesis and bridge.** Join only compatible, non-equivalent observations: Sections 5–6 identify a dominant persistent/constant rear-rotation reference plus a smaller dynamic correction under their own configuration and estimands; this section adds an independently acquired steady endpoint, a kinematics-first viscous hypothesis, and a failed inviscid closure. The shared interpretation is organizational rather than causal. Retain the theory-specific missing evidence only: amplitude-matched branch/intervention data, held-out endpoint and convergence qualification, wall-vorticity/separation diagnostics, and a closed same-control-volume momentum/mechanical-energy budget. Detailed synchronized experimental test design belongs to Section 9. Bridge calmly to the author-owned qualitative open-loop experiment and short conclusion; the experiment cannot repair mechanism or quantitative validation.
+
+## Selected equations and jobs
+
+1. **Imported/local wall definition:** $(U_w,V_w)=(-\omega r_y,\omega r_x)$ and $s=\Omega R/U_\infty$. Job: anchor the physical reading in actual surface motion. Keep inline unless Section 2 already makes the wall law fully reusable.
+2. **Steady endpoint metric:** quote the already defined full-cross-section $E_{\infty,\mathrm{vector}}$ and its $x/D=10$ value; do not spend a display re-deriving a reusable metric.
+3. **Required main display:** $\boldsymbol{u}=\boldsymbol{u}_N+\sum_j\Gamma_j\boldsymbol{h}_j$ with period normalization. Job: show the shortest useful multi-cylinder outer-solution form and explain prescribed circulations as boundary-compatible harmonic freedom.
+4. **Concise boundary:** one-circle tangential trace/no-slip incompatibility. Job: prevent streamline resemblance from being read as wall-spin closure. It may be inline or a compact second line adjacent to the main display.
+5. **Required numerical contrast:** $g_{opt}=+6.147944$ versus $g=-12.200554$. Job: fail the proposed finite-Re identification by sign. Full MFS, image-layer, multipole, strip-modal and control-volume derivations remain supplementary.
+
+## Figure reference and actual readiness
+
+Use one main-text figure, specified only in `docs/JFM_WYQ/FIGURE_PLANS/SECTION_08_STEADY_CONTEXT_AND_THEORY.md`. It combines CFD endpoint evidence, three outer-theory streamline states and a compact sign/closure panel; there is no action schematic. Current authenticated endpoint arrays, redraw CSVs, and raw/plate streamline and vorticity assets exist. They are `redraw-ready source material`, not publication-ready panels. The 40-image-layer strip field is `bounded outer-reference ready`: current authority records exact-circle/failure controls, bounded independently discretized unbounded MFS–Laurent agreement, and bounded finite-image consistency, but not exact strip convergence, rotating no-slip transfer or finite-Re mechanism validation.
+
+## Interfaces, exclusions and ownership
+
+**Imports:** Section 2 owns configuration/Re/action/metric definitions; Section 5 supplies only persistent-term ranking; Section 6 supplies only the descriptive persistent/constant-plus-smaller-correction organization and its noncausal ceiling; Section 7 supplies the non-transfer boundary.  
+**Owns:** uniform-inflow/free-slip-wall endpoint and sign, kinematics-first physical hypothesis, outer-solution boundary, sign failure and the late contextual synthesis.  
+**Exports:** one compatible-context statement, the failed circulation-closure boundary, theory-specific missing evidence and a qualitative experiment/conclusion bridge.
+
+**Exclude from main text:** repeated SR formula/deletion values; CCD percentages or mode construction; cross-configuration arithmetic; full endpoint sweep; action schematic; boat-tail analogy; MFS validation plots; error/vorticity galleries; NS budget panels; power/gap-flux bars; full MFS, image, multipole and CV derivations; causal, force-ownership, stability, optimum, robustness, momentum-restoration, universal-law or quantitative-experiment claims.
+
+## Exact author and production choices
+
+1. **Reader-facing title — retained recommendation:** `Steady control as physical context and a theory boundary`.
+2. **Main equation — author accepted for R5:** use the Hodge/period outer-solution display above, with the one-circle trace as its adjacent concise boundary. This supersedes the earlier one-circle-only main-text recommendation because the author explicitly requires the solution form.
+3. **Main figure — author accepted for R5:** one integrated composition; no action schematic; CFD evidence plus $g=0$, $g_{opt}$ and fitted-$g$ theory streamlines plus sign/closure box.
+4. **CFD evidence choice — production decision:** use accepted vorticity/streamlines plus a compact $x/D=10$ profile comparison against $q_{in}$; keep stationary/reverse as smaller comparators. If legibility forces compression, drop stationary field first but retain its profile/reference label.
+5. **Theory status label — required wording:** `prescribed-circulation outer reference (bounded finite-image consistency)`. Do not label it `validated MFS mechanism` or simply `potential prediction`.
+6. **Publication rendering — required:** redraw from authenticated arrays/CSVs with shared coordinates and explicit family/status labels; do not assemble the existing titled plates as the final figure.
+
+## Compact evidence ledger
+
+| State | Material | Operative use / boundary |
+|---|---|---|
+| `read` | Round-4 global ledger, display/equation inventory and prior style memo; Round-3 Section 8 synthesis and final audit; current steady README/RESULTS/DERIVATION/LITERATURE; endpoint manifest; v3b/display reports; strip and finite-Re analyses | Argument, values, mathematical status and claim ceiling verified against current authorities; reading is not new CFD or independent reproduction |
+| `read` | Current figure README/manifest, `field_metrics.csv`, profile/near-wake/gap CSVs, and actual streamline/vorticity/profile images | Exact source identities, coordinates, crops, arrays and redraw routes verified; images remain drafts/data |
+| `mapped` | Sections 5–6 current interface decisions from targeted Nowledge recall and controlling ownership ledger | Imports only; no values pooled and no old file presumed authoritative over current controls |
+| `not-covered` | New CFD/MFS generation, new budgets, stability, full raw-array recomputation, literature expansion | Outside Round 4; no claim inferred |
+| `external-inaccessible` | None needed for the selected source-material plan; authenticated external endpoint paths are hash-bound through manifests | Raw external paths remain production dependencies, not a text blocker |
+
+## Paragraph-level JFM literature/style enrichment — 2026-08-11
+
+**Scope and authority.** This dated style area enriches the five-paragraph `discuss` packet without selecting the main equation or figure composition reserved for Frank. It changes no endpoint, sign, theory status, mechanism ceiling, figure readiness or experiment interface. External literature supplies paragraph-scale rhetorical and comparator practice only.
+
+### Exact literature identities and read states
+
+- Cornejo Maceda et al., *Stabilization of the fluidic pinball with gradient-enriched machine learning control*, **JFM 917, A42** (2021), DOI `10.1017/jfm.2021.301` — `full-PDF read`; steady branch comparator, visually attractive mean fields with residual-dynamics caveat, and local-optimum language.
+- Chan et al., *Vortex suppression and drag reduction in the wake of counter-rotating cylinders*, **JFM 679, 343–382** (2011), DOI `10.1017/jfm.2011.134` — `full-PDF read`; explicit rotation direction before interpretation, non-equivalent normal/reverse branches, and local three-dimensional/optical limits.
+- Li & Zhang, *Reinforcement-learning-based control of confined cylinder wakes with stability analyses*, **JFM 932, A9** (2022), DOI `10.1017/jfm.2021.1045` — `full-PDF read`; separation of controlled performance from dedicated stability evidence and local failure at stronger instability or sparse sensing.
+- Marra, Meilán-Vila & Discetti, *Self-tuning model predictive control for wake flows*, **JFM 983, A26** (2024), DOI `10.1017/jfm.2024.47` — `full-PDF read`; objective-local control mechanisms and distinct front/rear actuation roles.
+- Ueda et al., *On the low-Reynolds-number flow about two rotating circular cylinders*, **JFM 495, 255–281** (2003), DOI `10.1017/S002211200300627X` — `full-PDF read`; no-slip rotation in a viscous asymptotic theory and explicit validity/failure boundaries.
+- Crowdy, *Analytical solutions for uniform potential flow past multiple cylinders*, **European Journal of Mechanics B/Fluids 25, 459–470** (2006), DOI `10.1016/j.euromechflu.2005.11.005` — `full-PDF read`, non-JFM mathematical comparator; exact irrotational scope, zero-circulation choice and streamline visualization discipline.
+- Sun & Ng, *Hydrodynamic interactions among multiple circular cylinders in an inviscid flow*, **JFM 712, 505–530** (2012), DOI `10.1017/jfm.2012.434` — `full-PDF read`; impenetrability-based multiple-cylinder solution, prescribed zero circulation and front-loaded assumptions.
+
+### Observed practice and manuscript recommendation by paragraph job
+
+1. **Late contextual return.** **Observed:** Cornejo Maceda et al. first establish the symmetric steady scan as a local branch map, then distinguish those states from later feedback results; Marra et al. associate boat-tailing and stagnation-point control with different objectives and actuators rather than collapsing them into one mechanism. **Recommend:** return to the steady seed in one sentence, immediately state the configuration break, and pose the narrow context question. Cite these pinball papers after the external comparator sentence, not after the imported SR/CCD observation. Avoid restarting the article's chronology or presenting a second control climax.
+
+2. **Separate steady endpoint.** **Observed:** Chan et al. define the two rotation directions and nondimensional rotation before comparing wake topology; their normal and reverse directions can both become steady but have different thresholds and structures. Cornejo Maceda et al. show that an apparently stabilized mean can hide residual shear-layer dynamics and use local-minimum language where the search does not prove globality. **Recommend:** define body order, wall-rotation sign and endpoint comparator before the deficit value. Order the evidence accepted endpoint → stationary unsteady reference → reverse non-amplitude-matched reference. Put `settling-qualified`, `endpoint` and the no-optimum/no-stability ceiling in the result paragraph itself. A caption identifies states, action signs, $s$, plane and metric; body explains why the accepted state is distinct.
+
+3. **Kinematic wall velocity versus viscous hypothesis.** **Observed:** Chan et al. establish no-slip and rotation direction before discussing virtual-body or separated-wake morphology. Ueda et al. connect prescribed rotation to viscous flow through the no-slip wall condition, but confine analytical conclusions to a stated low-$Re$ asymptotic domain. Li & Zhang reserve stabilization language for cases supported by a base-flow/stability test and report failures locally at stronger instability or insufficient sensing. **Recommend:** first state only the exact tangential wall motion and cardinal-point directions. Then enumerate observed endpoint fields. Introduce each arrow in the wall-shear/vorticity → shear-layer/inter-cylinder transport → separation/recirculation → downstream-transport chain as `may`, `could`, `is consistent with` or `motivates`; follow the chain immediately with the missing amplitude-matched intervention and closed-budget tests. Literature can motivate diagnostics, not certify the project mechanism.
+
+4. **Prescribed-circulation outer reference plus adjacent sign/closure failure.** **Observed:** Crowdy defines steady irrotational uniform flow, impermeable streamline boundaries and a chosen zero-circulation condition before showing streamlines. Sun & Ng likewise front-load impenetrability and no-vortex assumptions before deriving interaction results. Ueda et al. place the Jeffery-paradox and higher-order inertial limitations beside the analytical route. Across these papers, attractive mathematical figures do not silently acquire viscous no-slip status. **Recommend:** if Frank retains the multi-cylinder display, define $\boldsymbol u_N$, the harmonic circulation basis and prescribed periods before any streamline interpretation. In the same paragraph and visual composition, place (i) rotation invisibility to impermeability, (ii) one-circle no-slip incompatibility and (iii) the project sign failure. The surviving statement is geometrical/outer organization; the rejected statement is wall-spin-to-circulation closure. Do not place the failure in a later paragraph or caption footnote.
+
+5. **Compatible, non-equivalent synthesis and experiment bridge.** **Observed:** the closest JFM papers retain distinct branch/objective identities and use dedicated stability or asymptotic tests before stronger explanation. Negative routes narrow the valid envelope rather than being hidden after a visually successful case. **Recommend:** synthesize three non-equivalent items in order: Kármán persistent-plus-correction organization under its configuration; separate steady endpoint and viscous hypothesis; outer-reference utility constrained by failed closure/sign. State explicitly that compatibility is organizational, not continuation, validation or mechanism proof. End by asking what qualitative, open-loop observation can test or illustrate without repairing the missing intervention/budget evidence; do not preview an experimental result.
+
+### Wording ladder and local placement
+
+- **Green — licensed by this packet:** `separately acquired steady endpoint`; `settling-qualified`; `directly prescribed tangential wall velocity`; `the endpoint exhibits`; `is compatible with`; `motivates a viscous hypothesis`; `prescribed-circulation outer reference`; `geometrical organization`; `fails to close`; `opposite sign`; `compatible but non-equivalent`; `organizational rather than causal`.
+- **Amber — use only with an adjacent condition/test:** `steady branch` only for the observed/acquired endpoint, not a continued family; `stabilized` only with a dedicated perturbation or eigenvalue test; `gap jet` only after direction and field evidence; `separation change`, `vorticity production` and `transport pathway` as observations/hypotheses, not causal ownership; `outer approximation` only with its boundary conditions and finite-image status; `suggests` only when followed by the missing evidence.
+- **Red — delete from S8 body and captions:** `mechanism demonstrated`, `potential prediction`, `validated MFS mechanism`, `circulation selected by spin`, `inviscid cloak`, `momentum restored`, `base-bleed mechanism`, `boat-tail mechanism` for the accepted project branch, `optimal`, `stable branch`, `same control law`, `agreement`, `replication`, `continuation`, `validation across configurations`, or any quantitative experimental repair.
+
+### Citation, caveat, caption and bridge controls
+
+- Use a compact rotating-cylinder/pinball citation cluster after the external statement that opposite rotation directions can yield distinct steady organizations. Cite analytical papers immediately after defining the mathematical problem they actually solve; never attach them to the finite-$Re$ endpoint claim.
+- Put the no-intervention/no-budget caveat directly after the viscous chain. Put the no-slip incompatibility and sign failure in the same paragraph as the outer streamlines. This adjacency is the central style safeguard against theory being read as mechanism proof.
+- CFD caption: configuration, body order/action signs, endpoint/reference statuses, $s$, field/profile coordinates and metric. Theory caption: `prescribed-circulation outer reference`, circulation signs, boundary conditions, numerical-consistency status and explicit non-no-slip ceiling. Body: physical inference, failure and consequence. Neither caption should choose the unresolved equation/composition.
+- Preferred S8→S9 bridge job: ask what an author-owned qualitative open-loop comparison can make visible while remaining unable to establish forces, field matching, stability or mechanism. Preserve omission fallback if provenance does not close.
+
+### Anti-patterns and coverage
+
+- Avoid a detached theory subsection, streamline-led mechanism claims, a historical theory derivation, equation inventory in body prose, a single `steady` label for non-equivalent branches, action-direction decoding deferred to a caption, and a terminal limitations dump separated from the attractive figure.
+- `read`: the seven full PDFs listed above, with paragraph-, section-, equation- and figure-specific questions; DOI and PDF states confirmed in the DynamisLab Undermind workspace on 2026-08-11.
+- `observed practice`: branch-local comparators, condition-first interpretation, dedicated stability evidence, front-loaded mathematical assumptions, adjacent analytical failures and caption/body division.
+- `recommended practice`: the five paragraph orders, wording ladder, citation/caveat locations and bridge above; recommendation is a cross-paper synthesis, not a universal JFM rule.
+- `not covered`: no S8 equation/composition decision, new CFD/MFS calculation, endpoint promotion, mechanism/stability proof, figure generation, experiment observation or polished manuscript prose.
+
+## Round-5 author decision — 2026-08-11
+
+Frank accepted the multi-cylinder Hodge/period outer-solution main display with prescribed circulations, the adjacent one-circle rotating-no-slip incompatibility, and one integrated CFD + prescribed-circulation outer-reference + sign/closure-failure figure. This closes the equation/composition author-choice gate only. It does not approve equation artwork, exact source assets, provenance, caption, crop/scale, redraw, rendering, publication readiness, finite-$Re$ transfer or mechanism validation.

