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 Remove outdated figures and images that are no longer relevant to the current manuscript.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.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/CITATION_EXPANSION_BIBLIOGRAPHY_AUDIT.md

@@ -0,0 +1,99 @@
+# Citation Expansion Bibliography Audit
+
+**Project / unit:** DynamisLab JFM post-R5 / bibliography and key normalization only.  
+**Owned outputs:** `../R5_REFERENCES.bib`, `../00_CITATION_CANDIDATE_LEDGER.md`, this audit, and the stale-count correction in `../master/STATIC_PREFLIGHT.md`.  
+**Excluded:** section prose, citation insertion, final disposition, CFD, literature research, dependency installation and new scientific analysis.
+
+## Inventory and admission result
+
+`R5_REFERENCES.bib` now contains **49 entries and 49 unique keys**:
+
+- **16 preserved records** from the completed R5 citation integration;
+- **15 high-confidence local-wave records** from `CITATION_EXPANSION_LOCAL_SOURCE_AUDIT.md`;
+- **18 full-PDF-verified Undermind records** from `CITATION_EXPANSION_UNDERMIND_GAP_AUDIT.md`.
+
+The net change is **+33 records**. The four optional local rows were not added. No existing record was deleted, renamed or rewritten. The later authorized prose-integration pass is complete: all 33 additions are now cited, so current section prose activates all 49 records.
+
+## Net-new key map
+
+| Wave | Final key | Identity | DOI / stable identifier |
+|---|---|---|---|
+| Local | `Leonhardt2006OpticalMapping` | Leonhardt, “Optical conformal mapping” (2006) | `10.1126/science.1126493` |
+| Local | `Miller2006PerfectActiveCloaking` | Miller, “On perfect cloaking” (2006) | none recorded |
+| Local | `VasquezEtAl2009ActiveExterior` | Guevara Vasquez, Milton & Onofrei (2009) | `10.1103/PhysRevLett.103.073901` |
+| Local | `LinEtAl2021ActiveAcousticIllusions` | Lin et al. (2021) | none recorded |
+| Local | `ChenEtAl2024MultilayerHydrodynamicCloak` | Chen, Shen & Xu (2024) | none recorded |
+| Local | `RenEtAl2021HydrodynamicStealthDRL` | Ren, Wang & Tang (2021) | `10.1063/5.0060690` |
+| Local | `DengEtAl2021PinballForceModel` | Deng et al. (2021) | `10.1017/jfm.2021.299` |
+| Local | `FengEtAl2023PinballForceControl` | Feng et al. (2023) | `10.1063/5.0142949` |
+| Local | `BeemTriantafyllou2015WakeSensing` | Beem & Triantafyllou (2015) | none recorded |
+| Local | `SakoeChiba1978DynamicProgramming` | Sakoe & Chiba (1978) | `10.1109/TASSP.1978.1163055` |
+| Local | `Muller2015FundamentalsMusicProcessing` | Müller (2015) | none recorded |
+| Local | `FengWang2010SyntheticJetCylinder` | Feng & Wang (2010) | none recorded |
+| Local | `LiEtAl2017GeneticProgrammingControl` | Li et al. (2017) | none recorded |
+| Local | `BruntonEtAl2016SINDy` | Brunton, Proctor & Kutz (2016) | `10.1073/pnas.1517384113` |
+| Local | `Cranmer2023PySR` | Cranmer (2023) | arXiv `2305.01582` |
+| Undermind | `FanEtAl2020BluffBodyRL` | Fan et al. (2020) | `10.1073/pnas.2004939117` |
+| Undermind | `LiZhang2022ConfinedWakeRL` | Li & Zhang (2022) | `10.1017/jfm.2021.1045` |
+| Undermind | `MarraEtAl2024ActuationManifold` | Marra et al. (2024) | `10.1017/jfm.2024.593` |
+| Undermind | `JinEtAl2022SensorActuatorPlacement` | Jin, Illingworth & Sandberg (2022) | `10.1017/jfm.2021.948` |
+| Undermind | `GongEtAl2020VortexEstimation` | Gong, Monty & Illingworth (2020) | `10.1103/PhysRevFluids.5.023901` |
+| Undermind | `NairEtAl2021PhaseFlowControl` | Nair et al. (2021) | `10.1017/jfm.2021.735` |
+| Undermind | `WangEtAl2024DynamicFeatureDRL` | Wang et al. (2024) | `10.1017/jfm.2024.333` |
+| Undermind | `DebienEtAl2016GeneticProgrammingRamp` | Debien et al. (2016) | `10.1007/s00348-016-2126-8` |
+| Undermind | `CastellanosEtAl2022FewSensorMLControl` | Castellanos et al. (2022) | `10.1063/5.0087208` |
+| Undermind | `RenEtAl2019MachineLearningVIV` | Ren, Wang & Tang (2019) | `10.1063/1.5115258` |
+| Undermind | `ZolmanEtAl2025SINDyRL` | Zolman et al. (2025) | `10.1038/s41467-025-65738-4` |
+| Undermind | `JordanEtAl2007ObservableJetModes` | Jordan et al. (2007) | `10.2514/6.2007-3602` |
+| Undermind | `TadmorEtAl2010CylinderMeanField` | Tadmor et al. (2010) | `10.1063/1.3298960` |
+| Undermind | `DiscettiEtAl2018CorrelatedFieldEstimation` | Discetti, Raiola & Ianiro (2018) | `10.1016/j.expthermflusci.2017.12.011` |
+| Undermind | `Mittal2001RotatingControlCylinders` | Mittal (2001) | `10.1006/JFLS.2000.0337` |
+| Undermind | `ChanEtAl2011CounterRotatingCylinders` | Chan et al. (2011) | `10.1017/jfm.2011.134` |
+| Undermind | `SierraEtAl2020RotatingCylinderBifurcations` | Sierra et al. (2020) | `10.1017/jfm.2020.692` |
+| Undermind | `Watson1996TwoRotatingCylinders` | Watson (1996) | `10.1093/QJMAM/49.2.195` |
+
+## Collision, duplicate and exclusion checks
+
+- **Keys:** 49 parsed keys, 49 unique; no key collision.
+- **DOIs:** 40 DOI-bearing records, 40 normalized unique DOI strings; normalization lowercases and strips `doi:`, `https://doi.org/`, surrounding braces and trailing punctuation. No DOI duplicate was found.
+- **Legacy prohibition:** key `Urzhumov2012` is absent. The preserved active-cloak identity remains `UrzhumovSmith2012ActiveCloaks`.
+- **Loiseau collision:** optional `LoiseauBrunton2018ConstrainedGalerkin` was not added; current `LoiseauEtAl2018SparseROM` remains the only admitted Loiseau 2018 identity.
+- **Li collision:** `LiEtAl2017GeneticProgrammingControl` was added. The selected 18-row Undermind wave does not contain `Li16`, so no duplicate identity was imported.
+- **Optional local exclusions:** Monticone & Alù (2013), Zou et al. (2019), Brunton et al. (2020), and Loiseau & Brunton (2018) remain absent.
+
+## Provisional metadata flags
+
+The user authorized syntactically valid provisional metadata for local-wave records pending manual checking. The following local records have no recorded DOI and retain local-source metadata without invented DOI, pages, issue or article number:
+
+- `Miller2006PerfectActiveCloaking`;
+- `LinEtAl2021ActiveAcousticIllusions`;
+- `ChenEtAl2024MultilayerHydrodynamicCloak`;
+- `BeemTriantafyllou2015WakeSensing`;
+- `Muller2015FundamentalsMusicProcessing`;
+- `FengWang2010SyntheticJetCylinder`;
+- `LiEtAl2017GeneticProgrammingControl`.
+
+`Cranmer2023PySR` is an arXiv record and intentionally has no DOI. Existing repository authorities supplied DOI/article metadata for Leonhardt, Vasquez, Ren (2021), Deng (2021) and Feng (2023). Unverified pages/volumes were not invented for Undermind rows; notably, `NairEtAl2021PhaseFlowControl` carries the verified venue/volume/DOI but no article number.
+
+## Static validation
+
+The repository has no BibTeX/Biber executable and no designated standalone bibliography checker. A read-only Python static scanner was therefore used without dependency changes. It checked:
+
+- balanced braces across the file and within every entry;
+- top-level entry segmentation and parseable `@type{key,...}` headers;
+- unique keys;
+- normalized DOI uniqueness;
+- required identity fields (`author`, `title`, `year`) and a venue field appropriate to each entry type;
+- absence of prohibited/optional keys.