docs/JFM_WYQ/ROUND4_TOP_DOWN/SECTION_09_EXPERIMENT.md

@@ -0,0 +1,143 @@
+# Section 9 — Qualitative open-loop experiment
+
+## Status, authorship and scope
+
+**Status:** `author-interface`; R5 permits explicit author-field placeholder TeX only. This is not agent-authored manuscript prose, not a promoted experimental result, and not a section freeze. **Author:** Frank owns the substantive account, especially apparatus design, trial observations and final wording. **Provisional reader title:** **Qualitative open-loop experiment**. **Placement:** short main-text subsection near the end, following Section 8 and before the conclusion.
+
+**Purpose and evidence ceiling.** Reserve a compact interface for physical deployment and selected qualitative morphology observations. The present author-described materials are one apparatus schematic, one steady single-frame experimental/CFD pair, and several Illusion single-frame experimental/CFD pairs. Until exact files, trial identities, conditions, processing and permissions are supplied and audited, all remain `author-supply/provenance-gated`. A retained single frame may support only a named visible morphology observation. It cannot establish shedding frequency or any other temporal change, velocity, PIV, force, field matching, CFD validation, registration, quantitative agreement, robustness or repeatability. No experiment equation or quantitative results table is warranted.
+
+Section 8 exports only a boundary: synchronized quantitative experiment would be needed to cross the CFD evidence boundary. This section cannot repair the mechanism, closure or validation limits stated there.
+
+## Frank's author slots
+
+Supply only recoverable facts; mark unavailable fields explicitly rather than estimating them.
+
+- **Apparatus design:** [AUTHOR: functional schematic narrative; why the isolated test section, carriage, actuation and imaging path were designed this way].
+- **Facility and geometry:** [facility/test-section dimensions and boundaries; water state/preparation; cylinder diameter, spacing, span, blockage and coordinates; carriage travel and towing direction/speed; experiment-specific $Re_D$ definition if supported].
+- **Open-loop execution:** [software entry point actually used; command source; action units, body order/sign, physical conversion, ramp/start/stop timing; commanded versus measured motion; what was synchronized].
+- **Trial/case identities:** [steady reference/controlled trial IDs; selected Illusion trial IDs and named conditions; dates/setup version; repeat count; exclusion and frame-selection rule].
+- **Image provenance and observations:** [raw filename/hash, camera/view, scale, exposure, frame index/time, dye injection, crop/enhancement, permission; one exact morphology-only sentence per retained pair].
+- **Known limits/confounds:** [background/blank state, settling and repeat information; any calibration or synchronization absent]. Keep the full hardware/calibration inventory supplementary unless it changes how a shown image must be read.
+
+## Two–three paragraph job scaffold — Frank to write
+
+1. **Purpose, apparatus and execution boundary.** [In one paragraph, state why physical deployment is included; describe the apparatus by evidentiary function—facility/carriage, rotating pinball, command path and image acquisition—and identify open-loop execution. Do not turn this into a construction diary or import CFD performance values.]
+2. **Selected named morphology observations.** [For the steady pair and only the selected Illusion pair(s), identify condition, trial, frame and processing before describing visible streakline/wake morphology. Keep experimental and CFD views as corresponding named-condition images, not registered pairs. Do **not** state or imply that shedding frequency changed: a single frame contains no temporal estimator.]
+3. **Immediate ceiling and next test.** [State that these are qualitative open-loop, single-frame observations. Name the synchronized, calibrated, repeated force/PIV or declared image-sequence estimator needed for quantitative or temporal claims, then bridge to the short conclusion.]
+
+Paragraphs 1 and 3 may be compressed around one observation paragraph. No invented connective prose is supplied here.
+
+## Exact user-home CelerisLab inventory
+
+Inspection was read-only; no CFD, hardware, rendering or data generation was executed.
+
+| Exact path | What exists / actual capability | What it can inform | What it cannot establish |
+|---|---|---|---|
+| `/home/frank14f/CelerisLab/README.md` | Python CUDA-LBM API with runtime `sim.set_body(..., omega=...)`, action upload on `sim.run(...)`, and force/torque/sensor readback. | Generic CFD control-loop and body-action software path. | The physical apparatus execution path, a Raspberry Pi/motor deployment, or acquired/calibrated experiment telemetry. |
+| `/home/frank14f/CelerisLab/tests/postproc/experiment.ipynb` | Executed CFD notebook containing experiment-coordinate action sign conversion, three-body action ordering, analytic action generation, direct action upload, particle-streakline and vorticity rendering; embedded output is computational. | Candidate CFD counterpart workflow and sign/body-order questions Frank must reconcile. | Hardware control, physical trial provenance, or an experimental image. Notebook content/output is not a deployment script. |
+| `/home/frank14f/CelerisLab/tests/specs/exp_ctrl_matrix.md` | Recorded no-control, steady/`stealth`, and several `deceit` analytic signal-feature sets (means, cosine frequencies, amplitudes and phases). | Candidate identities for CFD named-condition counterparts after author mapping. | That these exact commands were played on the apparatus, or that repository slugs equal manuscript Illusion identities. |
+| `/home/frank14f/CelerisLab/tests/postproc/run_exp_ctrl_matrix_vorticity.py` | CFD batch runner: constructs the three-cylinder geometry, analytically evaluates the control matrix, applies a 5-s ramp, converts surface-speed commands to lattice angular velocity, swaps rear-body action mapping, runs simulation, and writes vorticity/checkpoint summaries. | Potential CFD counterpart generation path and explicit action playback logic in simulation. | Physical deployment/execution, policy playback, experiment matching, or a publication-ready counterpart. Do not run without authorization. |
+| `/home/frank14f/CelerisLab/tests/postproc/run_exp_ctrl_matrix_streakline.py` | CFD batch runner using the same analytic actions; continuously releases particles from step 20,000, samples velocity every 50 steps, renders at steps 40k/60k/80k/100k, and clears particles between early windows. | Potential single-frame CFD streakline counterpart generation and its sampling/render contract. | Dye physics, a registered experimental pair, temporal frequency measurement, or validation. Do not run without authorization. |
+| `/home/frank14f/CelerisLab/src/CelerisLab/common/streakline/{_streakline.py,_config.py,_integrate.py,_render.py}` | Passive particle streakline library: point/line/strip release; injection between sampled velocity frames; bilinear space and linear time interpolation; RK4 substeps; optional random-walk diffusion; domain/solid filtering; age-weighted density binning and Gaussian blur. | How a CFD streakline-like image is numerically produced and which processing fields must appear in a caption. | A calibrated dye concentration model, quantitative image estimator, experiment processing, or morphology agreement. |
+| `/home/frank14f/CelerisLab/docs/validation_specs/streakline_notes.md` | Design note distinguishing streaklines from streamlines/path-history plots and describing online/offline concepts and rendering choices. Some implementation-path examples use older filenames and must be checked against current package paths. | Terminology and disclosure of release, integration, diffusion and rendering choices. | Validation of the current experimental images or proof that offline replay/current output exists. |
+| `/home/frank14f/CelerisLab/tests/postproc/run_kan99b_streakline.py` | Online single-cylinder CFD streakline demonstration with configurable release/integration settings. | General test/example of the library. | A three-cylinder experiment counterpart or physical evidence. |
+| `/home/frank14f/CelerisLab/src/CelerisLab/body/action_smoother.py` | Generic linear/exponential action scaling utility. | Possible software concept for ramp disclosure. | Evidence that this utility was used in experiment or in the control-matrix scripts (which implement their own cosine ramp). |
+| `/home/frank14f/CelerisLab/tests/output/exp_ctrl_matrix_streak_nx1500/manifest.json` | Manifest records C0–C6 computational streakline PNG paths and four render steps per case; the referenced PNGs were not present in the inspected tree, while summaries/manifest remain. | Historical generated-output provenance and candidate regeneration route. | Located/audited image assets, experiment counterparts, or figure readiness. |
+| `/home/frank14f/CelerisLab/tests/output/exp_ctrl_matrix_vort_nx1500/manifest.json` | Manifest records C0–C6 computational vorticity PNG/checkpoint paths; the referenced PNG/HDF5 files were not present in the inspected tree, while summaries/manifest remain. | Historical CFD counterpart provenance. | Available publication assets or any experimental comparison. |
+| `/home/frank14f/CelerisLab/tests/output/stealth_steady_sweep_nx1500/manifest.json` | Manifest records a computational steady action sweep and streakline/vorticity/checkpoint paths; referenced assets were not present in the inspected tree. | Candidate steady CFD context after exact author mapping. | The author-described steady pair, experimental provenance, or agreement. |
+
+**Execution/deployment finding.** No Raspberry Pi, GPIO, serial, motor-driver, carriage/towing, or hardware-deployment script was located in the inspected `/home/frank14f/CelerisLab` tree or targeted user-home filename/content searches. The located execution code is CFD-side analytic action playback. Therefore the physical software execution path remains an author slot; code capability must not be reported as evidence that a given physical trial was executed.
+
+## Current material and figure interface
+
+- **Apparatus schematic:** author-described as existing; exact file not located. Role: configuration context only.
+- **Steady experimental/CFD single-frame pair:** author-described as existing; exact files not located/audited. Role ceiling: named visual morphology after provenance.
+- **Several Illusion experimental/CFD single-frame pairs:** author-described as existing; exact files not located/audited. Select only the pair(s) needed for the argument; all need case/trial/frame binding. Role ceiling: named visual morphology only.
+- **CFD manifests:** computational output records exist at the exact CelerisLab paths above, but referenced image assets were absent from the inspected tree. They are not automatically the author's corresponding CFD images.
+
+Detailed textual production plan: [`../FIGURE_PLANS/SECTION_09_EXPERIMENT.md`](../FIGURE_PLANS/SECTION_09_EXPERIMENT.md).
+
+## Integrated JFM style identities and reporting principles
+
+Literature supplies presentation practice only; it establishes no DynamisLab apparatus, observation or readiness. The former Style 09 selected these exact identities and read levels:
+
+1. Soumarup Bhattacharyya, Izhar Hussain Khan, Shivam Verma, Sanjay Kumar & Kamal Poddar, “Experimental investigation of three-dimensional modes in the wake of a rotationally oscillating cylinder”, *Journal of Fluid Mechanics* **950**, A10 (2022), DOI `10.1017/jfm.2022.792`, `full-PDF`.
+2. Izhar Hussain Khan, Puja Sunil, Soumarup Bhattacharyya, Rahul Yadav, Kamal Poddar & Sanjay Kumar, “Flow past two rotationally oscillating cylinders”, *Journal of Fluid Mechanics* **969**, A16 (2023), DOI `10.1017/jfm.2023.549`, `full-PDF`.
+3. Lukasz Klotz, Karol Bukowski & Konrad Gumowski, “Influence of porous material on the flow behind a backward-facing step: experimental study”, *Journal of Fluid Mechanics* **998**, A31 (2024), DOI `10.1017/jfm.2024.639`, `full-PDF`.
+4. Yu Zhou, Dewei Fan, Bingfu Zhang, Ruiying Li & Bernd R. Noack, “Artificial intelligence control of a turbulent jet”, *Journal of Fluid Mechanics* **897**, A27 (2020), DOI `10.1017/jfm.2020.392`, `full-PDF`.
+
+Use the compact sequence `purpose/boundary → apparatus, actuation and acquisition → named visual comparison → local measurement limit → next quantitative test`. Organize apparatus by evidentiary function. Give visualization its own contract (tracer, illumination, camera/view, field of view, frame selection and processing). State comparator and held conditions before morphology. Disclose when experimental and CFD views are separate realizations. Captions carry condition, view, frame, scale/calibration status and processing. Stronger temporal, field or force statements require a separate calibrated estimator. Main text selects representative comparisons rather than every condition; omit media if provenance fails.
+
+## Ownership, imports, exports and omissions
+
+**Imports:** Section 2 definitions only where Frank can bind experiment-specific geometry/action units; Section 7 supplies named Illusion terminology but no metric or positive SR claim; Section 8 supplies only the quantitative-boundary bridge. Computational results do not validate the apparatus.
+
+**Owns:** Frank's apparatus account, physical open-loop execution, trial/media provenance, exact morphology observations, local confounds, figure selection and the next experimental threshold.
+
+**Exports:** no result at present. If Frank later supplies a provenance-passing qualitative observation, the coordinator may decide after a new targeted provenance audit and explicit coordinator reopening decision whether one bounded qualitative sentence enters Section 10 D3 or C1; it may contain no number, agreement or validation wording. Otherwise omit experimental observation from Section 10.
+
+**Omit/prohibit:** shedding-frequency change from single frames; time-resolved-video claims; quantitative image estimator; CFD values; L2/DTW; PIV/velocity/vorticity inference from dye; force reduction; registered overlays; agreement/validation language; closed-loop experimental control; policy validation; mechanism, stability, optimum, robustness or generality. Keep brands, complete hardware/calibration tables and recoverability details supplementary unless needed to read a retained panel.
+
+## Compact provenance slots and quality gate
+
+For every retained experimental or CFD image bind: `[raw exact path/hash] [trial/case and role] [condition/action/body order/sign] [frame index/time or CFD step] [camera/view/FOV/scale] [dye/release protocol] [crop/enhancement/render processing] [selection/repeats] [permission] [missing fields]`.
+
+**Pass only if:**
+- Frank has written the apparatus/execution and observation text;
+- every shown image has exact source, condition, trial/case, frame/step, processing and permission;
+- each observation names visible morphology only and its comparator is local;
+- no sentence infers frequency or temporal behavior from one frame;
+- CFD and experiment are labelled separate, unregistered named-condition views;
+- open-loop and no-force/PIV/no-quantitative-estimator limits sit beside the observation;
+- absent provenance triggers pair omission, not reconstruction or digitization.
+
+## Coverage ledger
+
+| State | Material | Use/boundary |
+|---|---|---|
+| `read` | Round-4 global ledger, display inventory, former Style 09; Round-3 Section 9 interface; Round-4 Section 8 bridge | Architecture, ownership, former style identities and ceilings; read is not independent experiment verification. |
+| `read` | Exact CelerisLab code/docs/manifests listed above | Software capability and historical computational-output inventory; no execution or artifact regeneration. |
+| `mapped` | Author-described apparatus schematic and experimental/CFD pairs | Current-material slots only; exact files not located or audited. |
+| `not-covered` | New CFD/hardware/rendering/data generation; broad project rediscovery; exhaustive home-tree audit | Outside authorization/need; no inference. |
+| `unresolved` | Physical deployment scripts/logs, raw experiment media, apparatus schematic path, trial linkage, conditions, calibration/synchronization, repeats and permissions | Frank must supply or mark unavailable. |
+
+**Prior packet memory use:** targeted Nowledge recall was used for prior decisions, but the current author correction supersedes recalled frequency wording. At that packet stage, former Style 09's already-read literature identities were integrated without a new search. The dated enrichment below records the later library search and full-PDF reads. No thread recall or Working Memory was used.
+
+## Paragraph-level literature/style enrichment — 2026-08-11
+
+**Scope of enrichment.** This is author-facing reporting guidance only. It neither fills Frank's slots nor supplies an apparatus fact, trial condition, observation, experimental claim or promoted result. Three close JFM experimental papers were read at full-PDF level: S. Bhattacharyya *et al.*, “Experimental investigation of three-dimensional modes in the wake of a rotationally oscillating cylinder”, *JFM* **950**, A10 (2022), DOI `10.1017/jfm.2022.792`, `full-PDF`; I. H. Khan *et al.*, “Flow past two rotationally oscillating cylinders”, *JFM* **969**, A16 (2023), DOI `10.1017/jfm.2023.549`, `full-PDF`; and Y. Zhou *et al.*, “Artificial intelligence control of a turbulent jet”, *JFM* **897**, A27 (2020), DOI `10.1017/jfm.2020.392`, `full-PDF`. Literature establishes reporting practice only, not DynamisLab evidence.
+
+### Observed sample-bounded practice
+
+- The experimental narrative runs `purpose/question → facility and model → actuation and sensing/acquisition → processing or estimator → declared comparator → morphology or measured result`. Apparatus detail is grouped by what supplies, moves, actuates, images or measures rather than by construction chronology.
+- Methods use past tense for actions actually performed; figure-led observations and definitions generally use present tense. Visual evidence uses shape/location language (`visible`, `streakline`, `wake outline`, `near-wake`, `downstream`, `broader/narrower`, `deflected`, `separated`) while frequency, velocity, circulation and force language appears only after a named estimator or instrument.
+- Comparator conditions precede interpretation. The experimental papers separate flow visualization from hot-wire, PIV, spectral or load-cell evidence; quantitative diagnostics corroborate claims that visualization alone does not carry.
+- Acquisition and uncertainty information is placed beside the relevant method or result. Captions identify condition, orientation, scale/markers and panel-specific acquisition; body text supplies the inference and its local qualification.
+- The close papers often contain much more hardware specification than this short interface needs. That density is an observed option, not a recommendation here; the manuscript should retain only details needed to recover or correctly read the shown evidence.
+
+### Recommendation by paragraph job
+
+1. **Purpose and deployability.** Open with the narrow reason for including the physical realization, then state `[AUTHOR: what was physically deployed and what open-loop role the trial serves]`. Use `implemented`, `commanded`, `imaged` or `recorded` only after Frank binds each verb to a recoverable source. Do not call the trial a validation, reproduction or closed-loop demonstration.
+2. **Apparatus by evidentiary function.** Order placeholders as `[facility and background state] → [moving/holding geometry] → [rotation command and physical conversion] → [image path]`. Compress brands and construction history. Place each dimension or setting only where it changes the interpretation of a retained image.
+3. **Acquisition and provenance.** Before any morphology slot, require `[trial/case identity] [raw file/hash] [view/FOV/scale] [frame index/time] [dye/release] [exposure] [crop/enhancement] [selection rule/repeats] [permission]`. Use past tense only after these fields are supplied. If they remain unavailable, omit the pair rather than soften the provenance language.
+4. **Local morphology comparator.** Write the comparator first: `[AUTHOR: experimental condition A versus condition B, or experimental named condition versus separate CFD named-condition view]`; then allow one `[AUTHOR: directly visible shape/location observation]`. Keep `corresponding named-condition views` and `separate realizations`; prohibit `registered`, `agrees`, `matches`, `reproduces`, `confirms`, `suppresses shedding` or any frequency-change wording from a retained single frame.
+5. **Measurement ceiling and next threshold.** Put the caveat in the same paragraph as the observation: `[AUTHOR: qualitative, open-loop, single-frame ceiling and local confound]`. Then name only the measurement threshold required to support a stronger class of claim: synchronized calibrated force/PIV for force or field claims, repeated image sequences plus a declared temporal estimator for frequency claims. This is a local threshold, not a programme of promised results.
+
+### Caption/body and bridge controls
+
+- **Caption owns:** exact condition and role; experiment versus CFD identity; unregistered/separate-realization status; trial/case and frame/step; view, scale/calibration status and orientation; dye or numerical release; crop/enhancement/rendering; selection/repeat status; and any missing provenance that changes readability.
+- **Body owns:** why the physical interface matters; one comparator-led morphology statement per retained pair; the immediately adjacent evidence ceiling; and the consequence for what remains unclaimed. Do not narrate every panel or repeat caption metadata.
+- **Incoming bridge from Section 8:** `[INSTRUCTION: move from the CFD evidence boundary to the narrower question of physical command-and-image deployability; do not imply that experiment repairs the theory or validation boundary.]`
+- **Outgoing bridge to Section 10:** `[INSTRUCTION: close on the exact measurement ceiling and declared next evidentiary threshold. Do not export an observation to Section 10 unless the existing targeted provenance audit and explicit coordinator-reopening gate is later passed.]`
+
+### Lexicon and anti-patterns
+
+**Use after author supply:** `qualitative open-loop trial`, `retained frame`, `visible morphology`, `separate named-condition view`, `command path`, `image acquisition`, `frame selection`, `processing disclosed`, `measurement ceiling`, `would require`.
+
+**Avoid:** construction diary; equipment catalogue; CFD performance imported into apparatus prose; an observation before its comparator/provenance; `appears to validate`; single-frame temporal inference; PIV/velocity/vorticity language for dye streaklines; `agreement` for unregistered views; caveats deferred to the paragraph end or section end; generic `proof of feasibility`; promises of future validation.
+
+**Coverage:** `memory-recalled` — three narrow Nowledge searches; `read` — current Section-9 packet and controlling Round-4 ledgers; `full-PDF` — the three exact JFM papers above, each queried for apparatus/acquisition/morphology/caption/caveat practice; `mapped` — author-described experimental materials only, with no independent audit; `not-covered` — experiment media, physical deployment logs, new observations, CFD or artifact generation. The literature observations above are sample-bounded; recommendations are manuscript-specific.
+
+## Round-5 placeholder decision — 2026-08-11
+
+Frank permits a Section-9 TeX output containing clearly marked `AUTHOR FIELD` placeholders only. The placeholder may encode the existing paragraph jobs and requested fields but may not assert apparatus facts, execution details, trial identities, command provenance, media provenance, morphology observations or experimental results. This closes only the R5 output-format choice. Every author-supply, command-map, media/provenance, permission and result-import gate remains open; omission remains the fallback.