+
+Result: **PASS** — 49 entries, 49 unique keys, 40 unique normalized DOIs, balanced braces, and zero malformed entry headers. This is static parseability, not a runtime BibTeX compilation claim.
+
+## Closeout
+
+Section `.tex` files, legacy bibliographies and final-audit disposition were not edited. `master/STATIC_PREFLIGHT.md` was updated only because its exact 16-entry/all-cited claim became misleading after bibliography expansion. Nowledge was queried three times for navigation and yielded no newer bibliography-specific authority; repository records and existing full-PDF audit outputs controlled the work. Undermind was not called and no literature research was performed. No CFD, training, artifact generation, dependency installation or new scientific analysis occurred.
+
+## Post-integration current-state addendum — 2026-08-11
+
+The bibliography itself is unchanged from the normalization audit. A fresh current-file census finds 49 entries, 49 unique keys, 40 DOI-bearing records with 40 unique normalized DOI strings, balanced braces, zero malformed headers, zero undefined active keys and zero uncited entries. Section TeX contains 44 citation commands, 64 key occurrences and 49 unique active keys. All 33 net-new keys are cited; no prune-versus-placement decision is required on uncited-status grounds.
+
+The normalization audit's 33-uncited statement is superseded by this addendum. Runtime BibTeX validation remains unavailable, and the seven provisional no-DOI local records plus the 15-record local full-text verification gate remain as recorded in `CITATION_EXPANSION_FINAL_AUDIT.md`.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/CITATION_EXPANSION_FINAL_AUDIT.md

@@ -0,0 +1,56 @@
+# Citation Expansion Final Audit
+
+**Project / unit:** DynamisLab JFM / post-R5 citation integration global closeout.  
+**Date:** 2026-08-11.  
+**Disposition:** **CONDITIONAL PASS — static citation/key/governance integration complete; source-state, metadata and runtime gates remain.**  
+**Protected record:** `../00_R5_FINAL_AUDIT.md` overall disposition was not changed.
+
+## Scope and non-mutation boundary
+
+Current S1–S8 and S10, S9 placeholder, `R5_REFERENCES.bib`, review master, section audits, citation controls, three expansion audits, live-length ledger, final audit and static preflight were read. This closeout changed control/audit statements only. It did **not** edit section prose, the bibliography, the review master or the existing final-audit disposition. No citation was added merely to eliminate an uncited entry.
+
+## Authoritative current census
+
+- Uniform substantive words: **7,434** across S1–S8 and S10; S9 remains **0 substantive / 272 control tokens**.
+- Section counts: S1 **1,123**; S2 **1,612**; S3 **578**; S4 **793**; S5 **958**; S6 **745**; S7 **499**; S8 **591**; S10 **535**.
+- Kármán analytical centre S4–S6: **2,496 / 7,434 = 33.58%**. Provisional 42%/44% diagnostics are 3,122/3,271, gaps 626/775; no padding is authorized.
+- Source burden: **12** displayed equation environments, **31** nonblank displayed rows, **143** inline-math spans, **44** citation commands and **64** cited-key occurrences.
+- Citation/bibliography: **49 unique active keys / 49 bibliography entries**; undefined keys **0**; uncited entries **0**; duplicate keys **0**; DOI-bearing entries **40**, duplicate normalized DOIs **0**.
+- All **33/33** post-R5 keys are cited. There is no uncited-key prune list. Retain current placements subject to the source-state gates below; citation status alone is not a reason to prune or add text.
+
+## Master, S9 and static TeX
+
+The master has exactly nine active inputs in order: S1, S2, S3, S4, S5, S6, S7, S8 and S10. Every input resolves. S9 exists only as an author-field placeholder and is explicitly omitted; S10 imports no S9 result. The active bibliography commands remain `\bibliographystyle{jfm}` and `\bibliography{../R5_REFERENCES}`.
+
+Comment-aware static checks pass for balanced braces, nested/paired environments, math delimiters, section/document ownership, bibliography path/style, prohibited legacy key absence, and active-key resolution. No TeX engine, BibTeX/Biber executable or resolver is installed, so no runtime compile was attempted.
+
+## Governance and quarantine
+
+Static prose/control searches preserve the exact configuration separation, Kármán spelling and `action-correlated correction-field modes` terminology. The integration does not promote positive Illusion SR, Kármán-law transfer, OID/PCD claims, CCD causality/validation/superiority/triangulation, old configuration chains, experimental validation, internal-result verification by literature, or priority/first-of-kind wording. S4 PPO, S5 SR and S6 four-role quantities remain distinct acquisitions and estimands. S9 remains quarantined.