docs/JFM_WYQ/ROUND4_TOP_DOWN/SECTION_10_DISCUSSION_CONCLUSION.md

@@ -0,0 +1,230 @@
+# Round 4 integrated architecture — Discussion and conclusion
+
+## 1. Status and scope
+
+**Project/unit:** DynamisLab JFM manuscript, Round 4, Section 10 only.  
+**Status:** `frozen`; integrated author-facing architecture built after rereading the current Sections 1–8. The current author-owned `SECTION_09_EXPERIMENT.md` was read afterward for placeholder/interface status only. This is not manuscript prose, a new scientific audit, or a limitations/outlook design.  
+**Recommended reader-facing title:** **Discussion and conclusions**, with two short subunits: **Discussion** and **Conclusion**.
+
+**Primary role.** This final section summarizes and synthesizes the established manuscript answers. It does not reopen evidence, build a gap ledger, or convert local qualifications into a limitations programme. The current architecture is text-only: no new figure, evidence map, limitations table, equation, number, citation, or panel reference.
+
+**Current exclusions.** Limitations and future work are reserved for Frank and are outside this freeze. Only retained manuscript results enter Section 10. Vortex/Erase data inventories, experiment-provenance problems, theory test wish lists, and other former gap-ledger material are not summary content. Current Section 9 is an author-owned placeholder/interface that was read after the Section-10 synthesis architecture was built; no experimental result or observation is imported, and at most a neutral author-owned transition may be inserted later.
+
+## 2. Summary question and bounded answer
+
+**Question.** What does the paper establish when its breadth of learned rotational control, independent Kármán field evaluation, and action/field interpretation are read together under their distinct physical definitions?
+
+**Bounded answer.** The manuscript's physical progression begins with a prescribed steady endpoint and background recovery. Across the retained PPO tasks, force feedback plus sparse downstream velocity observations enables learned rotational control to approach periodic, event-relative, and declared non-zero-target hydrodynamic references under their own configurations, roles, clocks, and evaluation definitions. The periodic Kármán cloak in the parabolic-inflow/no-slip-wall channel configuration is the sole deep case: independent registered-field comparisons show that the controlled wake is substantially closer to its target than physical zero, and compatible but non-equivalent action-side and field-side analyses organize the result around persistent rear counter-rotation plus smaller corrections. The separate steady endpoint and failed outer closure provide physical context and a boundary on interpretation; they do not validate the periodic result or establish a causal mechanism.
+
+## 3. Current author decisions controlling Section 10
+
+1. Section 10 is primarily summary/synthesis, not a late limitations inventory.
+2. Retain a distinct short Conclusion after the Discussion.
+3. Omit all current limitations and future-work design. Frank may add those later without changing the frozen summary architecture.
+4. Delete the former D3 envelope/limitations, D4 decisive-tests, and C3 strongest-boundary/future-test jobs. Do not replace them with another research programme.
+5. Include only retained manuscript results in synthesis, boundaries, and conclusion.
+6. Do not import cross-case missing-data inventories, experimental provenance issues, theoretical test lists, or other gap ledgers as final-section content.
+7. Use only current Sections 1–8 as the scientific narrative authority. Read current Section 9 for author-owned placeholder/interface status only; it supplies no result or observation to the synthesis. The Round-3 final module is superseded provenance and boundary evidence only.
+8. Add no new display, table, equation, numerical value, citation, or scientific result.
+
+## 4. Compact reader-facing architecture
+
+### 10.1 Discussion — integrated synthesis (three paragraphs)
+
+#### D1 — Breadth to depth: capability first, then the sole deep case
+
+**Job.** Re-establish the paper's inferential order without replaying sections or values.
+
+**Required moves.**
+- Begin with the controlled object and task: three cylinder rotations use force feedback and sparse downstream velocity observations to approach a declared reference.
+- Distinguish the prescribed steady background-recovery endpoint from learned capability. The retained PPO tasks demonstrate periodic Kármán cloaking, an event-relative incoming-vortex response, and changed-target control under distinct configurations and task definitions.
+- State immediately that these results are not pooled and do not share one efficacy score or interpretation contract.
+- Narrow to the periodic Kármán cloak in the parabolic-inflow/no-slip-wall channel configuration as the sole case combining repeated learning evidence, signal histories, independent registered fields, executable policy reduction, and corrected field analysis.
+- Import Section 4's conclusion, not its numbers: all five tested stochastic realizations reach high-performing checkpoints in the selected uniform-inflow/free-slip-wall condition, while the separate parabolic-inflow/no-slip-wall reference case shows a controlled registered field substantially closer to target than physical zero in both complete-cycle mean and phase-resolved comparisons. Reward, DTW, and spatial field error retain their distinct meanings.
+
+**Claim ceiling.** This paragraph establishes named breadth and asymmetric depth. It does not imply pooled robustness, transfer, universal disturbance control, configuration continuation, or that signal matching alone proves field matching.
+
+**Bridge.** Once independent field improvement is established for the deep Kármán case, the synthesis can ask what recurring organization is visible in its actions and fields.
+
+#### D2 — Action and field organization in the deep Kármán case
+
+**Job.** Integrate Sections 5–6 into one bounded control-skeleton reading while preserving their different estimands.
+
+**Required moves.**
+- On the action side, summarize the executable compact law as persistent rear counter-rotation with smaller rear-lift modulation and a weak tested front correction. Closed-loop deployment and parent-relative term deletion support this ranking among the tested terms; do not replay the formula, coefficients, windows, or deletion values.
+- On the field side, summarize the passive mean streamwise deficit and the largely opposite-signed controlled addition relative to physical zero. State only that this is compatible with adding downstream streamwise velocity where the passive pinball produces a deficit; do not call it momentum restoration, cancellation, or a closed budget.
+- State that the constant reference accounts for most of the measured scalar mean-field error reduction, while a smaller remaining residual is organized descriptively through action coordinates.
+- Name **action-correlated correction-field modes** once if needed for terminological continuity, with the adjacent noncausal boundary. POD supplies the same-object modal boundary; it is not an inferior baseline.
+- Conclude that action-side and field-side results are compatible but non-equivalent. They support a bounded control-skeleton organization—persistent rear counter-rotation plus smaller correction—not causal or independent confirmation, a unique decomposition, or a governing law.
+
+**Near-collision control.** Section 4's standardized PPO field evaluation, Section 5's SR deployment, and Section 6's corrected four-role mean accounting remain different acquisitions/estimands. Their numerical proximity or conceptual compatibility cannot be described as replication, agreement, validation, closure, or triangulation.
+
+**Bridge.** The deep-case organization can now be placed beside the changed-task demonstrations and the separate steady endpoint without transferring its interpretation to them.
+
+#### D3 — Changed tasks and steady physical context
+
+**Job.** Close the Discussion positively by locating the deep interpretation within the manuscript's wider physical range.
+
+**Required moves.**
+- Summarize Section 7 at the level now supported by the current architecture: replacing periodic phase with an event-relative incoming-vortex problem demonstrates controlled response under a changed temporal definition, while the retained Illusion PPO case demonstrates retargeting toward a declared non-zero reference. Omit Erase from the established synthesis because its role-field material is not yet quantitatively bound.
+- State that these changed temporal and target tasks demonstrate breadth but do not inherit the Kármán control-skeleton interpretation.
+- Return briefly to Section 8's separate uniform-inflow/free-slip-wall steady endpoint. The accepted prescribed-rear-wall-motion case is a low-deficit mean-flow state and supplies physical context for why a persistent rear-rotation component is plausible as an organizing feature.
+- Keep the relation explicitly compatible and non-equivalent: the steady endpoint is neither continuation nor validation of the periodic result.
+- Close with the theory boundary as an established contextual finding, not a future-work prompt: the prescribed-circulation outer family organizes streamlines, but circulation is not selected by wall spin and the fitted/optimal sign mismatch defeats the proposed inviscid closure. The surviving synthesis is a bounded viscous, kinematics-first context rather than a causal mechanism.
+
+**Section-9 handling.** No Section-9 result is imported. If Frank later supplies a provenance-passing qualitative observation, the coordinator may decide after a new targeted provenance audit and explicit coordinator reopening decision whether one bounded qualitative sentence enters D3 or C1; no number, agreement or validation wording is allowed. Otherwise omit experimental observation. A navigation-only author transition may still be inserted outside this architecture.
+
+### 10.2 Conclusion — exact established answer (two short paragraphs)
+
+#### C1 — Capability and independent evaluation answer
+
+**Job.** Answer the paper question directly, without methods inventory, numbers, citations, or figure references.
+
+**Required content.** The prescribed steady endpoint establishes the historical background-recovery starting point. Across the retained PPO tasks, learned rotational control using forces and sparse downstream observations approaches periodic, event-relative, and declared non-zero-target hydrodynamic references under their own physical definitions. For the periodic Kármán cloak in the parabolic-inflow/no-slip-wall configuration, independent registered-field evaluation establishes that the controlled wake is closer to the target than physical zero; this is the paper's strongest field-level answer.
+
+#### C2 — Bounded interpretation and conceptual close
+
+**Job.** State the strongest integrated interpretation once and close on the paper's governing concept.
+
+**Required content.** In the deep Kármán case, executable action reduction and corrected mean/residual field organization support a bounded control skeleton consisting of persistent rear counter-rotation plus smaller corrections. These analyses are compatible but non-equivalent and remain descriptive rather than causal. Close by returning to hydrodynamic cloak and illusion as approach to a declared background or non-zero reference for specified observations and comparisons—not universal invisibility, configuration-independent transfer, or an unrestricted physical law.
+
+**Optional author-only insertion after C2, outside the frozen summary architecture:** `[AUTHOR PLACEHOLDER — limitations and outlook to be supplied later.]`
+
+## 5. Detailed synthesis and style controls
+
+### Evidence order and paragraph discipline
+
+- Preserve `prescribed steady starting point and learned-task breadth → sole deep-case field evaluation → action/field organization → changed-task breadth and steady context → compressed conclusion`.
+- Give each paragraph one inferential job. Do not produce one sentence per manuscript section or an abstract replay.
+- Use concrete physical subjects: controller, rotations, wake, field, target, physical zero, constant reference, wall motion, circulation. Avoid package names, internal provenance labels, and governance vocabulary in reader-facing prose.
+- State the comparator before the inference. A declared reference and physical-zero role must remain legible wherever field matching is summarized.
+- Use present tense for established interpretation and definitions; past tense only for what was performed. No future tense is needed in the frozen architecture.
+- Put the non-equivalence boundary in D2 beside the integrated action/field claim. Do not defer it to a caveat dump.
+
+### Strength and terminology
+
+**Preferred:** `demonstrates under the stated configuration`, `is closer to the declared target`, `supports`, `is compatible with`, `organizes`, `accounts for most of the measured scalar reduction`, `provides physical context`, `bounded control skeleton`, `action-correlated correction-field modes`.
+
+**Prohibited without new authority:** `causes`, `mechanism`, `restores momentum`, `cancels`, `confirms independently`, `triangulates`, `necessary`, `sufficient`, `unique`, `optimal`, `stable`, `robust`, `general`, `universal`, `transferable`, or any configuration-continuation claim.
+
+Use `parabolic-inflow/no-slip-wall channel configuration` and `uniform-inflow/free-slip-wall channel configuration` when configuration identity materially limits a sentence. Internal chain labels remain provenance-only. Spell `Kármán` exactly. The Conclusion should not repeat configuration detail more than needed to locate the sole deep case.
+
+### Compression and duplication deletion
+
+- Section 2 owns configurations, roles, action definitions, objective, DTW, spatial L2, ROI, masks, windows, and CFD qualification.
+- Section 3 owns the visual progression and first breadth statement.
+- Section 4 owns learning-range detail, exact rewards, L2/DTW values, field panels, and representative trajectories.
+- Section 5 owns the reflection map, symbolic law, coefficients, deployment details, and deletion values.
+- Section 6 owns four-role values, percentages, residual equation, CCD/POD construction, and mode displays.
+- Section 7 owns case-local Vortex/Illusion/Erase roles, event offsets, readiness, blocked quantitative-efficacy status, and strict non-transfer detail.
+- Section 8 owns steady values, wall kinematics, outer equations, sign values, and theory derivation.
+- Section 10 imports only result-level meanings and states the integrated answer once. Delete numerical replay, formulas, panel narration, source paths, acquisition histories, figure-readiness notes, and detailed nonclaim inventories from reader-facing prose.
+
+## 6. Exact selected JFM rhetorical identities and retained rationale
+
+External literature supports final-section organization only. It establishes no DynamisLab result, claim, validity, or artifact readiness. The following identities and read levels are the unique necessary record retained from the superseded style memo:
+
+1. N. Deng, B. R. Noack, M. Morzyński & L. R. Pastur (2020), “Low-order model for successive bifurcations of the fluidic pinball”, *Journal of Fluid Mechanics* **884**, A37. DOI `10.1017/jfm.2019.959`. `full-PDF`; final sections read for rhetorical architecture, not result reproduction.
+2. G. Y. Cornejo Maceda, Y. Li, F. Lusseyran, M. Morzyński & B. R. Noack (2021), “Stabilization of the fluidic pinball with gradient-enriched machine learning control”, *Journal of Fluid Mechanics* **917**, A42. DOI `10.1017/jfm.2021.301`. `full-PDF`; Discussion/Conclusion and closing synthesis displays read rhetorically, not result reproduction.
+3. Ji-chao Li & Mengqi Zhang (2022), “Reinforcement-learning-based control of confined cylinder wakes with stability analyses”, *Journal of Fluid Mechanics* **932**, A44. DOI `10.1017/jfm.2021.1045`. `full-PDF`; final physics/control integration read rhetorically, not result reproduction.
+4. J. Rabault, M. Kuchta, A. Jensen, U. Réglade & N. Cerardi (2019), “Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control”, *Journal of Fluid Mechanics* **865**, 281–302. DOI `10.1017/jfm.2019.62`. `full-PDF`; interpretation and compact Conclusion read rhetorically, not result reproduction.
+
+**Observed sample-bounded practice retained.** Structural labels vary, but interpretation is completed before the shortest closing unit; final synthesis follows the inferential order earned in the body; and a closing display is optional. For this manuscript, several non-equivalent evidence streams justify a distinct Discussion, while the Conclusion should compress only established answers. The former style memo's recommendations to organize a limitations envelope, decisive tests, a future-facing third conclusion paragraph, a limitations table, or an evidence map are superseded by the current author instruction and are intentionally not retained.
+
+## 7. Imports, ownership, and boundaries
+
+**Imports from current Sections 1–8; Section 9 interface read:**
+- Sections 1–2: declared-reference definitions, force-plus-sparse-observation execution, independent field evaluation, physical configuration names, and the three-level inferential order.
+- Sections 3–4: prescribed steady profile recovery as the physical starting point; retained PPO breadth under local task definitions; repeated high-performance learning in the tested five-realization condition; and independent mean/phase field improvement in the separate deep reference case.
+- Section 5: executable compact-law organization and tested persistent-rear-term ranking.
+- Section 6: passive deficit/opposite-signed controlled addition, constant-reference dominance in measured scalar accounting, smaller action-correlated residual, and strict noncausal/non-equivalent boundary.
+- Section 7: event-relative Vortex role material and positive Illusion PPO capability as bounded breadth, with Vortex/Erase quantitative efficacy blocked and strict non-transfer of Kármán interpretation.
+- Section 8: separate steady low-deficit endpoint, kinematics-first viscous context, prescribed-circulation outer organization, and failed wall-spin closure.
+- Section 9: current author-owned placeholder/interface status only; no experimental result or observation is imported.
+
+**Section 10 owns:** the compact integrated answer, the one-time action/field synthesis, the contextual placement of changed tasks and steady theory, and the two-paragraph conceptual close.
+
+**Out of scope now:** limitations architecture, future programme, experimental synthesis, Section-9 claims, raw-evidence rediscovery, new diagnostics, and any display/equation/citation production.
+
+## 8. Display, equation, citation, and placeholder controls
+
+- **Figure:** none.
+- **Evidence map:** omitted; former proposal superseded.
+- **Limitations table:** omitted; former proposal superseded.
+- **Equation:** none.
+- **New number:** none in reader-facing Section 10.
+- **Citation:** none in reader-facing Section 10. The records in §6 are author-facing style provenance only.
+- **Existing figure references:** omit unless later drafting finds one indispensable; safe default is no figure calls.
+- **Author placeholders:** allowed only after the frozen synthesis architecture and visibly marked as author-owned. They must not be populated by an agent without new instruction.
+
+## 9. Coverage ledger
+
+`read` means the current repository document was inspected for architecture and boundaries; it does not mean independent scientific verification or artifact audit. `mapped` is not `read`.
+
+- `memory-recalled`: three targeted Nowledge searches for manuscript-wide Round-4 decisions, Sections 3–8 interpretation boundaries, and final-section literature/style decisions. Memory guided navigation only; current repository authorities controlled every operative choice.
+- `read`: current Round-4 Sections 1–8 in full; current `SECTION_09_EXPERIMENT.md` for author-owned placeholder/interface status only; former Section-10 style memo; Round-3 `10_LIMITATIONS_CONCLUSION.md` as superseded provenance/boundary; Round-4 global decision and display/equation ledgers; Round-3 module interface ledger; Round-4 style crosswalk.
+- `mapped`: underlying authorities, manifests, and artifacts already owned by Sections 1–8; no broad raw-evidence rediscovery.
+- `not-covered` at freeze: raw fields/models; new CFD/training/SR/CCD/theory analysis; experiment media and their provenance audit; figures and manuscript prose. The dated enrichment below adds rhetorical full-PDF reads only and changes no scientific coverage.
+- `external-inaccessible`: none required for this architecture.
+
+## 10. Quality gate and disposition
+
+- [x] Rebuilt from current Sections 1–8 rather than the Round-3 limitation inventory.
+- [x] Current Section 7 exists and was read; no fallback to Round-3 M07 was needed.
+- [x] Current Section 9 was read as author-placeholder/interface only; no experimental result or observation enters the summary.
+- [x] Discussion has three compact synthesis jobs: breadth-to-depth, deep-case action/field organization, and changed-task/steady context.
+- [x] Conclusion has two short jobs and summarizes only established answers.
+- [x] Only retained manuscript results enter every Section-10 unit.
+- [x] No limitations architecture, future programme, strongest-boundary/future-test close, or substitute outlook was created.
+- [x] No Vortex/Erase missing-data inventory, experiment-provenance ledger, theory wish list, or other gap inventory enters the summary.
+- [x] Section 4/5/6 estimands remain non-equivalent; no replication, validation, agreement, closure, or triangulation wording.
+- [x] Cross-case breadth does not inherit Kármán interpretation; steady/theory remains context and boundary, not validation.
+- [x] No new figure, evidence map, limitations table, equation, number, citation, result, or artifact claim.
+- [x] Exact selected JFM identities, DOIs, read levels, and the still-relevant split/evidence-order rationale are integrated.
+- [x] Former limitations/future-work recommendations are explicitly recorded as superseded.
+- [x] Ownership and duplication deletion are explicit; the former style memo may be deleted.
+
+**Disposition:** `frozen`; coordinator second audit passed this text-only Section-10 architecture, which is Round-5 eligible. Pair A/E scientific content remains unchanged. Frank's later limitations/outlook may be added as a separate author-owned unit without reopening the frozen summary sequence unless it changes an established answer.
+
+## 11. Paragraph-level literature/style enrichment — 2026-08-11
+
+**Scope of enrichment.** This dated addition refines rhetoric only; it does not reopen the frozen `3 + 2` architecture or alter any scientific status, ownership, display, equation, control or claim. Four close JFM final sections were read at full-PDF level: N. Deng *et al.*, “Low-order model for successive bifurcations of the fluidic pinball”, *JFM* **884**, A37 (2020), DOI `10.1017/jfm.2019.959`, `full-PDF`; G. Y. Cornejo Maceda *et al.*, “Stabilization of the fluidic pinball with gradient-enriched machine learning control”, *JFM* **917**, A42 (2021), DOI `10.1017/jfm.2021.301`, `full-PDF`; J.-c. Li & M. Zhang, “Reinforcement-learning-based control of confined cylinder wakes with stability analyses”, *JFM* **932**, A44 (2022), DOI `10.1017/jfm.2021.1045`, `full-PDF`; and J. Rabault *et al.*, “Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control”, *JFM* **865**, 281–302 (2019), DOI `10.1017/jfm.2019.62`, `full-PDF`. These identities remain planning metadata only: reader-facing Section 10 still contains no citations.
+
+### Observed sample-bounded practice
+
+- Final sections compress around inferential or physical themes, not one sentence per preceding section. Control outcome is established before strategy interpretation; richer modelling, stability or flow analysis is then used for a distinct interpretive job.
+- Study actions are commonly past or present-perfect (`was evaluated`, `we have shown`), whereas established definitions and bounded interpretations use present tense (`supports`, `is compatible with`, `organizes`). Prospective tense belongs to outlook; it is unnecessary in this frozen no-outlook close.
+- Multiple evidence streams are often placed in a hierarchy or sequence. The close papers do not provide a universal wording rule for non-equivalence, but their separation of performance, actuation organization and flow/stability analysis supports making each evidentiary role explicit before synthesis.
+- Strong interpretive vocabulary in the sample is frequently supported by dedicated stability, intervention or quantitative analysis. Local qualifications occur next to the affected inference rather than in a terminal caveat list.
+- Closing units are shorter than the preceding synthesis and add no evidence. Some sampled papers end prospectively and some with a conceptual/application return; neither move requires replaying values. For this manuscript, the frozen declared-reference close is the relevant reusable pattern, while the sampled future-work closes are not.
+
+### Five-paragraph rhetorical controls
+
+- **D1 — breadth to depth.** Use one opening sentence for the controlled object and declared-reference task, one compressed breadth sentence, then narrow explicitly to the sole deep Kármán case. Put each comparator before its inference and preserve the evidence order `task definitions → retained breadth → deep-case independent registered-field result`. State non-pooling beside the breadth claim. Delete values, acquisition histories and section labels.
+- **D2 — compatible, non-equivalent synthesis.** Begin from the independently established deep-case result, then place action-side organization before field-side organization. Join them with `is compatible with` or `together support a bounded organization`, immediately followed by `non-equivalent` and `descriptive rather than causal`. Do not use the sampled stronger verbs (`discovers`, `stabilizes`, `confirms`) unless the owning DynamisLab evidence independently licenses them; it currently does not license causal or independent-confirmation wording.
+- **D3 — changed-task breadth plus steady context.** Move outward from the deep case: changed temporal definition, changed declared target, then the separate steady endpoint and theory boundary. Use `demonstrates under its stated task definition` for retained breadth and `provides physical context` for the steady relation. Keep the non-transfer and non-continuation boundaries local. End on the surviving bounded physical context, not on missing tests, a limitations inventory or Section 9.
+- **C1 — exact answer in evidence order.** Answer with `prescribed starting point → retained learned capability → strongest independent field-level result`. Use present tense for what the manuscript establishes and past tense only if a performed evaluation must be named. No methods list, values, citations, figure calls or new qualification should appear here; all needed qualification must already be embedded in the nouns and comparators.
+- **C2 — declared-reference close.** State the bounded control-skeleton interpretation once, place compatibility/non-equivalence and noncausality immediately beside it, and close by defining cloak/illusion through approach to a declared reference for specified observations and comparisons. The last sentence must contain no new result, example, citation, number, test or future promise.
+
+### Compression, lexicon, caveat and citation placement
+
+**Compression test:** each Discussion paragraph must have one governing comparison and one consequence; C1 answers `what was established`, C2 answers `how it is bounded and what concept survives`. Delete manuscript chronology, method inventory, repeated configuration detail, repeated sensor/reward definitions, metric names where no distinction is being made, and serial `first/second/finally` scaffolding.
+
+**Preferred register:** `establishes`, `demonstrates under the stated definition`, `is closer to the declared target`, `supports`, `is compatible with`, `organizes`, `provides physical context`, `bounded`, `distinct`, `non-equivalent`, `descriptive rather than causal`.
+
+**Reject:** `proves`, `validates`, `confirms independently`, `triangulates`, `causes`, `reveals the mechanism`, `restores`, `cancels`, `generalizes`, `transfers`, `unifies`, `universal`, `robust` or `optimal`; also reject ornamental intensifiers and sample-derived promotional language.
+
+**Caveat placement:** D1 carries non-pooling beside breadth; D2 carries estimand non-equivalence and noncausality beside the action/field synthesis; D3 carries non-transfer beside changed tasks and non-continuation beside steady context; C2 carries the declared-reference ceiling inside the close. Do not collect these into a limitations paragraph.
+
+**Citation placement:** none in reader-facing D1–D3 or C1–C2. The four identities above document author-facing rhetorical observation only. No literature citation may be inserted to make an internal result appear stronger or to decorate the final sentence.
+
+### Bridges and anti-patterns
+
+- **D1 → D2:** independent field improvement makes the action/field organization question legitimate.
+- **D2 → D3:** the bounded deep-case organization is placed beside wider tasks without transfer.
+- **D3 → C1:** finish the Discussion on established physical context so the Conclusion can answer rather than qualify.
+- **C1 → C2:** move from the strongest evidence-ordered answer to its single bounded interpretation and conceptual definition.
+
+**Anti-patterns:** abstract replay; one sentence per section; values or panel narration; action and field analyses described as mutual confirmation; steady endpoint described as validation or continuation; experimental placeholder imported as evidence; erased historical routes named in the synthesis; limitations or future-work programme; a third conclusion paragraph; citation-bearing close; new result or stronger verb in the last sentence.
+
+**Coverage:** `memory-recalled` — three narrow Nowledge searches; `read` — current frozen Section-10 packet and controlling Round-4 ledgers; `full-PDF` — the four exact JFM papers above, queried separately for final-section order, tense, compression, evidence relation, caveats, citations and closing moves; `mapped` — underlying Sections 1–8 evidence through their owning frozen packets only; `not-covered` — raw artifacts, new science, Section-9 observations, manuscript prose, displays, equations or tests. Observations are sample-bounded; recommendations are specific to the frozen DynamisLab architecture.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/00_AGENT_REGISTRY.md

@@ -0,0 +1,22 @@
+# Round-5 Agent Registry
+
+**Version:** `R5-REGISTRY-v4-citation-closeout`  
+**State:** post-R5 citation integration globally recounted and statically closed out; all units retain the unchanged final-audit disposition. Review assembly is conditionally admitted; literature-source, artifact and runtime gates remain separate.  
+**Rule:** one persistent writer per fixed unit. Agent IDs are recorded only after launch; later work resumes that ID with packet/registry plus an approved delta.
+
+| Unit | Sections | Canonical packet | Writer agent ID | Stage | Output ownership | Last accepted feedback | Unresolved gates | Next resume goal | Memory record |
+|---|---|---|---|---|---|---|---|---|---|
+| A | S1–S2 | `01_CONTEXT_PACKET_A_S01_S02.md` v1 | `5ddac128-c9c7-4b83-8fb9-aed05023a9b0` | `final-audit / S1–S2 conditional` | S1/S2 TeX and audits complete; shared `audits/WAVE_S02_S03_INTERFACE_AUDIT.md` | Final bounded S1 correction integrated: execution for Kármán/Illusion/Vortex, target-relative field proximity Kármán-only; Erase unresolved; no research | S2 selected-run/source/display/Reynolds gates; S1 bibliography and global interface closed | Runtime compile/production gates only; no prose reopening absent approved delta | Nowledge checkpoints previously read; no new memory/research |
+| B | S3–S4 | `02_CONTEXT_PACKET_B_S03_S04.md` v1 | `a7e7df0b-a529-47e3-85b5-95683f9e7e24` | `final-audit / S3–S4 conditional` | `sections/SECTION_03_CAPABILITY_PROGRESSION.tex`; `sections/SECTION_04_KARMAN_RESULTS.tex`; corresponding audits | First drafts, bounded five-paper full-PDF diagnosis, integrated rewrite and approved S2--S3 interface findings and targeted S4/S5 evidence reconciliation complete; interface owner: `audits/SECTION_02_AUDIT.md#s2s3-interface-audit` | S3 exact panel/offset/role provenance, captions and art; S4 redraw/crops/scales, spectrum window, action units and zero portrait; $U_0/U_{ref}$ conversion remains unbound | Enter gated figure production only under source/provenance controls | `no durable write; repository audit is authority` |
+| C | S5–S6 | `03_CONTEXT_PACKET_C_S05_S06.md` v1 | `cf2466a8-c3c8-44e9-a926-e7d826a8816a` | `final-audit / S5–S6 conditional` | S5/S6 TeX and audits; `audits/WAVE_S04_S06_INTERFACE_AUDIT.md` | Final Unit-C definitions integrated: S5 reflection/force/error contracts and S6 T/0/C/D plus consistent two-component velocity/residual notation; estimand separation retained | Historical velocity reference; PPO/SR mismatch; action extraction/figures; S6 mean window `[480,840)` bound; composites/POD production remains open; bibliography keys closed | Enter gated figure production; retain estimand and U0/Uref boundaries | `e06d3316-e9ea-4efa-9db9-7bee3f685904` |
+| D | S7–S8 | `04_CONTEXT_PACKET_D_S07_S08.md` v1 | `da7067b9-38f7-4b50-bc3b-e128f9f08a2f` | `final-audit / S7–S8 conditional` | `sections/SECTION_07_CROSS_CASE_EXTENSIONS.tex`; `audits/SECTION_07_AUDIT.md`; `sections/SECTION_08_STEADY_THEORY.tex`; `audits/SECTION_08_AUDIT.md` | Final Unit-D delta integrated: exact bounded S7 no-scalar claim and S8 cross-sectional two-component L-infinity endpoint metric; $g$ basis/sign/units retained | S7 evaluators/role visuals and Illusion scalar binding; S8 equation source/typesetting, source/provenance/artwork, finite-image labeling and mechanism gates | Enter gated role-field/steady production; quantitative Vortex/Erase remain blocked | `repository-recorded; no durable-memory write` |
+| E | S9–S10 | `05_CONTEXT_PACKET_E_S09_S10.md` v1 | `b238cc11-c57b-481f-922d-ae7c410171c0` | `final-audit / S9 blocked, S10 pass` | S9 author-field placeholder/audit and S10 five-paragraph TeX/audit complete | Accepted S7–S9 finding retained; S10 four-paper rhetoric diagnosis and bounded causal-wording fix integrated | All S9 author/provenance/result-import gates; S1/S10/global interface closed | Retain S9 omission; reopen only after author supply and targeted provenance audit | `repository audits written; no durable-memory write` |
+
+## Stage vocabulary
+`controls-ready → preflight → first-draft → literature-diagnosed → rewrite-approved → integrated-rewrite → section-audit → section-pass → global-pass → final-audit`.
+
+A stage never closes an evidence or artifact gate. Wave auditors report findings only; the coordinator approves deltas and the original writer integrates them.
+
+## Final normalization — 2026-08-11
+
+All units remain at `final-audit` after the delta-only refresh. This stage records prose/static disposition only: S1 and S10 pass; S2–S8 are conditional on their listed source/artifact gates; S9 substantive prose is blocked and omitted. Runtime compilation remains a conditional production gate. The authoritative post-integration count is 7,434 substantive words (S4–S6: 2,496 / 33.58%) in `00_LIVE_LENGTH_SHARE_LEDGER.md`. All 49 bibliography keys are active and resolve. `00_R5_FINAL_AUDIT.md` deliberately retains its pre-expansion overall disposition; post-integration findings are in `audits/CITATION_EXPANSION_FINAL_AUDIT.md`.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/00_CITATION_CANDIDATE_LEDGER.md