+
+## Exact unresolved items
+
+1. **Local-source full-text gate:** all 15 local-wave identities remain `local-bibliography/context mapped, full text unverified`. Their present sentences are bounded to broad genealogy, method identity or context. Before submission, verify each exact proposition in full text or prune/rewrite that placement; literature cannot support internal project facts.
+2. **Provisional metadata:** seven local records have no recorded DOI and retain provisional local metadata: `Miller2006PerfectActiveCloaking`, `LinEtAl2021ActiveAcousticIllusions`, `ChenEtAl2024MultilayerHydrodynamicCloak`, `BeemTriantafyllou2015WakeSensing`, `Muller2015FundamentalsMusicProcessing`, `FengWang2010SyntheticJetCylinder`, and `LiEtAl2017GeneticProgrammingControl`. Confirm metadata manually; do not invent missing fields. `Cranmer2023PySR` is intentionally identified by arXiv `2305.01582`, not treated as a missing-DOI defect.
+3. **Runtime bibliography/TeX gate:** compile with the documented JFM class/style path when a TeX/BibTeX toolchain is available; inspect undefined-citation/reference warnings, bibliography rendering, class/package compatibility, fonts, line numbering, bad boxes and layout.
+4. **Existing scientific/production gates:** all source/provenance/caption/artwork/evaluator/S9 and historical-$U_0/U_{ref}$ conditions in `../00_R5_FINAL_AUDIT.md` remain unchanged. This citation closeout does not close them.
+
+## Files changed by this closeout
+
+- `../00_LIVE_LENGTH_SHARE_LEDGER.md`
+- `GLOBAL_PASS3_LENGTH_DISPLAY_AUDIT.md`
+- `../00_CITATION_CANDIDATE_LEDGER.md`
+- `CITATION_EXPANSION_BIBLIOGRAPHY_AUDIT.md`
+- `../master/STATIC_PREFLIGHT.md`
+- `SECTION_01_AUDIT.md`
+- `SECTION_02_AUDIT.md`
+- `SECTION_05_AUDIT.md`
+- `SECTION_06_AUDIT.md`
+- `SECTION_08_AUDIT.md`
+- `../README.md`
+- `../00_AGENT_REGISTRY.md`
+- this file, `CITATION_EXPANSION_FINAL_AUDIT.md`
+
+## Evidence and computation closeout
+
+Nowledge was searched three times narrowly and used for navigation only; current repository files controlled every count and status. Working Memory, thread recall and `mem_fs` were not used. Existing local and Undermind audit source states were read; no literature search or PDF read was performed in this closeout. Static Python counting/parsing and governance searches were used. No CFD, training, evaluator, rendering, artifact generation, dependency change or scientific re-analysis occurred.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/CITATION_EXPANSION_LOCAL_FULLTEXT_VERIFICATION.md

@@ -0,0 +1,46 @@
+# Citation Expansion Local Full-Text Verification
+
+**Unit:** all 15 frozen local-wave papers currently cited in S1/S2/S5.  
+**Owned outputs:** this audit, `../00_CITATION_CANDIDATE_LEDGER.md`, and source-state overrides in `CITATION_EXPANSION_LOCAL_SOURCE_AUDIT.md`.  
+**Excluded:** section TeX, `R5_REFERENCES.bib`, final disposition, broad literature search and internal verification.
+
+## Current-sentence basis
+
+Current exact citation-bearing sentences were read first in S1 paragraphs 1--4, S2 actuation/observation/DTW blocks, and S5 paragraphs 2--3. The verification question for each paper was limited to the proposition already written there.
+
+## Paper-by-paper result
+
+| 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. |
+
+## Required later prose action
+
+- **Remove `Miller2006PerfectActiveCloaking`** from the S1 active-construction cluster unless its exact PDF is supplied. The surviving Vasquez and Lin papers support source-based linear-PDE cloaking and non-zero acoustic illusion; do not retain a Miller-specific sensor/source claim from metadata alone.
+- **Remove both `BeemTriantafyllou2015WakeSensing` uses** and delete or soften the biological analogy. The accessible `Bee15b` arXiv precursor is not silently substituted for the exact JFM record.