@@ -0,0 +1,194 @@
+# Round-5 Authoritative Citation and Bibliography Ledger
+
+**Project / round / unit:** DynamisLab JFM / Round 5 / bibliography integration.  
+**Owned bibliography:** `R5_REFERENCES.bib`.  
+**Scope:** reader-facing external propositions only. Literature never verifies a DynamisLab result, number, acquisition, artifact, mechanism, priority or readiness state. Section prose remains unchanged in this task.
+
+## Approved reader-facing key map
+
+| Key | Exact identity | DOI | Approved section / exact proposition placement | Source state |
+|---|---|---|---|---|
+| `PendryEtAl2006Controlling` | J. B. Pendry, D. Schurig & D. R. Smith, “Controlling electromagnetic fields,” *Science* 312, 1780–1782 (2006) | `10.1126/science.1125907` | **S01-P01**, after “linear or effectively linear field problems”: transformation-based cloak genealogy only | `full-PDF` repository record |
+| `MaEtAl2013ActiveCloaking` | Qian Ma, Zhong Lei Mei, Shou Kui Zhu, Tian Yu Jin & Tie Jun Cui, “Experiments on active cloaking and illusion for Laplace equation,” *Physical Review Letters* 111, 173901 (2013) | `10.1103/PhysRevLett.111.173901` | **S01-P01**, same sentence: active cloak and non-zero illusion genealogy for a Laplace problem | `full-PDF` repository record |
+| `UrzhumovSmith2012ActiveCloaks` | Yaroslav A. Urzhumov & David R. Smith, “Flow stabilization with active hydrodynamic cloaks,” *Physical Review E* 86, 056313 (2012) | `10.1103/PhysRevE.86.056313` | **S01-P01**, at the finite-Re transfer boundary: restricted active hydrodynamic-cloak precedent; not evidence for this plant | `full-PDF` repository record |
+| `DengEtAl2020PinballBifurcations` | Nan Deng, Bernd R. Noack, Marek Morzyński & Luc R. Pastur, “Low-order model for successive bifurcations of the fluidic pinball,” *Journal of Fluid Mechanics* 884, A37 (2020) | `10.1017/jfm.2019.959` | **S01-P02**, after “rich nonlinear wake dynamics”: external fluidic-pinball dynamical context | `full-PDF` reread in R5 |
+| `CornejoMacedaEtAl2021PinballControl` | G. Y. Cornejo Maceda, Y. Li, F. Lusseyran, M. Morzyński & B. R. Noack, “Stabilization of the fluidic pinball with gradient-enriched machine learning control,” *Journal of Fluid Mechanics* 917, A42 (2021) | `10.1017/jfm.2021.301` | **S01-P02**, after “open- and closed-loop control”: prior pinball control context only | `full-PDF` reread in R5 |
+| `RabaultEtAl2019DRLFlowControl` | Jean Rabault, Miroslav Kuchta, Atle Jensen, Ulysse Réglade & Nicolas Cerardi, “Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control,” *Journal of Fluid Mechanics* 865, 281–302 (2019) | `10.1017/jfm.2019.62` | **S01-P03**, after the model-free DRL motivation; optionally **S02-P02b** after the first PPO sentence if the method section needs a direct flow-control precedent | `full-PDF` reread in R5 |
+| `SchulmanEtAl2017PPO` | John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford & Oleg Klimov, “Proximal policy optimization algorithms,” arXiv:1707.06347 (2017) | no DOI; stable arXiv identifier `1707.06347` | **S02-P02b**, immediately after the clipped-surrogate PPO definition; algorithm definition only | exact repository identity checked |
+| `ParisEtAl2021RobustFlowControl` | Romain Paris, Samir Beneddine & Julien Dandois, “Robust flow control and optimal sensor placement using deep reinforcement learning,” *Journal of Fluid Mechanics* 913, A25 (2021) | `10.1017/jfm.2020.1170` | **S01-P03**, with Rabault after the limited-observation DRL precedent; not evidence that the present sensors are optimal | `full-PDF` reread in R5 |
+| `GautierEtAl2015ClosedLoop` | N. Gautier, J.-L. Aider, T. Duriez, B. R. Noack, M. Segond & M. Abel, “Closed-loop separation control using machine learning,” *Journal of Fluid Mechanics* 770, 442–457 (2015) | `10.1017/jfm.2015.95` | **S01-P04** and **S05-P01**, at the proposition that a readable learned feedback law must be judged in execution; external methodological precedent only | `full-PDF` reread in R5 |
+| `LoiseauEtAl2018SparseROM` | Jean-Christophe Loiseau, Bernd R. Noack & Steven L. Brunton, “Sparse reduced-order modelling: sensor-based dynamics to full-state estimation,” *Journal of Fluid Mechanics* 844, 459–490 (2018) | `10.1017/jfm.2018.147` | **S01-P04** and **S06-P05**, at the sparse-information/full-field and observable-associated versus energy-ranked basis distinction | `full-PDF` reread in R5 |
+| `RaibaudoEtAl2020PinballControl` | Cédric Raibaudo, Peng Zhong, Bernd R. Noack & Robert J. Martinuzzi, “Machine learning strategies applied to the control of a fluidic pinball,” *Physics of Fluids* 32, 015108 (2020) | `10.1063/1.5127202` | **S01-P02** and **S02-P02b**, prior fluidic-pinball control and downstream sensing context only | repository bibliography identity; DOI metadata checked |
+| `LiEtAl2022FlowEstimation` | Songqi Li, Wenpeng Li & Bernd R. Noack, “Machine-learned control-oriented flow estimation for multi-actuator multi-sensor systems exemplified for the fluidic pinball,” *Journal of Fluid Mechanics* 952, A36 (2022) | `10.1017/jfm.2022.908` | **S01-P04** and **S02-P02b**, sparse-sensor-to-full-field estimation and pinball sensing context; no validation of the present sensor arrangement | repository bibliography identity; publisher metadata checked |
+| `SchlegelEtAl2012LeastOrder` | Michael Schlegel, Bernd R. Noack, Peter Jordan, Andreas Dillmann, Elmar Gröschel, Wolfgang Schröder, Mingjun Wei, Jonathan B. Freund, Oliver Lehmann & Gilead Tadmor, “On least-order flow representations for aerodynamics and aeroacoustics,” *Journal of Fluid Mechanics* 697, 367–398 (2012) | `10.1017/jfm.2012.70` | **S06-P05**, after the POD/observable-oriented basis distinction; methodological context only | `full-PDF` reread in R5 |
+| `Boree2003ExtendedPOD` | J. Borée, “Extended proper orthogonal decomposition: a tool to analyse correlated events in turbulent flows,” *Experiments in Fluids* 35, 188–192 (2003) | `10.1007/S00348-003-0656-3` | **S06-P05**, same methodological sentence: external precedent for correlation-oriented modal analysis; no CCD validation or causal role | `full-PDF` reread in R5 |
+| `Crowdy2006MultipleCylinders` | Darren G. Crowdy, “Analytical solutions for uniform potential flow past multiple cylinders,” *European Journal of Mechanics B/Fluids* 25, 459–470 (2006) | `10.1016/j.euromechflu.2005.11.005` | **S08-P04**, after introducing the irrotational impermeability family: mathematical outer-flow precedent only | `full-PDF` reread in R5 |
+| `UedaEtAl2003RotatingCylinders` | Y. Ueda, A. Sellier, T. Kida & M. Nakanishi, “On the low-Reynolds-number flow about two rotating circular cylinders,” *Journal of Fluid Mechanics* 495, 255–281 (2003) | `10.1017/S002211200300627X` | **S08-P04**, after the rotating no-slip incompatibility / validity boundary: viscous rotating-cylinder boundary context only, not finite-$Re_D=50$ verification | `full-PDF` reread in R5 |
+
+These sixteen records remain the legacy cited subset. Bibliography normalization added 15 high-confidence local identities and 18 full-PDF-verified Undermind identities, producing 49 unique records. The subsequent prose-integration pass is complete: all 49 keys are now active.
+
+## Full-PDF audit identities not admitted as reader-facing citations
+
+The section audits also used the following papers for rhetoric, ordering, caveat adjacency, caption discipline or placeholder design. They are deliberately excluded from `R5_REFERENCES.bib` because no retained reader-facing external proposition currently needs them:
+
+- S02–S04 diagnosis: Mao et al. (2015), `10.1017/jfm.2015.304`; Xia et al. (2024), `10.1017/jfm.2024.69`; Wang et al. (2024), `10.1017/jfm.2024.333`. Paris et al. is admitted only for the S01 placement above.
+- S05–S06 diagnosis: Cornejo Maceda et al., Gautier et al., Loiseau et al., Schlegel et al. and Borée are admitted only for the placements above; no additional diagnosis paper is retained merely for rhetoric.
+- S07–S08 diagnosis: Deng et al. (2021), `10.1017/jfm.2021.299`; Chan et al. (2011), `10.1017/jfm.2011.134`. Cornejo Maceda et al., Crowdy and Ueda et al. are admitted only for the placements above.
+- S09 reporting diagnosis: Bhattacharyya et al. (2022), `10.1017/jfm.2022.792`; Khan et al. (2023), `10.1017/jfm.2023.549`; Klotz et al. (2024), `10.1017/jfm.2024.639`; Zhou et al. (2020), `10.1017/jfm.2020.392`. S09 remains an author-field placeholder and retains no reader-facing citation.
+- S10 rhetoric diagnosis: Li & Zhang (2022), `10.1017/jfm.2021.1045`, plus Deng, Cornejo Maceda and Rabault. S10 imports established internal synthesis only and retains no reader-facing citation.
+
+## Existing bibliography suitability and stale/conflicting records
+
+Neither existing repository bibliography is safe as the authoritative R5 source:
+
+- `../My_JFM/jfm.bib` is coupled to the stale `FLMguide.tex` draft and mixes useful records with incomplete or conflicting metadata. Most importantly, key `Urzhumov2012` actually contains the different 2011 paper “Fluid flow control with transformation media,” omits its DOI, and cannot represent the intended 2012 active-hydrodynamic-cloak source. `Gautier2015` and `Boree2003` also omit verified DOIs. Several later JFM records omit volume/article metadata. The file is left untouched.
+- `../../My_Confirmation/mybibfile.bib` uses a separate thesis-era key scheme, omits DOIs on relevant records such as Pendry and Rabault, and contains at least one visibly malformed author token (`<? format?>`). It is not imported into R5.
+- The old key `Urzhumov2012` is therefore **prohibited** in R5. Use `UrzhumovSmith2012ActiveCloaks` for DOI `10.1103/PhysRevE.86.056313`. The distinct 2011 transformation-media paper is not currently needed.
+- Current section prose still contains legacy keys because prose was explicitly out of scope. The required insertion-pass substitutions are: `Deng2020` → `DengEtAl2020PinballBifurcations`; `CornejoMaceda2021` → `CornejoMacedaEtAl2021PinballControl`; `Raibaudo2020` → `RaibaudoEtAl2020PinballControl`; `Rabault2019` → `RabaultEtAl2019DRLFlowControl`; `Paris2021` → `ParisEtAl2021RobustFlowControl`; `Schulman2017` → `SchulmanEtAl2017PPO`; `Gautier2015` → `GautierEtAl2015ClosedLoop`; `Loiseau2018` → `LoiseauEtAl2018SparseROM`; `Li2022` → `LiEtAl2022FlowEstimation`; and `Boree2003` → `Boree2003ExtendedPOD`. These legacy keys are not approved for final R5 assembly.
+
+## Validation record
+
+- Key policy: descriptive `AuthorEtAlYearTopic` keys; no collision with either existing bibliography.
+- DOI uniqueness: one normalized DOI per DOI-bearing R5 entry; no duplicate DOI within `R5_REFERENCES.bib`.
+- Identity check: titles/authors/DOIs cross-checked against current repository authorities and publisher/index metadata where needed. The misleading web-search result for `10.1017/S002211200300627X` was rejected; repository full-PDF authority and the publisher DOI resolve to Ueda, Sellier, Kida & Nakanishi.
+- Syntax: no `bibtex`/`biber` executable is installed in the repository environment. The bibliography-normalization audit records the current static parser result for 49 entries, unique keys, normalized DOI uniqueness and balanced braces.
+- Section prose and `../My_JFM/FLMguide.tex` were not edited.
+
+## Evidence-use closeout
+
+Nowledge was searched narrowly for R5 citation/bibliography decisions and used only for navigation. All ten section audits, the current R5 citation ledger, R5 section TeX, both repository `.bib` files and the legacy TeX bibliography calls were inspected. Literature states above preserve the audits' `full-PDF` versus repository-record distinctions. No literature source is assigned an internal-result role.
+
+## Post-R5 local-source expansion control
+
+This control originally reopened citation planning only. Bibliography normalization admitted the selected 15 high-confidence local identities and all 18 verified Undermind identities; the later authorized prose-integration pass then placed all 33. The current closeout does not alter section prose, bibliography or `00_R5_FINAL_AUDIT.md`.
+
+The merged local-source audit is `audits/CITATION_EXPANSION_LOCAL_SOURCE_AUDIT.md`. It consolidates the Confirmation and legacy `My_JFM` source maps, gives paragraph-level jobs, preserves duplicate and identity-collision handling, and separates the local wave from later Undermind gap searches.
+
+### Expansion arithmetic and decision boundary
+
+- Confirmation source: 35 unique cited keys / 53 cited-key occurrences; exact current-R5 duplicates are `pendry2006controlling` → `PendryEtAl2006Controlling` and `rabault2019artificial` → `RabaultEtAl2019DRLFlowControl`.
+- Legacy `My_JFM`: 9 active keys / 10 cited-key occurrences; exact current-R5 duplicates are `Gautier2015` → `GautierEtAl2015ClosedLoop`, `Deng2020` → `DengEtAl2020PinballBifurcations`, `CornejoMaceda2021` → `CornejoMacedaEtAl2021PinballControl`, and `Boree2003` → `Boree2003ExtendedPOD`.
+- High-confidence local wave: **15 net-new identities** (12 Confirmation identities, including one conventional AFC example, plus 3 legacy identities). If all pass full-text verification and sentence-level review, the bibliography would rise from **16 to 31 unique references**.
+- Optional local wave: **4 additional net-new identities**. If all high-confidence and optional identities pass, the ceiling would be **19 net-new / 35 total unique references**.
+- Planned substantive uplift: S1 +300–430, S2 +130–200, S5 +90–150 and S6 +15–35 words, for **+535–815 words** overall. Applied to the current uniform count of 6,303, the projected body is **6,838–7,118 words**. These are controlled ranges, not padding quotas.
+- S3, S4, S7, S8 and S10 receive no expansion by default. A future insertion requires a proposition that cannot be owned more precisely by S1, S2, S5 or S6.
+
+Every local-wave record remains `local-bibliography/context mapped, full text unverified` and is now bibliography-added with provisional metadata flags retained in the audits. Bibliographic mapping can nominate an external proposition; it cannot support detailed wording until full text is read. It cannot verify any internal result, number, acquisition, artifact, mechanism, readiness state or priority claim.
+
+### Collision and quarantine controls
+
+- `Urzhumov2012` in legacy `jfm.bib` is the **2011** PRL article “Fluid flow control with transformation media”; current `UrzhumovSmith2012ActiveCloaks` is the **2012** PRE article “Flow stabilization with active hydrodynamic cloaks”. They are distinct identities. The old misleading key remains prohibited and the 2011 paper is not in the local import wave.
+- `Loiseau2018Galerkin` (Loiseau & Brunton, JFM 838, constrained sparse Galerkin regression) and current `LoiseauEtAl2018SparseROM` (Loiseau, Noack & Brunton, JFM 844, sensor-based dynamics to full-state estimation) are distinct papers, not duplicate keys.
+- Hard quarantine: positive Illusion SR; cross-case SR transfer; OID; CCD causality, validation or triangulation; old Reynolds/configuration chains; internal old numbers; experiment validation; and priority or first-of-kind wording. No local source can lift those ceilings.
+
+### Verification gate
+
+Before any bibliography or TeX insertion: verify identity and DOI/identifier; read the full text for the exact proposition; replace the provisional key with a collision-safe R5 key; check paragraph ownership and adjacent caveat; rerun unique-key, unresolved-key, uncited-entry and word-count checks; then issue a new audit. The current final-audit `CONDITIONAL PASS` is unchanged.
+
+
+## Post-R5 Undermind verified gap wave
+
+**Scope:** existing DynamisLab workspace only; no new search or deep search. This verified pool is now bibliography-added; citation insertion remains a separate prose unit. Each selected identity was checked with `get_paper_info(show_doi=true)` and its exact proposition with a bounded `read_pdfs` question.
+
+| Job | Key | Identity / DOI | Verified proposition and passage | Safe topic | State |
+|---|---|---|---|---|---|
+| S1-P3 | `Fan20b` | Fan et al., PNAS 117 (2020), `10.1073/pnas.2004939117` | RL operated rotating control cylinders in towing experiments and simulations (abstract p.1; Technical Approach p.2). | Experimental/numerical bluff-body RL precedent only. | `full-PDF`, pp.1--2 |
+| S1-P3 | `Li21` | Li & Zhang, JFM 932 (2022), `10.1017/jfm.2021.1045` | Sparse-probe control depends on placement relative to a Re/blockage-dependent sensitive region (abstract p.1; §§5.6, 6.1, pp.26--28). | Sparse RL sensing is task- and flow-dependent. | `full-PDF`, pp.1, 26--28 |
+| S2 observation | `Mar24b` | Marra et al., JFM 996 (2024), `10.1017/jfm.2024.593` | A five-dimensional pinball manifold separates three actuation and two shedding amplitude/phase coordinates; estimation uses two deliberately placed measurements, without placement optimization (§§4.2--5, pp.8--11). | Actuation and sparse wake data encode complementary coordinates. | `full-PDF`, pp.8--11 |
+| S2 placement | `Jin21` | Jin et al., JFM 932 (2022), `10.1017/jfm.2021.948` | Estimation-optimal placement is not feedback-optimal because downstream information growth trades against convective lag (§§4.1.3, 4.4.2, pp.15, 25). | Estimation and feedback impose different placement requirements. | `full-PDF`, pp.15, 25 |
+| S2 periodic sensing | `Gon19` | Gong et al., Phys. Rev. Fluids 5 (2020), `10.1103/PhysRevFluids.5.023901` | One sensor plus a harmonic model estimates periodic shedding in the studied regimes; higher harmonics are inferred from the fundamental (abstract p.1; §V.D p.11). | A phase-bearing sparse signal can encode dominant periodic dynamics. | `full-PDF`, pp.1, 11 |
+| S2 force/phase | `Nai20b` | Nair et al., JFM 927 (2021), `10.1017/jfm.2021.735` | Lift/drag time series replace online full-field access for periodic phase control, including rotary actuation (abstract p.1; §2.1 p.6; §3.2.1 p.11; conclusions p.18). | Force signals can carry periodic phase information. | `full-PDF`, pp.1, 6, 11, 18 |
+| S2 DRL state | `Wan23b` | Wang et al., JFM (2024), `10.1017/jfm.2024.333` | Dynamic lifting adds 30 prior action-step pressure histories, with dimensionality/redundancy caveats (§2.2 p.8; §4.2.1 p.16). | Temporal features can enrich sparse instantaneous sensing at a cost. | `full-PDF`, pp.8, 16 |
+| S5-P1 | `Deb15` | Debien et al., Exp. Fluids 57 (2016), `10.1007/s00348-016-2126-8` | GP expression-tree laws are compiled and evaluated in the real-time experimental loop (§§3.1--3.2, pp.8--9). | Direct symbolic controller search differs from post-hoc distillation. | `full-PDF`, pp.8--9 |
+| S5-P1/P5 | `Cas22` | Castellanos et al., Phys. Fluids 34 (2022), `10.1063/5.0087208` | DRL and LGPC are tested closed-loop; LGPC yields compact sensor-selective laws while DRL is less interpretable and has different tested robustness (§§II.C--E pp.5--6; §III.A p.7; conclusions p.13). | Policy-map interpretability and executed behaviour are distinct. | `full-PDF`, pp.5--7, 13 |
+| S5-P1 | `Ren19` | Ren et al., Phys. Fluids 31 (2019), `10.1063/1.5115258` | GP symbolic expressions become explicit nonlinear feedback laws and each is evaluated in coupled CFD/FSI (§II.C pp.4--5; law p.7). | Direct GP laws are plant-evaluated controllers. | `full-PDF`, pp.4--7 |
+| S5-P5 | `Zol25` | Zolman et al., Nat. Commun. 16 (2025), `10.1038/s41467-025-65738-4` | SINDy-RL jointly learns surrogate dynamics/reward and policy in Dyna-style model-based RL, including a flow example (abstract p.1; framework p.3). | Joint model-based sparse RL is not fixed-policy post-hoc regression. | `full-PDF`, pp.1, 3 |
+| S6 lineage | `Jor07` | Jordan et al., AIAA 2007-3602, `10.2514/6.2007-3602` | MOD resolves velocity modes by linear observability in far-field pressure rather than kinetic energy (abstract p.1; §II pp.3, 5). | Observable-ranked decomposition ancestry only. | `full-PDF`, pp.1, 3, 5 |
+| S6 mean field | `Tad10` | Tadmor et al., Phys. Fluids 22 (2010), `10.1063/1.3298960` | Shift modes represent changing means and their bilateral coupling with fluctuations in natural/actuated wakes (abstract p.1; §III.B.2 p.8). | Mean and fluctuation organization are distinct coupled objects. | `full-PDF`, pp.1, 8 |
+| S6 lineage | `Dis18` | Discetti et al., Exp. Therm. Fluid Sci. 93 (2018), `10.1016/j.expthermflusci.2017.12.011` | POD point/field temporal modes are correlated for reconstruction, with uncorrelated/noisy contributions truncated (abstract p.2; §2.2 pp.5--6). | Correlation-conditioned reconstruction is descriptive, not causal. | `full-PDF`, pp.2, 5--6 |
+| S8 boundary | `Mit01` | Mittal, J. Fluids Struct. 15 (2001), `10.1006/JFLS.2000.0337` | Rotating-control-cylinder wake outcomes vary with Re, gap, speed and dimensionality; 2-D is an upper bound (abstract p.291; conclusions p.324). | Rotating-cylinder control is parameter/dimension dependent. | `full-PDF`, pp.291, 324 |
+| S8 boundary | `Cha11b` | Chan et al., JFM 679 (2011), `10.1017/jfm.2011.134` | Suppression speed depends on Re, gap and rotation sense; a viscous high-rotation state only resembles a potential doublet (§§1, 4.1, 4.3, 5, pp.345, 352, 367, 375). | Potential-flow resemblance does not establish inviscid closure. | `full-PDF`, cited pages |
+| S8 boundary | `Sie20` | Sierra et al., JFM 905 (2020), `10.1017/jfm.2020.692` | Rotating-cylinder flow below Re 200 has multiple steady/cyclic and bistable regimes (abstract p.1; §§3.4--3.5 pp.8--11; conclusions p.16). | Viscous behaviour is branch- and parameter-dependent. | `full-PDF`, pp.1, 8--11, 16 |
+| S8 boundary | `Wat96` | Watson, QJMAM 49 (1996), `10.1093/QJMAM/49.2.195` | Slow uniform flow past two rotating cylinders matches inner Stokes and outer Oseen solutions; rotation can make lift the same order as drag (summary/introduction p.195). | Even low-Re pairs require viscous inner/outer matching. | `full-PDF`, p.195 |
+
+### Balance, arithmetic and gate
+
+- **18 net-new identities beyond both current R5 and the frozen local high-confidence 15:** S1 2; S2 5; S5 4; S6 3; S8 4. S3/S4/S7/S9/S10 receive none.
+- Projected union: 16 current + 15 local + 18 Undermind = **49 unique references**. Recommended final range: **46--49**, not mechanical 50; 50--51 requires an exact essential sentence gap.
+- Estimated wave uplift: S1 +35--55, S2 +140--210, S5 +120--180, S6 +55--85, S8 +85--125 = **+435--655 words**. Combined local+Undermind uplift is **+970--1,470**, projecting 6,303 to **7,273--7,773** before recount.
+- Duplicates (zero count): `Rai20`, `Mac20`, `Loi17`, `Bor03`, `Sch12`, `Par20`, `Rab18b`, `Ued03`. Local-wave collisions: `Li16`, `Loi16`. Rejected: redundant `Bru16`; unused `Ish17`; inaccessible `Mau01`/`Bon94`; redundant additional inviscid multiple-cylinder papers.
+- Hard quarantine persists: no OID; no PCD/CCD causality, validation, superiority or triangulation; retain `action-correlated correction-field modes` as descriptive/noncausal; no positive Illusion SR, cross-case SR transfer, project-number/artifact verification, experiment validation or priority wording.
+- Before prose insertion: use the final R5 keys below, check sentence ownership/caveat adjacency, and rerun unresolved-key, uncited-entry and word-count checks. The final disposition remains unchanged. See `audits/CITATION_EXPANSION_UNDERMIND_GAP_AUDIT.md`.
+
+## Final post-R5 bibliography key mapping
+
+**Bibliography state:** added. `R5_REFERENCES.bib` now contains 49 unique records: 16 preserved legacy records plus the following 33 net-new identities. All 33 additions are now cited after the separate section-prose integration unit.
+
+| Wave | Source key | Final collision-safe R5 key | DOI / identifier | Bibliography state | Metadata state |
+|---|---|---|---|---|---|
+| Local | `leonhardt2006optical` | `Leonhardt2006OpticalMapping` | `10.1126/science.1126493` | Added | Local metadata; DOI available |
+| Local | `miller2006perfect` | `Miller2006PerfectActiveCloaking` | none recorded | Added | Provisional local metadata |
+| Local | `vasquez2009active` | `VasquezEtAl2009ActiveExterior` | `10.1103/PhysRevLett.103.073901` | Added | Local metadata; DOI available |
+| Local | `lin2021active` | `LinEtAl2021ActiveAcousticIllusions` | none recorded | Added | Provisional local metadata |
+| Local | `chen2024multilayered` | `ChenEtAl2024MultilayerHydrodynamicCloak` | none recorded | Added | Provisional local metadata |
+| Local | `ren2021bluff` | `RenEtAl2021HydrodynamicStealthDRL` | `10.1063/5.0060690` | Added | DOI recovered from existing R1 record |
+| Local | `deng2021galerkin` | `DengEtAl2021PinballForceModel` | `10.1017/jfm.2021.299` | Added | DOI recovered from existing R1 record |
+| Local | `feng2023control` | `FengEtAl2023PinballForceControl` | `10.1063/5.0142949` | Added | DOI/article recovered from existing R1/R4 record |
+| Local | `beem2015wake` | `BeemTriantafyllou2015WakeSensing` | none recorded | Added | Provisional local metadata |
+| Local | `sakoe1978dynamic` | `SakoeChiba1978DynamicProgramming` | `10.1109/TASSP.1978.1163055` | Added | Local metadata; DOI available |
+| Local | `muller2015fundamentals` | `Muller2015FundamentalsMusicProcessing` | none recorded | Added | Provisional local book metadata |
+| Local | `feng2010circular` | `FengWang2010SyntheticJetCylinder` | none recorded | Added | Provisional local metadata |
+| Local | `Li2017` | `LiEtAl2017GeneticProgrammingControl` | none recorded | Added | Provisional local metadata; distinct from excluded `Li16` |
+| Local | `Brunton2016` | `BruntonEtAl2016SINDy` | `10.1073/pnas.1517384113` | Added | Local metadata; DOI available |
+| Local | `Cranmer2023` | `Cranmer2023PySR` | arXiv `2305.01582` | Added | Local arXiv identity |
+| Undermind | `Fan20b` | `FanEtAl2020BluffBodyRL` | `10.1073/pnas.2004939117` | Added | Full-PDF verified |
+| Undermind | `Li21` | `LiZhang2022ConfinedWakeRL` | `10.1017/jfm.2021.1045` | Added | Full-PDF verified |
+| Undermind | `Mar24b` | `MarraEtAl2024ActuationManifold` | `10.1017/jfm.2024.593` | Added | Full-PDF verified |
+| Undermind | `Jin21` | `JinEtAl2022SensorActuatorPlacement` | `10.1017/jfm.2021.948` | Added | Full-PDF verified |
+| Undermind | `Gon19` | `GongEtAl2020VortexEstimation` | `10.1103/PhysRevFluids.5.023901` | Added | Full-PDF verified |
+| Undermind | `Nai20b` | `NairEtAl2021PhaseFlowControl` | `10.1017/jfm.2021.735` | Added | Full-PDF verified; no unverified article number added |
+| Undermind | `Wan23b` | `WangEtAl2024DynamicFeatureDRL` | `10.1017/jfm.2024.333` | Added | Full-PDF verified |
+| Undermind | `Deb15` | `DebienEtAl2016GeneticProgrammingRamp` | `10.1007/s00348-016-2126-8` | Added | Full-PDF verified; year normalized to publication year |
+| Undermind | `Cas22` | `CastellanosEtAl2022FewSensorMLControl` | `10.1063/5.0087208` | Added | Full-PDF verified |
+| Undermind | `Ren19` | `RenEtAl2019MachineLearningVIV` | `10.1063/1.5115258` | Added | Full-PDF verified |
+| Undermind | `Zol25` | `ZolmanEtAl2025SINDyRL` | `10.1038/s41467-025-65738-4` | Added | Full-PDF verified |
+| Undermind | `Jor07` | `JordanEtAl2007ObservableJetModes` | `10.2514/6.2007-3602` | Added | Full-PDF verified |
+| Undermind | `Tad10` | `TadmorEtAl2010CylinderMeanField` | `10.1063/1.3298960` | Added | Full-PDF verified |
+| Undermind | `Dis18` | `DiscettiEtAl2018CorrelatedFieldEstimation` | `10.1016/j.expthermflusci.2017.12.011` | Added | Full-PDF verified |
+| Undermind | `Mit01` | `Mittal2001RotatingControlCylinders` | `10.1006/JFLS.2000.0337` | Added | Full-PDF verified |
+| Undermind | `Cha11b` | `ChanEtAl2011CounterRotatingCylinders` | `10.1017/jfm.2011.134` | Added | Full-PDF verified |
+| Undermind | `Sie20` | `SierraEtAl2020RotatingCylinderBifurcations` | `10.1017/jfm.2020.692` | Added | Full-PDF verified |
+| Undermind | `Wat96` | `Watson1996TwoRotatingCylinders` | `10.1093/QJMAM/49.2.195` | Added | Full-PDF verified |
+
+The four optional local identities remain excluded: `MonticoneAlu2013CloakScatteringBounds`, `ZouEtAl2019WaterWaveCloak`, `BruntonEtAl2020MachineLearningFluids` and `LoiseauBrunton2018ConstrainedGalerkin`. The prohibited legacy key `Urzhumov2012` is absent. The selected Undermind wave does not include `Li16`; therefore `LiEtAl2017GeneticProgrammingControl` has no imported duplicate.
+
+
+## Post-integration global closeout — 2026-08-11
+
+Static current-file census: **44 citation commands, 64 cited-key occurrences and 49 unique active keys**. `R5_REFERENCES.bib` has **49 entries / 49 unique keys**. Undefined keys = 0; uncited bibliography entries = 0; duplicate keys = 0; duplicate normalized DOIs = 0. All **33/33** post-R5 additions are cited. No key is recommended for pruning merely to clear an uncited state, and no citation was added in this closeout.
+
+The 15 local-wave identities retain their audited `local-bibliography/context mapped, full text unverified` source state; their current placements remain bounded to broad genealogy/method/context propositions. The 18 Undermind identities retain `full-PDF` state. Seven local records retain provisional metadata without a recorded DOI (`Miller2006PerfectActiveCloaking`, `LinEtAl2021ActiveAcousticIllusions`, `ChenEtAl2024MultilayerHydrodynamicCloak`, `BeemTriantafyllou2015WakeSensing`, `Muller2015FundamentalsMusicProcessing`, `FengWang2010SyntheticJetCylinder`, `LiEtAl2017GeneticProgrammingControl`); `Cranmer2023PySR` is intentionally identified by arXiv `2305.01582`. Exact unresolved controls and disposition are in `audits/CITATION_EXPANSION_FINAL_AUDIT.md`.
+
+## Local-wave full-text evidence gate — 2026-08-11
+
+This table supersedes every earlier `local-bibliography/context mapped, full text unverified` state for the 15 cited local-wave records. It records the exact current S1/S2/S5 proposition, identity check, readable passage and insertion disposition; it does not edit TeX or bibliography.
+
+| R5 key | Verified identity / DOI or link | PDF/read state | Current proposition and passage | Disposition |
+|---|---|---|---|---|
+| `Leonhardt2006OpticalMapping` | U. Leonhardt, “Optical conformal mapping,” *Science* 312, 1777--1780 (2006), DOI `10.1126/science.1126493` | `full-PDF` | Conformal mapping supplies an effectively two-dimensional optical-invisibility construction; p.1778 (PDF p.2), mapping to empty space. | **KEEP**; current transformation-optics genealogy is bounded and supported. |
+| `Miller2006PerfectActiveCloaking` | D. A. B. Miller, “On perfect cloaking,” *Optics Express* 14, 12457--12466 (2006), DOI `10.1364/OE.14.012457` | `mapped/inaccessible`; exact PDF retrieval failed | Metadata/abstract says sensors and active sources near the region, but no page passage was readable. | **REMOVE** from the current citation cluster unless full text is supplied; Vasquez supports the retained source-based linear-PDE example. |
+| `VasquezEtAl2009ActiveExterior` | F. Guevara Vasquez, G. W. Milton & D. Onofrei, “Active exterior cloaking for the 2D Laplace and Helmholtz equations,” *PRL* 103, 073901 (2009), DOI `10.1103/PhysRevLett.103.073901` | `full-PDF` | One exterior device is proved for 2-D quasistatics and three are demonstrated numerically for 2-D Helmholtz; abstract p.1, equations/limits pp.1--3. | **KEEP**; current “specified linear equations” wording is safe. |
+| `LinEtAl2021ActiveAcousticIllusions` | C. Lin et al., “Active acoustic cloaking and illusions of sound-hard bodies using the boundary element method,” *JASA* 149, 1803--1812 (2021), DOI `10.1121/10.0003556` | `full-PDF` | Active MIMO acoustic control reproduces a target scattered field to misrepresent object size/orientation; §III.B pp.1805--1806 and §IV pp.1810--1811. | **KEEP**; retain acoustic/non-zero-signature scope. |
+| `ChenEtAl2024MultilayerHydrodynamicCloak` | M. Chen, X. Shen & L. Xu, “A multilayered homogeneous hydrodynamic cloak realized in a free fluid flow,” *APL* 125, 171701 (2024), DOI `10.1063/5.0230069` | `full-PDF` | Ten-layer nonporous free-fluid cloak tested numerically and experimentally at `Re≈0.08`; pp.2--4, with `Re≪1` limit on p.5. | **KEEP**; “restricted low-inertia or free-flow regimes” is supported. |
+| `RenEtAl2021HydrodynamicStealthDRL` | F. Ren, C. Wang & H. Tang, “Bluff body uses deep-reinforcement-learning trained active flow control to achieve hydrodynamic stealth,” *Phys. Fluids* 33, 093602 (2021), DOI `10.1063/5.0060690` | `full-PDF` | DRL controls WSLB actuation using 33 velocity sensors on a line 2D downstream; the sensor-line uniform-flow discrepancy defines the objective; §§I--II, pp.1--5. | **KEEP**; current text correctly refuses equivalence with full-field target comparison. |
+| `DengEtAl2021PinballForceModel` | N. Deng et al., “Galerkin force model for transient and post-transient dynamics of the fluidic pinball,” *JFM* 918, A4 (2021), DOI `10.1017/jfm.2021.299` | `full-PDF` | Force is a constant-linear-quadratic function of Galerkin amplitudes and is applied across transient/post-transient regimes; abstract p.1, eq.2.18 p.6, conclusion p.25. | **KEEP**; “informative reduced object, not field substitute” is a safe manuscript distinction. |
+| `FengEtAl2023PinballForceControl` | H. Feng et al., “How to control hydrodynamic force on fluidic pinball via deep reinforcement learning,” *Phys. Fluids* 35 (2023), DOI `10.1063/5.0142949` | `full-PDF` | DRL controls force extrema and drag tracking using cylinder rotation and force-history state; abstract p.1, §III p.5, §IV.A pp.9--10, conclusion p.13. | **KEEP**; paper explicitly leaves wake/velocity information to future work. |
+| `BeemTriantafyllou2015WakeSensing` | H. Beem & M. S. Triantafyllou, “Wake-induced ‘slaloming’ response explains exquisite sensitivity of seal whisker-like sensors,” *JFM* 783, 306--322 (2015), DOI `10.1017/jfm.2015.513` | `mapped/inaccessible`; exact JFM PDF retrieval failed | No page passage available. Workspace `Bee15b` is an accessible arXiv precursor with a different title, not the exact cited record. | **REMOVE** both current uses unless the exact full text is supplied and read; delete/soften the biological analogy. |
+| `SakoeChiba1978DynamicProgramming` | H. Sakoe & S. Chiba, “Dynamic programming algorithm optimization for spoken word recognition,” *IEEE TASSP* 26, 43--49 (1978), DOI `10.1109/TASSP.1978.1163055` | `full-PDF` | DP minimizes time-normalized distance; monotonicity, continuity, boundary and slope constraints define admissible warping; §§I--II.B, pp.43--44. | **KEEP**; current generic origin/path statement is supported. |
+| `Muller2015FundamentalsMusicProcessing` | M. Müller, *Fundamentals of Music Processing*, Springer (2015), DOI `10.1007/978-3-319-21945-5` | PDF listed available, but two bounded `read_pdfs` attempts timed out; `mapped/unread` | No chapter/page passage returned for local cost, accumulated cost or path constraints. | **REMOVE** pending readable chapter evidence; reword the sentence to rely only on Sakoe--Chiba’s verified path/constraint definitions. |
+| `FengWang2010SyntheticJetCylinder` | L. H. Feng & J. J. Wang, “Circular cylinder vortex-synchronization control with a synthetic jet positioned at the rear stagnation point,” *JFM* 662, 232--259 (2010), DOI `10.1017/S0022112010003174` | `mapped/inaccessible`; exact PDF retrieval failed | No page passage available. Workspace `Fen10` is a different three-author *Journal of Fluids and Structures* paper and cannot verify this citation. | **REMOVE** this key and the localized-AFC clause unless exact full text is supplied. |
+| `LiEtAl2017GeneticProgrammingControl` | R. Li et al., “Drag reduction of a car model by linear genetic programming control,” *Exp. Fluids* 58, 103 (2017), DOI `10.1007/s00348-017-2382-2` | `full-PDF` | Symbolic GP uses the experiment to evolve explicit feedback laws; inner-loop plant evaluation and outer-loop learning are stated in §1 p.3 and §3.3 p.9. | **KEEP**; direct-controller-search versus post-hoc regression is exact. |
+| `BruntonEtAl2016SINDy` | S. L. Brunton, J. L. Proctor & J. N. Kutz, “Discovering governing equations from data by sparse identification of nonlinear dynamical systems,” *PNAS* 113, 3932--3937 (2016), DOI `10.1073/pnas.1517384113` | `full-PDF` | Sparse regression selects active terms from a candidate nonlinear-function library to form parsimonious dynamical equations; abstract/significance p.1 and method p.2. | **KEEP**; current text correctly distinguishes dynamics identification from policy mapping. |
+| `Cranmer2023PySR` | M. Cranmer, “Interpretable Machine Learning for Science with PySR and SymbolicRegression.jl,” arXiv:2305.01582 (2023), DOI `10.48550/arXiv.2305.01582` | `full-PDF` | PySR uses a multi-population evolve--simplify--optimize loop and explicitly optimizes constants; pp.1, 5--6. | **KEEP**; method/software role is supported and does not validate the present law. |
+
+**Gate result:** 11 **KEEP**, 0 **REWORD-only**, 4 **REMOVE pending readable full text** (`Miller2006PerfectActiveCloaking`, `BeemTriantafyllou2015WakeSensing`, `Muller2015FundamentalsMusicProcessing`, `FengWang2010SyntheticJetCylinder`). The four failures are evidence gates, not identity uncertainty: all identities/DOIs were resolved, but exact full-text passages were unavailable. Detailed audit: `audits/CITATION_EXPANSION_LOCAL_FULLTEXT_VERIFICATION.md`.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/00_CLAIM_EVIDENCE_ARTIFACT_LEDGER.md

@@ -0,0 +1,19 @@
+# Round-5 Claim, Evidence and Artifact Binding Ledger
+
+**State:** initialized from R4 controls. `OPEN` means the dependency remains unresolved; no row is artifact approval.
+
+| ID | Section | Licensed claim/job | Controlling authority | Artifact/equation dependency | Gate/status | Fallback |
+|---|---|---|---|---|---|---|
+| CEA-01 | S1 | framing and evidence-exact ladder: periodic Kármán cloak; positive non-zero-target Illusion PPO; Vortex event-relative execution/morphology; Erase unresolved; Kármán-only Levels 2–3 | S1 packet + approved global evidence delta + global ledger + current S2–S9 TeX/audits | citations only; no owned display/equation/table | PROSE AUDITED; bibliography and S1/S10/global pass OPEN | generic bounded hook; omit unkeyed citations; never promote Erase |
+| CEA-02 | S2 | definitions/method/evaluation/compact qualification | S2 packet + training pipeline/design decisions + sensor/L2 report + CelerisLab dossier | schematics/tables/run rows/Re labels | PROSE AUDITED; OPEN source/production, parabolic inherited Re and bibliography keys | omit unbound values; textual display fallbacks; supplement provenance |
+| CEA-03 | S3 | non-equivalent capability progression | S3 packet + S3 Figure Plans + named steady/Kármán authorities | two-figure interface accepted; exact Vortex offset, role leaves, hashes, units, crop/scale/caption/artwork | prose drafted; author choice CLOSED; artifact gates OPEN | source-bound previews; omit failed panel claim, preserve interface |
+| CEA-04 | S4 | tested learning and reference-wake evaluation | S4 packet + latest training summaries + role metadata/DTW + wake-L2 authority + time-window package | redraw/hash/unit/scale/crop/spectrum window/action unit/zero portrait | prose/values source-bound; historical $U_0$ normalization/time explicit; S4/S5 target+zero shared, controlled acquisitions distinct; production OPEN | scalar prose/compact table; omit spectrum then optional DTW display |
+| CEA-05 | S5 | executable Kármán/cloak-only surrogate and parent-relative deletion; 5/5 jobs drafted | S5 packet + `SR_analysis/CLAIMS.md`, results index, standardized SR summary and deletion table (`read`) | equations retained; rolling native DTW window/aggregation bound; action extraction, figure and velocity conversion | PROSE DRAFTED; production/source OPEN; PPO comparison BLOCKED | omit PPO/envelope; retain bound law, standardized SR/zero and deletion |
+| CEA-06 | S6 | four-role scalar/residual/descriptive CCD organization; target mean notation and non-residual-norm 6.58% semantics fixed; 6/6 jobs drafted | S6 packet + corrected math/final-results/figure-data authorities; current ROI parent `campaign.json` manifests + `dynamic_increment.py::_dense_statistics` (`read`) | share/residual equations retained; four-role `[480,840)` mean window and action-coordinate convention bound; composites/POD artwork | PROSE DRAFTED; exact four-role mean-window binding CLOSED; artwork/provenance OPEN; operator omitted | scalar ledger + residual definition |
+| CEA-07 | S7 | changed-contract role-field/morphology envelope | S7 packet + role metadata/evaluator authority | Vortex/Erase evaluators; role visuals | prose DRAFTED/AUDITED; quantitative BLOCKED; visuals OPEN | qualitative physical-variable role fields only |
+| CEA-08 | S8 | separate steady context, prescribed-circulation outer reference and failed closure | S8 packet + steady README/RESULTS/DERIVATION + `theory.py` unit-$L^2$ basis + strip/finite-Re analyses + figure manifest | Hodge/period equation + adjacent incompatibility accepted; dimensionless $g$ basis/sign/units bound; integrated figure | prose/equations DRAFTED/AUDITED; presentation rounding authorized; exact provenance retained; source/provenance/artwork and mechanism gates OPEN | endpoint/sign table + concise outer boundary |
+| CEA-09 | S9 | explicit author-field placeholder interface only; three evidence cards complete | S9 packet + R4 S9 packet + S9 Figure Plan + S9 audit | author facts, command map, trial/media provenance, processing, permissions and any result import | placeholder format/audit COMPLETE; all substantive evidence and artifact gates OPEN | drop failing media/observation or omit entire section; no S10 import absent targeted audit and reopening |
+| CEA-10 | S10 | established text-only synthesis; five one-to-one evidence cards complete | S10 packet + current retained S2–S8 TeX/audits + S10 audit | none | prose DRAFTED/AUDITED; S1/global interface pending; no artifact/citation dependency | omit S9, Erase and any import not retained after global edits |
+
+## Closure rule
+Only an explicitly accepted author choice is marked closed here. Source, provenance, evaluator, equation-source, caption, artwork, rendering and publication-readiness gates require their own audit and may not be closed by prose drafting or by this ledger.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/00_LIVE_LENGTH_SHARE_LEDGER.md

@@ -0,0 +1,59 @@
+# Round-5 Live Length and Share Ledger
+
+## Authority and counting boundary
+
+No authorized article-wide body-text target is recorded. The Round-2 `8,000–10,000 word-equivalents` range and its 8,700 midpoint are historical planning aids, not an active quota. Section proportions are diagnostics and do not authorize padding or narration of unavailable displays.
+
+The sole current authoritative count is the **post-citation-integration uniform static recount dated 2026-08-11** from the current section files. The counter removes unescaped comments, section/subsection headings, complete display-math bodies, inline-math spans and citation commands; it then removes remaining TeX commands/braces and counts lexical alphanumeric tokens with internal hyphens or apostrophes. A display embedded in a prose source block does not create substantive words. S1–S8 and S10 are substantive; the visibly marked S9 author instructions are classified separately as control tokens. This is a reproducible source-text count, not a typeset word-equivalent estimate.
+
+## Final authoritative current recount
+
+| Section | Reader-facing jobs | Design share | Uniform substantive words | Actual share of 7,434 | Deviation | Disposition note |
+|---|---:|---:|---:|---:|---:|---|
+| S1 Introduction | 5 | 9% | 1,123 | 15.11% | +6.11 pp | citation-integrated framing; Kármán-only field-proximity ceiling explicit |
+| S2 Problem, control and evaluation | 13 prose blocks / 5 jobs | 16% | 1,612 | 21.68% | +5.68 pp | largest prose/equation/citation burden |
+| S3 Capability/progression | 6 | 8% | 578 | 7.78% | −0.22 pp | figure-heavy; artifacts gated |
+| S4 Kármán performance | 9 blocks / 7 jobs | 13% | 793 | 10.67% | −2.33 pp | exact-result owner |
+| S5 Kármán compact law | 11 prose blocks / 5 jobs | 15% | 958 | 12.89% | −2.11 pp | citation-integrated method distinctions |
+| S6 Kármán mean fields/CCD | 7 prose blocks / 6 jobs | 16% | 745 | 10.02% | −5.98 pp | citation-integrated noncausal lineage |
+| S7 Cross-case extensions | 5 | 8% | 499 | 6.71% | −1.29 pp | exact bounded morphology ceiling |
+| S8 Steady context/theory | 6 prose blocks / 5 jobs | 6% | 591 | 7.95% | +1.95 pp | endpoint metric and viscous boundary explicit |
+| S9 Qualitative experiment interface | 3 author fields | 3% held | 0 substantive; 272 control tokens | excluded | n/a | omission branch active |
+| S10 Discussion/conclusion | 5 | 6% | 535 | 7.20% | +1.20 pp | text-only synthesis |
+| **Current substantive total** | **67 prose blocks** | **97% excluding held S9** | **7,434** | **100%** | — | **S9 control text excluded** |
+
+## Protected Kármán analytical centre
+
+S4–S6 contain **2,496 / 7,434 = 33.58%** of current substantive prose. The provisional 42% lower design protection and 44% target correspond to approximately **3,122** and **3,271** words, arithmetic gaps of **626** and **775** words. These are design diagnostics, not drafting instructions. S4–S6 carry five displayed equations, fourteen nonblank displayed rows and 82 inline-math spans, so prose share understates their typeset footprint; no equation-to-word conversion is authorized.
+
+Do not close the diagnostic gap by padding, repeated caveats, derivations, inventory prose or narration of gated artwork.
+
+## Display, equation and citation burden
+
+Current section source contains **12 displayed equation environments, 31 nonblank displayed rows, 143 inline-math spans, 44 citation commands and 64 cited-key occurrences (49 unique active keys)**. `R5_REFERENCES.bib` contains 49 entries and 49 unique keys; citation mapping is **49/49**, with zero undefined keys, zero uncited entries, zero duplicate keys and zero duplicate normalized DOIs. No figure or table environment appears in the section TeX.
+
+| Section | Display equations | Nonblank rows | Inline math | Citation commands | Cited-key occurrences |
+|---|---:|---:|---:|---:|---:|
+| S1 | 0 | 0 | 0 | 19 | 29 |
+| S2 | 5 | 11 | 34 | 10 | 14 |
+| S3 | 0 | 0 | 4 | 0 | 0 |
+| S4 | 0 | 0 | 15 | 0 | 0 |
+| S5 | 2 | 8 | 41 | 5 | 8 |
+| S6 | 3 | 6 | 26 | 4 | 6 |
+| S7 | 0 | 0 | 2 | 0 | 0 |
+| S8 | 2 | 6 | 21 | 6 | 7 |
+| S9 | 0 | 0 | 0 | 0 | 0 |
+| S10 | 0 | 0 | 0 | 0 | 0 |
+
+All 33 post-R5 bibliography additions are actively cited. Citation status alone therefore supplies no prune recommendation; retention still depends on the source-state and metadata gates recorded in `audits/CITATION_EXPANSION_FINAL_AUDIT.md`.
+
+## S9 branch effect
+
+- Current omission branch: **7,434** substantive words; S4–S6 = **33.58%**; 272 S9 control tokens remain excluded.
+- Author supply of 150–300 admitted substantive words: **7,584–7,734** words; unchanged centre = **32.91–32.27%**.
+
+Any S9 prose must pass author-supply, provenance, permission and reopening gates. It is admitted for evidence and article logic, never to manipulate a percentage.
+
+## Current control and confined snapshots
+
+For every future text change, rerun this exact method and replace this table rather than appending a competing snapshot. All earlier method-local or pre-integration figures—including totals **6,092**, **6,152**, **6,537**, **6,176** and **6,303**—are superseded for current arithmetic. They must not be summed or mixed with the authoritative **7,434** count; this paragraph is their only retained current-ledger role.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/00_PARAGRAPH_SECTION_INVENTORY.md

@@ -0,0 +1,19 @@
+# Round-5 Paragraph and Section Inventory
+
+**State:** initial planning inventory. IDs are stable; budgets are provisional ranges, not quotas. Evidence cards are required before drafting.
+
+| Section | Output | Paragraph IDs / jobs | Initial words | Admission/status |
+|---|---|---|---:|---|
+| S1 | `sections/SECTION_01_INTRODUCTION.tex` | S01-P01 fascination/gap; P02 pinball/force mediation; P03 nonlinear feedback/PPO/hooks; P04 realization/evaluation/action+field analysis; P05 question/contribution/contracts bridge | 750–950 | 5/5 drafted/audited; 700 words; evidence-exact Level 1; bibliography/global audit gates open |
+| S2 | `sections/SECTION_02_PROBLEM_METHOD_EVALUATION.tex` | S02-P01 configurations/numerics; P02 action/observations/PPO; P03 target/objective; P04 signal+field evaluation; P05 compact CFD qualification | 1000–1250 | drafted/audited: 1212 words; local source/display gates retained |
+| S3 | `sections/SECTION_03_CAPABILITY_PROGRESSION.tex` | S03-P01 viewing/non-equivalence; P02 steady seed; P03 Kármán centre; P04 Vortex preview; P05 Illusion preview; P06 depth selection | 800–1000 | 6/6 drafted; 662 uniform words (110/110/127/97/98/120); two-figure choice fixed, artifacts gated; S2--S3 audit linked from S2 audit |
+| S4 | `sections/SECTION_04_KARMAN_RESULTS.tex` | S04-P01 tested conditions; P02 five learning realizations; P03 reward/field hinge; P04 reference roles; P05 mean/phase fields; P06 sensor/DTW/action; P07 bounded consequence | 900–1150 | 7/7 drafted; 851 uniform words (150/118/88/153/125/115/102); artifacts gated |
+| S5 | `sections/SECTION_05_KARMAN_SR_LAW.tex` | S05-P01 reduction question; P02 executable law; P03 imposed symmetry; P04 deployment/result; P05 deletion/boundary | 1250–1500 | 5/5 drafted; 614 prose words; evidence cards in audit; source/artifact gates retained |
+| S6 | `sections/SECTION_06_KARMAN_FIELD_CCD.tex` | S06-P01 four-role means; P02 signed differences; P03 scalar accounting; P04 residual; P05 CCD/POD; P06 synthesis/bridge | 1250–1500 | 6/6 drafted; 635 prose words; evidence cards in audit; figures gated |
+| S7 | `sections/SECTION_07_CROSS_CASE_EXTENSIONS.tex` | S07-P01 changed contracts; P02 Vortex; P03 Illusion PPO/negative SR; P04 Erase role fields; P05 strict envelope | 640–800 controlling provisional range | 5/5 drafted and audited; quantitative evaluators blocked; role visuals gated |
+| S8 | `sections/SECTION_08_STEADY_THEORY.tex` | S08-P01 contextual return; P02 finite-Re endpoint; P03 kinematics/viscous hypothesis; P04 Hodge+incompatibility+sign failure; P05 synthesis/bridge | 480–600 controlling provisional range | 5/5 drafted and audited; author choice fixed; source/artwork/mechanism gates retained |
+| S9 | `sections/SECTION_09_EXPERIMENT_PLACEHOLDER.tex` | S09-P01 AUTHOR FIELD purpose/apparatus/execution; P02 AUTHOR FIELD provenance/morphology; P03 AUTHOR FIELD ceiling/threshold | 150–300 placeholder tokens | 3/3 placeholder jobs complete; 0 substantive words; all author/evidence gates open; audited |
+| S10 | `sections/SECTION_10_DISCUSSION_CONCLUSION.tex` | S10-D01 breadth/depth; D02 Kármán action/field synthesis; D03 changed-task+steady boundary; C01 capability/evaluation answer; C02 bounded close | 480–600 controlling provisional range | 5/5 drafted and audited; 544 words (125/140/130/76/73); text-only established synthesis; S1/global interface pending |
+
+## Evidence-card fields
+`paragraph ID | job | claim ceiling | comparator/definition | internal authority and state | metric/estimand/ROI/window if relevant | external citation role/state | display/equation dependency | local caveat | fallback/author field | target/actual words | status`.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/00_R5_FINAL_AUDIT.md