+- **Remove `Muller2015FundamentalsMusicProcessing`** pending readable chapter pages. Recast the DTW definition around the verified Sakoe--Chiba monotonicity, continuity, boundary and slope constraints.
+- **Remove `FengWang2010SyntheticJetCylinder`** and its localized synthetic-jet clause pending exact full text. `Fen10` is a different paper and is prohibited as a surrogate identity.
+
+These are recommendations for a separately authorized TeX/bibliography correction pass; this unit makes no such edits.
+
+## Identity corrections recovered
+
+Previously provisional DOI fields are now resolved: Miller `10.1364/OE.14.012457`; Lin `10.1121/10.0003556`; Chen `10.1063/5.0230069`; Beem `10.1017/jfm.2015.513`; Müller `10.1007/978-3-319-21945-5`; Feng & Wang `10.1017/S0022112010003174`; Li et al. `10.1007/s00348-017-2382-2`; Cranmer `10.48550/arXiv.2305.01582`. These metadata corrections are recorded here only because bibliography editing is excluded.
+
+## Tool and evidence closeout
+
+Undermind `get_orientation` and `list_workspaces` reconfirmed the existing DynamisLab workspace. `get_paper_info(show_doi=true)` was called for all known workspace identities; exact metadata lookup was used only for Miller, Cranmer, Feng--Wang and the exact Beem JFM record, followed by mandatory `get_paper_info`. Bounded `read_pdfs` questions addressed only current propositions. Eleven papers yielded `full-PDF` passages. Miller, Feng--Wang and exact Beem were `mapped/inaccessible`; Müller was PDF-listed but two bounded reads timed out and remains `mapped/unread`. No broad or deep search ran. No literature evidence was used to verify project values, artifacts, configurations or performance, and no CFD/training/evaluator/rendering operation ran.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/CITATION_EXPANSION_LOCAL_SOURCE_AUDIT.md

@@ -0,0 +1,152 @@
+# Citation Expansion Local-Source Audit
+
+**Project / stage / unit:** DynamisLab JFM / post-R5 citation expansion / merged local-source control.  
+**Deliverable:** paragraph-level candidate and verification plan only.  
+**Owned outputs:** this audit and `../00_CITATION_CANDIDATE_LEDGER.md`.  
+**Excluded:** section TeX, `R5_REFERENCES.bib`, legacy bibliographies, final-audit disposition, literature promotion, and all internal scientific re-analysis.
+
+## 1. Source census and merge rule
+
+The local census is accepted as the starting control:
+
+- `docs/My_Confirmation`: 35 unique cited keys and 53 cited-key occurrences across Chapters 1–5. Two identities are exact duplicates of current R5 records: `pendry2006controlling` and `rabault2019artificial`.
+- `docs/JFM_WYQ/My_JFM/FLMguide.tex`: 9 active keys and 10 cited-key occurrences. Four identities are exact duplicates of current R5 records: `Gautier2015`, `Deng2020`, `CornejoMaceda2021` and `Boree2003`.
+- The current reader-facing set remains 16 unique references. Duplicate source keys are mapped to the descriptive R5 keys and never counted as net-new.
+- Identity is determined by authors/title/year/venue/DOI where recorded, not by source key spelling. The local files nominate propositions but are not full-text evidence.
+
+## 2. Paragraph-level expansion plan
+
+| Priority | Section / paragraph job | Planned uplift | Local-source function | Boundary |
+|---|---|---:|---|---|
+| 1 | **S1-P1**, cloak/illusion genealogy and finite-Re boundary | 150–200 | Add the independent 2006 transformation-optics branch; distinguish active source-based cloaking and non-zero illusion; add low-Re passive hydrodynamic and finite-Re active precedents | No universality, no project novelty, no claim that linear-wave control transfers to Navier–Stokes |
+| 1 | **S1-P2**, pinball/AFC lineage and force-versus-field distinction | 65–100 | Add pinball force-model and force-control context plus one conventional shedding-control example | External work motivates the distinction; it does not establish this plant, configuration or field result |
+| 1 | **S1-P3**, DRL, sensing and observer-space motivation | 45–70 | Add broad ML-fluid context and a biological wake-sensing precedent | No optimal-sensor, full-observability or natural-stealth claim |
+| 1 | **S1-P4**, interpretable feedback and sparse identification | 40–60 | Position genetic-programming control, SINDy and PySR as distinct interpretability precedents | Do not call SR governing-equation discovery or use literature to validate the executable law |
+| 2 | **S2 actuation/observation blocks**, pinball force control and wake sensing | 55–85 | Explain why integrated force and sparse downstream velocity are complementary measurement classes | No sensor optimality, no transfer from another pinball setup, no configuration-chain inference |
+| 2 | **S2 DTW evaluation block** | 75–115 | State DTW's dynamic-programming origin and standard path interpretation | External DTW sources define the method only; exact normalization, channels, windows and constraints remain repository-owned |
+| 3 | **S5 opening and execution-test blocks** | 55–85 | Contrast direct genetic-programming control, sparse dynamical-system identification and post-hoc policy-map regression | Closed-loop necessity remains argued from the present task; no external source validates the present law |
+| 3 | **S5 reflection/model-selection block** | 35–65 | Use constrained sparse regression as optional context for structural constraints and dynamical assessment | Distinct from the current imposed reflection map and from CFD execution |
+| 4 | **S6 CCD/POD distinction and closing caveat** | 15–35 | One compact literature-facing sentence locating correlation-oriented modes among data-driven flow-analysis tools | No OID import, no causal language, no superiority, no triangulation, no energy interpretation |
+
+The ranges sum to S1 +300–430, S2 +130–200, S5 +90–150 and S6 +15–35, or +535–815 overall. S3/S4/S7/S8/S10 are no-padding zones unless a later full-text read identifies a proposition with unique paragraph ownership.
+
+## 3. High-confidence local wave — 15 net-new identities
+
+All rows have source state **`local-bibliography/context mapped, full text unverified`**. `Candidate` means eligible for bounded full-text verification, not approved for import.
+
+| Provisional key | Recorded identity | Origin | Proposed R5 location / paragraph job | Exact external proposition to verify | Proposed sentence role | Status / source state | Duplicate mapping | Caveat |
+|---|---|---|---|---|---|---|---|---|
+| `Leonhardt2006OpticalMapping` | U. Leonhardt, “Optical conformal mapping,” *Science* 312, 1777–1780 (2006) | Confirmation `leonhardt2006optical`; Ch.1 passive-cloak context | S1-P1, transformation genealogy | Conformal-coordinate mapping furnished an independent 2006 route to optical cloaking | Pair with current Pendry record to avoid a single-paper origin story | Candidate; local-bibliography/context mapped, full text unverified | Net-new; related, not duplicate, to `PendryEtAl2006Controlling` | Linear optics only; do not transfer feasibility to viscous flow |
+| `Miller2006PerfectActiveCloaking` | D. A. B. Miller, “On perfect cloaking,” *Optics Express* 14, 12457–12466 (2006) | Confirmation `miller2006perfect`; Ch.1 active-cloak context | S1-P1, active construction branch | Active cloaking can be formulated with sensing and distributed sources rather than a passive transformation shell | Introduce the sensor–controller–source architecture | Candidate; local-bibliography/context mapped, full text unverified | Net-new | “Perfect” is the paper title/theoretical objective, not an attained project property |
+| `VasquezEtAl2009ActiveExterior` | F. Guevara Vasquez, G. W. Milton & D. Onofrei, “Active exterior cloaking for the 2D Laplace and Helmholtz equations,” *Physical Review Letters* 103, 073901 (2009) | Confirmation `vasquez2009active`; Ch.1 | S1-P1, active exterior-cloak example | Exterior active devices can create a protected region and control scattering for specified linear PDEs | Concrete active-field precedent after Miller | Candidate; local-bibliography/context mapped, full text unverified | Net-new | Restricted 2-D linear equations; no finite-Re inference |