@@ -0,0 +1,100 @@
+# Round-5 Final Audit
+
+**Date:** 2026-08-11  
+**Status:** **CONDITIONAL PASS — admitted review assembly; runtime compile and production artifacts remain gated.**  
+**Scope:** current S1–S8 and S10 TeX; S9 placeholder/audit and omission branch; all section/interface/global audits; current ledgers, bibliography, review master/static preflight, R4 controls and JFM rules. No section prose, literature search/read, CFD, training, evaluator run, rendering, artifact generation or new scientific analysis was performed.
+
+## Decisive determination
+
+The current nine-section reader-facing TeX body is admitted to `master/R5_REVIEW_MASTER.tex` for review assembly. Static TeX structure, exact input order, safe S9 omission, bibliography integration and the bounded scientific claim architecture pass. This is not submission-master approval and not artifact readiness. Runtime compilation is **CONDITIONAL**, not passed: no TeX engine, BibTeX executable or resolver is installed, so engine execution and warning/layout inspection remain a production gate.
+
+## Audit basis
+
+- **PASS:** all 65 R4 jobs have substantive prose or the authorized S9 author-field/omission fallback.
+- **PASS:** current section audits, two wave audits, structure/logic pass, final uniform length/display recount, citation ledger, claim/evidence ledger, master and static preflight were read.
+- **PASS:** internal facts and numbers remain under their current section authority and caveats; this final audit did not independently recompute scientific results or promote mapped artifacts.
+- **PASS:** all 16 approved external identities map to 16 R5 bibliography keys; all active citations resolve and all bibliography entries are cited.
+- **PASS:** final current count is 6,303 substantive words; S9 contributes zero substantive words and 272 excluded control tokens.
+- **CONDITIONAL:** source/provenance/caption/artwork/rendering gates remain open as itemized below; no planned display is publication-ready merely because its prose or equation is admitted.
+
+## Section determination
+
+`PASS` means the current prose and static TeX are admitted to the review assembly under stated scientific ceilings. `CONDITIONAL` retains a local source, normalization or production dependency. `BLOCKED` means no substantive manuscript section is admitted.
+
+| Section | Prose / evidence / citations / static TeX | Final disposition | Remaining local gate |
+|---|---|---|---|
+| S1 | Five-job framing; exact ladder; citations resolve; no display | **PASS** | Final title remains author-provisional; no scientific blocker |
+| S2 | Definitions, action/sign, roles, objective, estimands and compact qualification pass | **CONDITIONAL** | Historical parabolic `U_0/U_{ref}` conversion, exact selected-run reproduction metadata and required method-display production |
+| S3 | Progression and preview/detail split pass; exact Illusion ceiling retained | **CONDITIONAL** | Panel source/provenance/crop/scale/caption/artwork; no scalar Illusion result |
+| S4 | Tested-condition, five-history and target-relative Kármán results pass with separate estimands | **CONDITIONAL** | Redraw/crop/scale/caption, exact spectrum window/action display; `U_0/U_{ref}` remains unbound |
+| S5 | Kármán/cloak-only executable SR, explicit measured-state reflection/force definitions, same-case spatial errors, execution and parent-relative deletion pass | **CONDITIONAL** | Historical velocity conversion, action extraction and figure production; PPO/SR controlled acquisitions remain distinct |
+| S6 | Locally defined T/0/C/D roles, consistent two-component velocity notation, scalar accounting, separately centred residual and descriptive CCD/POD boundary pass | **CONDITIONAL** | Composite provenance/caption/artwork; no causal, energy-share or residual-norm inference |
+| S7 | Changed-contract text passes at the exact bounded execution/morphology ceiling; no scalar change is reported | **CONDITIONAL** | Quantitative Vortex/Erase evaluators remain **BLOCKED**; Illusion scalar and role-visual production remain open |
+| S8 | Separate steady endpoint with explicit cross-sectional two-component L-infinity metric, kinematic hypothesis and failed inviscid closure pass | **CONDITIONAL** | Equation source/typesetting recheck, figure provenance/rendering and all intervention/budget/stability mechanism tests |
+| S9 | Author instructions only; zero manuscript evidence; correctly omitted from master | **BLOCKED** for substantive prose | Author facts, command map, trials/media/provenance/permissions/morphology, targeted audit and explicit reopening |
+| S10 | Three Discussion plus two Conclusion paragraphs; no new evidence/citation/display; no S9 import | **PASS** | No local artifact gate; remains contingent on retained S1–S8 meanings |
+
+## Global findings
+
+### Structure, ownership and style — PASS
+
+Frozen order, paragraph jobs, preview/detail splits, bridges and ownership are intact. Reader-facing prose contains no Legacy/V5 naming, bare `Re100`, package chronology or governance diction. `Kármán`, exact configuration names and `action-correlated correction-field modes` are used consistently. Strong words occur only in explicit denials or bounded hypotheses. No prose edit was warranted.
+
+### Evidence and claim ceilings — PASS WITH PRODUCTION CONDITIONS
+
+- S4 PPO, S5 SR and S6 corrected four-role quantities remain distinct acquisitions/estimands; no replication, validation or triangulation is inferred.
+- S1 now states execution for Kármán, Illusion and Vortex while reserving target-relative field proximity to Kármán; Erase remains unresolved for the same comparison.
+- S5 now defines measured-state reflection parity, signed native solver force coefficients, rear symmetric/antisymmetric combinations, and the same-case target/physical-zero spatial-error lineage before inference.
+- S6 now defines roles $T,0,C,D$, uses $\boldsymbol{u}$ consistently for two-component fields, and keeps the non-paired residual distinct from scalar differences.
+- S7 reports no scalar change for Illusion and morphology only for Vortex/Erase. S8 binds $E_{\infty,\mathrm{vector}}$ to the common-cross-section maximum of the two-component Euclidean difference, not quadrature weighting.
+- The exact Illusion ceiling passes: policy execution toward a declared non-zero target and morphology consistent with retargeting; **no scalar change in target-relative error is reported, and no positive Illusion SR is claimed**.
+- The 93.42%/6.58% quantities are scalar error differences. The separately centred residual is a velocity field, is non-paired, and is not assigned the 6.58% magnitude. S10 preserves this distinction.
+- Article-wide `U_{ref}` is defined by configuration. Historical parabolic `U_0` remains unbound to `U_{ref}`; S3–S5 retain source-normalized notation and use only conversion-free relative reductions where appropriate.
+- Vortex and Erase remain morphology-only where evaluators are absent. Steady remains separate-chain context and failed closure, not mechanism or cross-configuration validation.
+
+### Length and display — PASS AS DIAGNOSTIC
+
+Final authoritative substantive count: **6,303 words**. Shares are S1 11.80%, S2 20.85%, S3 8.98%, S4 12.58%, S5 11.77%, S6 10.01%, S7 7.82%, S8 7.85%, S10 8.33%; S9 is excluded. S4–S6 contain **2,166 words / 34.36%**. The gap from provisional 42%/44% centre diagnostics is 481/607 words, but equation/display density is high and no padding is authorized.
+
+Source burden is 12 display equations, 31 displayed rows, 143 inline-math spans, 14 citation commands and 24 citation-key occurrences. There are no completed figure/table environments; all planned figures/tables retain their production gates.
+
+### Citations and bibliography — PASS 16/16
+
+`R5_REFERENCES.bib` has 16 unique entries. Current TeX cites 16 unique keys in 24 occurrences; unresolved keys = 0; uncited entries = 0; duplicate keys = 0. Literature supports external propositions only and does not verify internal results.
+
+### S9 omission and master input — PASS
+
+The review master actively inputs exactly S1–S8 and S10 in order, explicitly omits `SECTION_09_EXPERIMENT_PLACEHOLDER.tex`, and imports no S9 result into S10. S9 remains an internal author interface rather than manuscript prose.
+
+### Static TeX — PASS; runtime compile — CONDITIONAL
+
+A comment-aware scanner found balanced braces, paired/nested environments and balanced math delimiters for the master plus all ten section files. All active inputs resolve; the JFM class/style support files and bibliography path are statically present under the documented `TEXINPUTS` assumption. No executable exists for `latexmk`, `pdflatex`, `lualatex`, `xelatex`, `bibtex`, `biber` or `kpsewhich`. Runtime compilation, citation/reference warning inspection, class/package/font/line-number behavior, bad boxes and page layout therefore remain an explicit production gate.
+
+## Overall TeX assembly admission
+
+**CONDITIONAL PASS.** The current S1–S8/S10 body is admitted to the R5 review master. The condition is runtime compilation and production inspection, plus local source/artifact gates for dependent displays. S9 substantive content is blocked and omitted. This audit does not authorize migration to the submission master or claim publication readiness.
+
+## Post-R5 Figure eligibility
+
+Post-R5 figure production may begin **only under the existing per-display admission workflow**. Eligible work is source/provenance audit, processing-contract completion, redraw/layout, caption-ceiling review and static validation for already licensed displays. No figure is automatically eligible for publication:
+
+- **Eligible to enter gated production:** S2 method schematics/tables; accepted S3 composition; S4 Kármán displays; S5 law/deletion and Target/SR/zero fallback; S6 mean/residual/CCD composition; S7 role-field visuals only; accepted S8 endpoint/outer-reference/sign-failure composition.
+- **Blocked from quantitative production:** S7 Vortex/Erase efficacy/scalars; any Illusion scalar not exact-role/evaluator bound; mismatched PPO/SR common-comparison panel; MFS mechanism/validation panel; any S9 experiment panel absent author provenance/permission/reopening.
+- **No post-R5 figure work:** S1 and S10 own no empirical display under the current contract; S9 omission remains active.
+
+## Remaining gates
+
+1. Runtime TeX/BibTeX compilation and warning/layout inspection in an environment with the documented JFM class/style path.
+2. Author confirmation or continued provisional use of the article title before submission-master migration.
+3. Historical parabolic `U_0/U_{ref}` identification before any physical conversion or inherited Reynolds relabeling.
+4. Per-display exact source/provenance, roles/comparator, estimand, ROI/mask/window, crop/scale, caption, rendering and final production review.
+5. Versioned evaluators before quantitative Vortex/Erase; exact binding before any Illusion scalar.
+6. Author supply, physical command map, trial/media provenance, permissions, morphology wording, targeted audit and reopening before any S9 content or S10 import.
+
+## Knowledge and computation closeout
+
+Nowledge was used with three narrow `DynamisLab JFM Round 5` searches for final-audit, claim-ceiling and S9/master decisions; repository controls superseded recall. Working Memory, thread recall and `mem_fs` were not used. Current R5 TeX/audits/ledgers/master/bibliography and R4 controls/rules were read; current source was statically counted and structurally scanned. Previously recorded literature states were read as repository metadata only; no literature search or PDF read occurred. Internal artifacts were not independently reverified, and no CFD, training, evaluator, rendering, artifact generation or new scientific analysis ran.
+
+## Concise final closeout memory proposal
+
+> DynamisLab JFM Round 5 closed on 2026-08-11 with a conditional review-assembly pass. Current S1–S8/S10 substantive TeX totals 6,303 uniform words; S9 is an omitted author interface with 0 substantive words and 272 control tokens. Static TeX and bibliography integration pass 16/16; runtime compile remains conditional because no TeX/BibTeX toolchain is installed. Exact ceilings remain: Illusion PPO execution/morphology only with no reported scalar target-relative error change or positive SR; Vortex/Erase quantitative claims blocked; S4/S5/S6 estimands distinct; 6.58% is a scalar increment, not residual magnitude; historical parabolic U0/Uref conversion unbound. Post-R5 figures may enter gated production, but no artifact is publication-ready and blocked evaluator/experiment/MFS/mismatched-comparison panels remain ineligible.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/01_CONTEXT_PACKET_A_S01_S02.md

@@ -0,0 +1,17 @@
+# Canonical Context Packet A — Sections 1–2
+
+**Version:** v1 | **Unit:** A | **Order:** S2 then S1 | **Outputs:** `sections/SECTION_02_PROBLEM_METHOD_EVALUATION.tex`, `audits/SECTION_02_AUDIT.md`, then `sections/SECTION_01_INTRODUCTION.tex`, `audits/SECTION_01_AUDIT.md`.
+
+## Controlling reads
+R4 global ledger, closeout audit and README; `SECTION_02_PROBLEM_METHOD_EVALUATION.md`; `SECTION_01_INTRODUCTION_FREEZE.md`; current R5 rules; exact authorities named by each paragraph job. Recall narrowly, then verify.
+
+## Contract
+- S2: five jobs—configurations/numerics; action/observations/PPO; target/objective; independent evaluation; compact CFD qualification. Frozen architecture, conditional local sources/displays.
+- S1: five-paragraph frozen framing/contribution architecture, written after evidence-owning sections. Literature may support external framing only.
+- Preserve configuration names, article-wide Reynolds convention and canonical-versus-earlier implementation boundary. No canonical rerun/equivalence claim.
+
+## Gates and exclusions
+Bind every run-specific value, inherited Reynolds label, policy/normalizer/target and qualification row before printing. Method schematics/tables remain production-gated. S1 owns no detailed values or empirical display. No final-master work, evidence creation, CFD/training/analysis or invented citation keys.
+
+## Acceptance
+Every paragraph has an evidence card, comparator/definition before inference, local caveat, owned equation/display dependency, fallback, word count and bridge. S2→S3 interface is audited before section-pass.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/02_CONTEXT_PACKET_B_S03_S04.md

@@ -0,0 +1,15 @@
+# Canonical Context Packet B — Sections 3–4
+
+**Version:** v1 | **Unit:** B | **Outputs:** `sections/SECTION_03_CAPABILITY_PROGRESSION.tex`, `audits/SECTION_03_AUDIT.md`, `sections/SECTION_04_KARMAN_RESULTS.tex`, `audits/SECTION_04_AUDIT.md`.
+
+## Controlling reads
+R4 global ledger, closeout audit and README; current S3 and S4 packets; S3/S4 Figure Plans; current R5 rules; exact source authorities named by the packets.
+
+## Accepted decision and jobs
+S3 author choice is closed at the two-figure safe default: S3-1 steady two-component profile; S3-2 Kármán-dominant target/controlled/physical-zero wake composition with one source-bound Vortex and one 0.75-radius Illusion preview. Its six jobs remain viewing contract, steady seed, Kármán centre, event preview, changed-target preview and depth selection. S4 owns tested-condition learning and reference-wake field/trajectory detail.
+
+## Ceilings and gates
+The acceptance fixes interface/count, not panels, source files, provenance, crop, scale, caption, rendering or approval. S3 cannot pool efficacy or inherit S4/S7 detail. Vortex is morphology-only; Illusion is positive PPO only. S4 figures, field redraw, hash/scale/crop and shared spectrum window remain gated. Reward, L2 and DTW remain distinct.
+
+## Acceptance
+Evidence cards identify role, configuration, temporal coordinate and owner. S3 preview/details are deduplicated against S4 and S7. Missing panels use the recorded omission/source-bound fallback without changing the two-figure interface.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/03_CONTEXT_PACKET_C_S05_S06.md

@@ -0,0 +1,15 @@
+# Canonical Context Packet C — Sections 5–6
+
+**Version:** v1 | **Unit:** C | **Outputs:** `sections/SECTION_05_KARMAN_SR_LAW.tex`, `audits/SECTION_05_AUDIT.md`, `sections/SECTION_06_KARMAN_FIELD_CCD.tex`, `audits/SECTION_06_AUDIT.md`.
+
+## Controlling reads
+R4 global ledger, closeout audit and README; current S5/S6 packets and Figure Plans; current R5 rules; exact SR, DTW and corrected-field authorities named locally.
+
+## Jobs and ownership
+S5 owns the executable Kármán/cloak-only surrogate, imposed symmetry, deployment scope and parent-relative deletion. S6 owns four-role means, exact scalar accounting, non-paired residual and descriptive `action-correlated correction-field modes` with the POD boundary. Preserve five S5 and six S6 jobs.
+
+## Hard boundaries
+Positive SR is Kármán/cloak-only; symmetry is imposed after PPO data collection and proves neither PPO nor plant symmetry. Historical `U0` conversion remains source-gated. Omit PPO from mismatched same-target panel. S4 PPO L2, S5 standardized SR L2 and S6 corrected mean estimand are non-equivalent. S6 percentages are measured scalar-error accounting, not energy, causality or physical contribution. CCD is descriptive/noncausal; POD is supplemental.
+
+## Acceptance
+All formulas, actions, role order, windows and values bind exact authority. Figures remain production-gated. Audit S4–S6 estimand separation before section-pass.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/04_CONTEXT_PACKET_D_S07_S08.md

@@ -0,0 +1,15 @@
+# Canonical Context Packet D — Sections 7–8
+
+**Version:** v1 | **Unit:** D | **Outputs:** `sections/SECTION_07_CROSS_CASE_EXTENSIONS.tex`, `audits/SECTION_07_AUDIT.md`, `sections/SECTION_08_STEADY_THEORY.tex`, `audits/SECTION_08_AUDIT.md`.
+
+## Controlling reads
+R4 global ledger, closeout audit and README; current S7/S8 packets and Figure Plans; current R5 rules; exact role-field, steady endpoint and theory authorities named locally.
+
+## Accepted decision and jobs
+S7 retains five changed-contract jobs with role-field/morphology ceilings. S8 author choice is closed: display the multi-cylinder Hodge/period outer solution with prescribed circulations, place the one-circle rotating-no-slip incompatibility adjacent, and use one integrated CFD + prescribed-circulation outer-reference + sign/closure-failure figure. Preserve S8's five paragraph jobs and kinematics-first ordering.
+
+## Gates and exclusions
+S7 quantitative Vortex/Erase efficacy remains blocked without versioned evaluators; role visuals remain gated. S8 choice does not approve equation artwork, panels, sources, provenance, caption, crop/scale, redraw or rendering. The outer field is a bounded prescribed-circulation reference, not finite-Re mechanism validation. Steady is separate-chain context; no continuation, amplitude agreement, optimum, stability, causal mechanism or momentum restoration.
+
+## Acceptance
+The S8 outer-reference paragraph places prescribed-period meaning, no-slip incompatibility and sign failure together. Audit S7→S8 non-transfer and S8→S9 quantitative-boundary bridge.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/05_CONTEXT_PACKET_E_S09_S10.md

@@ -0,0 +1,15 @@
+# Canonical Context Packet E — Sections 9–10
+
+**Version:** v1 | **Unit:** E | **Order:** S9 placeholder then S10 | **Outputs:** `sections/SECTION_09_EXPERIMENT_PLACEHOLDER.tex`, `audits/SECTION_09_AUDIT.md`, `sections/SECTION_10_DISCUSSION_CONCLUSION.tex`, `audits/SECTION_10_AUDIT.md`.
+
+## Controlling reads
+R4 global ledger, closeout audit and README; current S9/S10 packets and S9 Figure Plan; current R5 rules. S9 author fields are the only admitted TeX content until reopening.
+
+## S9 placeholder contract
+Create explicit TeX placeholders for: purpose; apparatus/facility/geometry; open-loop command path and units/sign/timing; trial/case IDs; raw media/frame/view/processing/permission; comparator-led morphology wording; confounds/limits; next evidentiary threshold. Every placeholder must be visibly marked `AUTHOR FIELD` and contain no asserted apparatus fact, execution verb, trial identity, provenance, morphology observation or experiment result. It may preserve structural transitions as instructions, not manuscript claims.
+
+## S10 contract
+Three Discussion plus two short Conclusion paragraphs, text-only, established Sections 1–8 synthesis. No new evidence, value, citation, equation, display, limitations programme or Section-9 result. A Section-9 observation can enter only after author supply, targeted provenance audit and explicit reopening.
+
+## Gates and acceptance
+All S9 author/media/command/provenance/permission gates remain open; omission remains fallback. The audit checks that placeholders cannot be mistaken for results and S10 imports only retained established meanings.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/R5_REFERENCES.bib

@@ -0,0 +1,503 @@
+% Round-5-owned bibliography for reader-facing external propositions only.
+% Do not use these sources to support DynamisLab results, values, artifacts, or readiness.
+
+@article{PendryEtAl2006Controlling,
+  author  = {Pendry, J. B. and Schurig, D. and Smith, D. R.},
+  title   = {Controlling electromagnetic fields},
+  journal = {Science},
+  year    = {2006},
+  volume  = {312},
+  number  = {5781},
+  pages   = {1780--1782},
+  doi     = {10.1126/science.1125907}
+}
+
+@article{MaEtAl2013ActiveCloaking,
+  author  = {Ma, Qian and Mei, Zhong Lei and Zhu, Shou Kui and Jin, Tian Yu and Cui, Tie Jun},
+  title   = {Experiments on active cloaking and illusion for {Laplace} equation},
+  journal = {Physical Review Letters},
+  year    = {2013},
+  volume  = {111},
+  number  = {17},
+  pages   = {173901},
+  doi     = {10.1103/PhysRevLett.111.173901}
+}
+
+@article{UrzhumovSmith2012ActiveCloaks,
+  author  = {Urzhumov, Yaroslav A. and Smith, David R.},
+  title   = {Flow stabilization with active hydrodynamic cloaks},
+  journal = {Physical Review E},
+  year    = {2012},
+  volume  = {86},
+  number  = {5},
+  pages   = {056313},
+  doi     = {10.1103/PhysRevE.86.056313}
+}
+
+@article{DengEtAl2020PinballBifurcations,
+  author  = {Deng, Nan and Noack, Bernd R. and Morzy{\'n}ski, Marek and Pastur, Luc R.},
+  title   = {Low-order model for successive bifurcations of the fluidic pinball},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2020},
+  volume  = {884},
+  pages   = {A37},
+  doi     = {10.1017/jfm.2019.959}
+}
+
+@article{CornejoMacedaEtAl2021PinballControl,
+  author  = {Cornejo Maceda, G. Y. and Li, Y. and Lusseyran, F. and Morzy{\'n}ski, M. and Noack, B. R.},
+  title   = {Stabilization of the fluidic pinball with gradient-enriched machine learning control},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2021},
+  volume  = {917},
+  pages   = {A42},
+  doi     = {10.1017/jfm.2021.301}
+}
+
+@article{RabaultEtAl2019DRLFlowControl,
+  author  = {Rabault, Jean and Kuchta, Miroslav and Jensen, Atle and R{\'e}glade, Ulysse and Cerardi, Nicolas},
+  title   = {Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2019},
+  volume  = {865},
+  pages   = {281--302},
+  doi     = {10.1017/jfm.2019.62}
+}
+
+@article{SchulmanEtAl2017PPO,
+  author        = {Schulman, John and Wolski, Filip and Dhariwal, Prafulla and Radford, Alec and Klimov, Oleg},
+  title         = {Proximal policy optimization algorithms},
+  journal       = {arXiv preprint arXiv:1707.06347},
+  year          = {2017},
+  eprint        = {1707.06347},
+  archiveprefix = {arXiv},
+  primaryclass  = {cs.LG}
+}
+
+
+@article{ParisEtAl2021RobustFlowControl,
+  author  = {Paris, Romain and Beneddine, Samir and Dandois, Julien},
+  title   = {Robust flow control and optimal sensor placement using deep reinforcement learning},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2021},
+  volume  = {913},
+  pages   = {A25},
+  doi     = {10.1017/jfm.2020.1170}
+}
+
+@article{GautierEtAl2015ClosedLoop,
+  author  = {Gautier, N. and Aider, J.-L. and Duriez, T. and Noack, B. R. and Segond, M. and Abel, M.},
+  title   = {Closed-loop separation control using machine learning},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2015},
+  volume  = {770},
+  pages   = {442--457},
+  doi     = {10.1017/jfm.2015.95}
+}
+
+@article{LoiseauEtAl2018SparseROM,
+  author  = {Loiseau, Jean-Christophe and Noack, Bernd R. and Brunton, Steven L.},
+  title   = {Sparse reduced-order modelling: sensor-based dynamics to full-state estimation},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2018},
+  volume  = {844},
+  pages   = {459--490},
+  doi     = {10.1017/jfm.2018.147}
+}
+
+@article{RaibaudoEtAl2020PinballControl,
+  author  = {Raibaudo, C{\'e}dric and Zhong, Peng and Noack, Bernd R. and Martinuzzi, Robert J.},
+  title   = {Machine learning strategies applied to the control of a fluidic pinball},
+  journal = {Physics of Fluids},
+  year    = {2020},
+  volume  = {32},
+  number  = {1},
+  pages   = {015108},
+  doi     = {10.1063/1.5127202}
+}
+
+@article{LiEtAl2022FlowEstimation,
+  author  = {Li, Songqi and Li, Wenpeng and Noack, Bernd R.},
+  title   = {Machine-learned control-oriented flow estimation for multi-actuator multi-sensor systems exemplified for the fluidic pinball},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2022},
+  volume  = {952},
+  pages   = {A36},
+  doi     = {10.1017/jfm.2022.908}
+}
+
+@article{SchlegelEtAl2012LeastOrder,
+  author  = {Schlegel, Michael and Noack, Bernd R. and Jordan, Peter and Dillmann, Andreas and Gr{\"o}schel, Elmar and Schr{\"o}der, Wolfgang and Wei, Mingjun and Freund, Jonathan B. and Lehmann, Oliver and Tadmor, Gilead},
+  title   = {On least-order flow representations for aerodynamics and aeroacoustics},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2012},
+  volume  = {697},
+  pages   = {367--398},
+  doi     = {10.1017/jfm.2012.70}
+}
+
+@article{Boree2003ExtendedPOD,
+  author  = {Bor{\'e}e, J.},
+  title   = {Extended proper orthogonal decomposition: a tool to analyse correlated events in turbulent flows},
+  journal = {Experiments in Fluids},
+  year    = {2003},
+  volume  = {35},
+  number  = {2},
+  pages   = {188--192},
+  doi     = {10.1007/S00348-003-0656-3}
+}
+
+@article{Crowdy2006MultipleCylinders,
+  author  = {Crowdy, Darren G.},
+  title   = {Analytical solutions for uniform potential flow past multiple cylinders},
+  journal = {European Journal of Mechanics B/Fluids},
+  year    = {2006},
+  volume  = {25},
+  number  = {4},
+  pages   = {459--470},
+  doi     = {10.1016/j.euromechflu.2005.11.005}
+}
+
+@article{UedaEtAl2003RotatingCylinders,
+  author  = {Ueda, Y. and Sellier, A. and Kida, T. and Nakanishi, M.},
+  title   = {On the low-{Reynolds}-number flow about two rotating circular cylinders},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2003},
+  volume  = {495},
+  pages   = {255--281},
+  doi     = {10.1017/S002211200300627X}
+}
+
+% Post-R5 high-confidence local-source wave (metadata provisional where audited).
+@article{Leonhardt2006OpticalMapping,
+  author  = {Leonhardt, Ulf},
+  title   = {Optical conformal mapping},
+  journal = {Science},
+  year    = {2006},
+  volume  = {312},
+  number  = {5781},
+  pages   = {1777--1780},
+  doi     = {10.1126/science.1126493}
+}
+
+@article{Miller2006PerfectActiveCloaking,
+  author  = {Miller, David A. B.},
+  title   = {On perfect cloaking},
+  journal = {Optics Express},
+  year    = {2006},
+  volume  = {14},
+  number  = {25},
+  pages   = {12457--12466}
+}
+
+@article{VasquezEtAl2009ActiveExterior,
+  author  = {Guevara Vasquez, Fernando and Milton, Graeme W. and Onofrei, Daniel},
+  title   = {Active exterior cloaking for the {2D} {Laplace} and {Helmholtz} equations},
+  journal = {Physical Review Letters},
+  year    = {2009},
+  volume  = {103},
+  number  = {7},
+  pages   = {073901},
+  doi     = {10.1103/PhysRevLett.103.073901}
+}
+
+@article{LinEtAl2021ActiveAcousticIllusions,
+  author  = {Lin, Cikai and Liu, Daipei and Eggler, Daniel and Kessissoglou, Nicole},
+  title   = {Active acoustic cloaking and illusions of sound-hard bodies using the boundary element method},
+  journal = {The Journal of the Acoustical Society of America},
+  year    = {2021},
+  volume  = {149},
+  number  = {3},
+  pages   = {1803--1812}
+}
+
+@article{ChenEtAl2024MultilayerHydrodynamicCloak,
+  author  = {Chen, Mengyao and Shen, Xiangying and Xu, Lei},
+  title   = {A multilayered homogeneous hydrodynamic cloak realized in a free fluid flow},
+  journal = {Applied Physics Letters},
+  year    = {2024},
+  volume  = {125},
+  number  = {17}
+}
+
+@article{RenEtAl2021HydrodynamicStealthDRL,
+  author  = {Ren, Feng and Wang, Chenglei and Tang, Hui},
+  title   = {Bluff body uses deep-reinforcement-learning trained active flow control to achieve hydrodynamic stealth},
+  journal = {Physics of Fluids},
+  year    = {2021},
+  volume  = {33},
+  number  = {9},
+  pages   = {093602},
+  doi     = {10.1063/5.0060690}
+}
+
+@article{DengEtAl2021PinballForceModel,
+  author  = {Deng, Nan and Noack, Bernd R. and Morzy{\'n}ski, Marek and Pastur, Luc R.},
+  title   = {Galerkin force model for transient and post-transient dynamics of the fluidic pinball},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2021},
+  volume  = {918},
+  pages   = {A4},
+  doi     = {10.1017/jfm.2021.299}
+}
+
+@article{FengEtAl2023PinballForceControl,
+  author  = {Feng, Haodong and Wang, Yue and Xiang, Hui and Jin, Zhiyang and Fan, Dixia},
+  title   = {How to control hydrodynamic force on fluidic pinball via deep reinforcement learning},
+  journal = {Physics of Fluids},
+  year    = {2023},
+  volume  = {35},
+  number  = {4},
+  pages   = {045137},
+  doi     = {10.1063/5.0142949}
+}
+
+@article{BeemTriantafyllou2015WakeSensing,
+  author  = {Beem, Heather R. and Triantafyllou, Michael S.},
+  title   = {Wake-induced `slaloming' response explains exquisite sensitivity of seal whisker-like sensors},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2015},
+  volume  = {783},
+  pages   = {306--322}
+}
+
+@article{SakoeChiba1978DynamicProgramming,
+  author  = {Sakoe, H. and Chiba, S.},
+  title   = {Dynamic programming algorithm optimization for spoken word recognition},
+  journal = {IEEE Transactions on Acoustics, Speech, and Signal Processing},
+  year    = {1978},
+  volume  = {26},
+  number  = {1},
+  pages   = {43--49},
+  doi     = {10.1109/TASSP.1978.1163055}
+}
+
+@book{Muller2015FundamentalsMusicProcessing,
+  author    = {M{\"u}ller, Meinard},
+  title     = {Fundamentals of Music Processing: Audio, Analysis, Algorithms, Applications},
+  publisher = {Springer},
+  year      = {2015},
+  volume    = {5}
+}
+
+@article{FengWang2010SyntheticJetCylinder,
+  author  = {Feng, Li Hao and Wang, Jin Jun},
+  title   = {Circular cylinder vortex-synchronization control with a synthetic jet positioned at the rear stagnation point},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2010},
+  volume  = {662},
+  pages   = {232--259}
+}
+
+@article{LiEtAl2017GeneticProgrammingControl,
+  author  = {Li, R. and Noack, B. R. and Cordier, L. and Bor{\'e}e, J. and Harambat, F.},
+  title   = {Drag reduction of a car model by linear genetic programming control},
+  journal = {Experiments in Fluids},
+  year    = {2017},
+  volume  = {58},
+  pages   = {103}
+}
+
+@article{BruntonEtAl2016SINDy,
+  author  = {Brunton, Steven L. and Proctor, Joshua L. and Kutz, J. Nathan},
+  title   = {Discovering governing equations from data by sparse identification of nonlinear dynamical systems},
+  journal = {Proceedings of the National Academy of Sciences},
+  year    = {2016},
+  volume  = {113},
+  pages   = {3932--3937},
+  doi     = {10.1073/pnas.1517384113}
+}
+
+@article{Cranmer2023PySR,
+  author        = {Cranmer, Miles},
+  title         = {Interpretable machine learning for science with {PySR} and {SymbolicRegression.jl}},
+  journal       = {arXiv preprint arXiv:2305.01582},
+  year          = {2023},
+  eprint        = {2305.01582},
+  archiveprefix = {arXiv}
+}
+
+% Post-R5 Undermind full-PDF verified gap wave.
+@article{FanEtAl2020BluffBodyRL,
+  author  = {Fan, Dixia and Yang, Liu and Wang, Zhicheng and Triantafyllou, Michael S. and Karniadakis, George E.},
+  title   = {Reinforcement learning for bluff body active flow control in experiments and simulations},
+  journal = {Proceedings of the National Academy of Sciences},
+  year    = {2020},
+  volume  = {117},
+  number  = {42},
+  pages   = {26091--26098},
+  doi     = {10.1073/pnas.2004939117}
+}
+
+@article{LiZhang2022ConfinedWakeRL,
+  author  = {Li, Ji-chao and Zhang, Mengqi},
+  title   = {Reinforcement-learning-based control of confined cylinder wakes with stability analyses},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2022},
+  volume  = {932},
+  pages   = {A44},
+  doi     = {10.1017/jfm.2021.1045}
+}
+
+@article{MarraEtAl2024ActuationManifold,
+  author  = {Marra, L. and Cornejo Maceda, G. Y. and Meil{\'a}n-Vila, A. and Guerrero, V. and Rashwan, S. and Noack, B. R. and Discetti, S. and Ianiro, A.},
+  title   = {Actuation manifold from snapshot data},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2024},
+  volume  = {996},
+  pages   = {A26},
+  doi     = {10.1017/jfm.2024.593}
+}
+
+@article{JinEtAl2022SensorActuatorPlacement,
+  author  = {Jin, B. and Illingworth, S. J. and Sandberg, R. D.},
+  title   = {Optimal sensor and actuator placement for feedback control of vortex shedding},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2022},
+  volume  = {932},
+  pages   = {A2},
+  doi     = {10.1017/jfm.2021.948}
+}
+
+@article{GongEtAl2020VortexEstimation,
+  author  = {Gong, Jiwen and Monty, Jason P. and Illingworth, Simon J.},
+  title   = {Model-based estimation of vortex shedding in unsteady cylinder wakes},
+  journal = {Physical Review Fluids},
+  year    = {2020},
+  volume  = {5},
+  pages   = {023901},
+  doi     = {10.1103/PhysRevFluids.5.023901}
+}
+
+@article{NairEtAl2021PhaseFlowControl,
+  author  = {Nair, Aditya G. and Taira, Kunihiko and Brunton, Bingni W. and Brunton, Steven L.},
+  title   = {Phase-based control of periodic flows},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2021},
+  volume  = {927},
+  doi     = {10.1017/jfm.2021.735}
+}
+
+@article{WangEtAl2024DynamicFeatureDRL,
+  author  = {Wang, Qiulei and Yan, Dong and Hu, Guodong and Chen, Gang and Rabault, Jean and Noack, Bernd R.},
+  title   = {Dynamic feature-based deep reinforcement learning for flow control of circular cylinder with sparse surface pressure sensing},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2024},
+  volume  = {988},
+  pages   = {A4},
+  doi     = {10.1017/jfm.2024.333}
+}
+
+@article{DebienEtAl2016GeneticProgrammingRamp,
+  author  = {Debien, Antoine and von Krbek, Kai A. F. F. and Mazellier, Nicolas and Duriez, Thomas and Cordier, Laurent and Noack, Bernd R. and Abel, Markus W. and Kourta, Azeddine},
+  title   = {Closed-loop separation control over a sharp edge ramp using genetic programming},
+  journal = {Experiments in Fluids},
+  year    = {2016},
+  volume  = {57},
+  number  = {3},
+  doi     = {10.1007/s00348-016-2126-8}
+}
+
+@article{CastellanosEtAl2022FewSensorMLControl,
+  author  = {Castellanos, R. and Cornejo Maceda, G. Y. and de la Fuente, I. and Noack, B. R. and Ianiro, A. and Discetti, S.},
+  title   = {Machine-learning flow control with few sensor feedback and measurement noise},
+  journal = {Physics of Fluids},
+  year    = {2022},
+  volume  = {34},
+  number  = {4},
+  pages   = {047118},
+  doi     = {10.1063/5.0087208}
+}
+
+@article{RenEtAl2019MachineLearningVIV,
+  author  = {Ren, Feng and Wang, Chenglei and Tang, Hui},
+  title   = {Active control of vortex-induced vibration of a circular cylinder using machine learning},
+  journal = {Physics of Fluids},
+  year    = {2019},
+  volume  = {31},
+  pages   = {093601},
+  doi     = {10.1063/1.5115258}
+}
+
+@article{ZolmanEtAl2025SINDyRL,
+  author  = {Zolman, Nicholas and Lagemann, Christian and Fasel, Urban and Kutz, J. Nathan and Brunton, Steven L.},
+  title   = {{SINDy-RL} for interpretable and efficient model-based reinforcement learning},
+  journal = {Nature Communications},
+  year    = {2025},
+  volume  = {16},
+  pages   = {10714},
+  doi     = {10.1038/s41467-025-65738-4}
+}
+
+@inproceedings{JordanEtAl2007ObservableJetModes,
+  author    = {Jordan, P. and Schlegel, M. and Stalnov, O. and Noack, B. R. and Tinney, C. E.},
+  title     = {Identifying noisy and quiet modes in a jet},
+  booktitle = {13th AIAA/CEAS Aeroacoustics Conference},
+  year      = {2007},
+  number    = {AIAA 2007-3602},
+  doi       = {10.2514/6.2007-3602}
+}
+
+@article{TadmorEtAl2010CylinderMeanField,
+  author  = {Tadmor, Gilead and Lehmann, Oliver and Noack, Bernd R. and Morzy{\'n}ski, Marek},
+  title   = {Mean field representation of the natural and actuated cylinder wake},
+  journal = {Physics of Fluids},
+  year    = {2010},
+  volume  = {22},
+  number  = {3},
+  pages   = {034102},
+  doi     = {10.1063/1.3298960}
+}
+
+@article{DiscettiEtAl2018CorrelatedFieldEstimation,
+  author  = {Discetti, Stefano and Raiola, Marco and Ianiro, Andrea},
+  title   = {Estimation of time-resolved turbulent fields through correlation of non-time-resolved field measurements and time-resolved point measurements},
+  journal = {Experimental Thermal and Fluid Science},
+  year    = {2018},
+  volume  = {93},
+  pages   = {119--130},
+  doi     = {10.1016/j.expthermflusci.2017.12.011}
+}
+
+@article{Mittal2001RotatingControlCylinders,
+  author  = {Mittal, Sanjay},
+  title   = {Control of flow past bluff bodies using rotating control cylinders},
+  journal = {Journal of Fluids and Structures},
+  year    = {2001},
+  volume  = {15},
+  number  = {2},
+  pages   = {291--326},
+  doi     = {10.1006/JFLS.2000.0337}
+}
+
+@article{ChanEtAl2011CounterRotatingCylinders,
+  author  = {Chan, A. S. and Dewey, P. A. and Jameson, A. and Liang, C. and Smits, A. J.},
+  title   = {Vortex suppression and drag reduction in the wake of counter-rotating cylinders},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2011},
+  volume  = {679},
+  pages   = {343--382},
+  doi     = {10.1017/jfm.2011.134}
+}
+
+@article{SierraEtAl2020RotatingCylinderBifurcations,
+  author  = {Sierra, J. and Fabre, D. and Citro, V. and Giannetti, F.},
+  title   = {Bifurcation scenario in the two-dimensional laminar flow past a rotating cylinder},
+  journal = {Journal of Fluid Mechanics},
+  year    = {2020},
+  volume  = {905},
+  pages   = {A2},
+  doi     = {10.1017/jfm.2020.692}
+}
+
+@article{Watson1996TwoRotatingCylinders,
+  author  = {Watson, E. J.},
+  title   = {Slow viscous flow past two rotating cylinders},
+  journal = {The Quarterly Journal of Mechanics and Applied Mathematics},
+  year    = {1996},
+  volume  = {49},
+  number  = {2},
+  pages   = {195--216},
+  doi     = {10.1093/QJMAM/49.2.195}
+}
+