+| `LinEtAl2021ActiveAcousticIllusions` | C. Lin, D. Liu, D. Eggler & N. Kessissoglou, “Active acoustic cloaking and illusions of sound-hard bodies using the boundary element method,” *JASA* 149, 1803–1812 (2021) | Confirmation `lin2021active`; Ch.1 | S1-P1, non-zero illusion definition | An active acoustic architecture can retarget the exterior field toward a different non-zero scattering signature | Sharpen cloak-versus-illusion distinction | Candidate; local-bibliography/context mapped, full text unverified | Net-new | Acoustic/BEM setting; no hydrodynamic Illusion efficacy or SR implication |
+| `ChenEtAl2024MultilayerHydrodynamicCloak` | M. Chen, X. Shen & L. Xu, “A multilayered homogeneous hydrodynamic cloak realized in a free fluid flow,” *Applied Physics Letters* 125(17) (2024) | Confirmation `chen2024multilayered`; Ch.1 | S1-P1, passive-hydrodynamic boundary | Multilayer passive hydrodynamic cloaking has been demonstrated in a restricted low-inertia/free-flow regime | Establish a fluid precedent before the separated finite-Re gap | Candidate; local-bibliography/context mapped, full text unverified | Net-new | Verify exact regime and wording; do not claim absence of separation from bibliography alone |
+| `RenEtAl2021HydrodynamicStealthDRL` | F. Ren, C. Wang & H. Tang, “Bluff body uses deep-reinforcement-learning trained active flow control to achieve hydrodynamic stealth,” *Physics of Fluids* 33(9) (2021) | Confirmation `ren2021bluff`; Ch.1 | S1-P1/P3, finite-Re active precedent | DRL-controlled active flow manipulation has previously been framed as hydrodynamic stealth for a bluff body | Narrow the claimed research gap and motivate model-free control | Candidate; local-bibliography/context mapped, full text unverified | Net-new | No priority wording; do not import thesis-era percentages or equate its metric with the present target-field measure |
+| `DengEtAl2021PinballForceModel` | N. Deng, B. R. Noack, M. Morzyński & L. R. Pastur, “Galerkin force model for transient and post-transient dynamics of the fluidic pinball,” *JFM* 918, A4 (2021) | Confirmation `deng2021galerkin`; Ch.2 | S1-P2 and S2 actuation/force context | Fluidic-pinball force dynamics admit a reduced modelling treatment across transient and post-transient behaviour | Support force as a meaningful intermediary, distinct from full-field matching | Candidate; local-bibliography/context mapped, full text unverified | Net-new; distinct from `DengEtAl2020PinballBifurcations` | No transfer of model coefficients, bifurcation thresholds or configuration labels |
+| `FengEtAl2023PinballForceControl` | H. Feng, Y. Wang, H. Xiang, Z. Jin & D. Fan, “How to control hydrodynamic force on fluidic pinball via deep reinforcement learning,” *Physics of Fluids* 35(4) (2023) | Confirmation `feng2023control`; Ch.2 | S1-P2 and S2 target/force role block | DRL has been applied to fluidic-pinball hydrodynamic-force control and tracking objectives | Contrast integrated-force objectives with the present independent field test | Candidate; local-bibliography/context mapped, full text unverified | Net-new | Do not infer identical geometry, action decoding, observations or field performance |
+| `BeemTriantafyllou2015WakeSensing` | H. R. Beem & M. S. Triantafyllou, “Wake-induced ‘slaloming’ response explains exquisite sensitivity of seal whisker-like sensors,” *JFM* 783, 306–322 (2015) | Confirmation `beem2015wake`; Ch.2 | S1-P3 and S2 observation block | Wake-carried velocity/vortical information can be detected by sparse biological-inspired appendage sensing | Motivate downstream wake information as an observer quantity | Candidate; local-bibliography/context mapped, full text unverified | Net-new | Analogy only; no claim that three probes reproduce seal sensing or are optimal |
+| `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 | Confirmation `sakoe1978dynamic`; Ch.3 | S2 DTW block | DTW computes a monotone dynamic-programming alignment under specified path constraints | Supply method genealogy for the recurrence/path description | Candidate; local-bibliography/context mapped, full text unverified | Net-new | Speech origin only; does not authorize current normalization or physical interpretation |
+| `Muller2015FundamentalsMusicProcessing` | M. Müller, *Fundamentals of Music Processing: Audio, Analysis, Algorithms, Applications*, vol. 5, Springer (2015) | Confirmation `muller2015fundamentals`; Ch.3 | S2 DTW block | Standard DTW exposition defines local costs, accumulated cost and admissible warping paths | Reader-facing algorithm reference paired with original source | Candidate; local-bibliography/context mapped, full text unverified | Net-new | Book metadata need verification; no claim that unconstrained DTW is appropriate here |
+| `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) | Confirmation `feng2010circular`; Ch.1 AFC context | S1-P2, conventional AFC comparator | Conventional AFC can use localized actuation to alter vortex shedding/synchronization around a bluff body | One concrete comparator before distinguishing observer-centred wake matching | Candidate; local-bibliography/context mapped, full text unverified | Net-new | Single example, not an AFC review; do not claim conventional AFC never evaluates fields |
+| `LiEtAl2017GeneticProgrammingControl` | R. Li, B. R. Noack, L. Cordier, J. Borée & F. Harambat, “Drag reduction of a car model by linear genetic programming control,” *Experiments in Fluids* 58, 103 (2017) | Legacy `Li2017`; active FLMguide S1 interpretability paragraph | S1-P4 and S5 opening | Genetic-programming control searches directly over explicit feedback expressions and evaluates them in a flow-control setting | Precedent for readable feedback maps, contrasted with post-hoc PPO distillation | Candidate; local-bibliography/context mapped, full text unverified | Net-new | Different plant/objective/search loop; no validation of the present SR map |
+| `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 | Legacy `Brunton2016`; active FLMguide S1 | S1-P4 and S5 opening | SINDy identifies parsimonious dynamical equations from a prescribed function library | Distinguish dynamical-system identification from policy-map regression | Candidate; local-bibliography/context mapped, full text unverified | Net-new | Do not describe the present symbolic controller as a discovered governing equation |
+| `Cranmer2023PySR` | M. Cranmer, “Interpretable machine learning for science with PySR and SymbolicRegression.jl,” arXiv:2305.01582 (2023) | Legacy `Cranmer2023`; active FLMguide methods | S1-P4 and S5 model-selection block | PySR performs symbolic-regression search over expression structure and coefficient optimization, yielding loss–complexity candidates | Software/method identity for the post-hoc regression stage | Candidate; local-bibliography/context mapped, full text unverified | Net-new | Verify version and preferred archival citation; software identity does not establish scientific validity |
+
+## 4. Optional local wave — 4 net-new identities
+
+These records require a sharper paragraph need or may be superseded by an Undermind full-text source. Their source state is also **`local-bibliography/context mapped, full text unverified`**.
+
+| Provisional key | Recorded identity | Origin | Proposed R5 location / paragraph job | Exact external proposition to verify | Proposed sentence role | Status / source state | Duplicate mapping | Caveat |
+|---|---|---|---|---|---|---|---|---|
+| `MonticoneAlu2013CloakScatteringBounds` | F. Monticone & A. Alù, “Do cloaked objects really scatter less?” *Physical Review X* 3, 041005 (2013) | Confirmation `monticone2013cloaked`; Ch.1 | Optional S1-P1 | Passive linear cloaks face broadband/integrated-scattering trade-offs | Explain why active architectures are attractive without overloading the genealogy | Optional candidate; local-bibliography/context mapped, full text unverified | Net-new | Exact theorem assumptions must be read; no direct Navier–Stokes implication |
+| `ZouEtAl2019WaterWaveCloak` | S. Zou et al., “Broadband waveguide cloak for water waves,” *Physical Review Letters* 123, 074501 (2019) | Confirmation `zou2019broadband`; Ch.1 | Optional S1-P1 | Water-wave cloaking can be realized through a waveguide/bathymetric construction in a linear-wave regime | Add a fluid-wave bridge if needed | Optional candidate; local-bibliography/context mapped, full text unverified | Net-new | Surface-wave manipulation is not bulk separated-flow cloaking |
+| `BruntonEtAl2020MachineLearningFluids` | S. L. Brunton, B. R. Noack & P. Koumoutsakos, “Machine learning for fluid mechanics,” *Annual Review of Fluid Mechanics* 52, 477–508 (2020) | Confirmation `brunton2020machine`; Ch.3 | Optional S1-P3 or S6 caveat | ML in fluid mechanics spans control, modelling and reduced representations, with method choice tied to the physical task | Broad review anchor, used only if a synthesis sentence needs one | Optional candidate; local-bibliography/context mapped, full text unverified | Net-new | Broad review cannot support a detailed PPO, CCD or mechanism claim |
+| `LoiseauBrunton2018ConstrainedGalerkin` | J.-C. Loiseau & S. L. Brunton, “Constrained sparse Galerkin regression,” *JFM* 838, 42–67 (2018), DOI 10.1017/jfm.2017.823 | Legacy `Loiseau2018Galerkin`; active FLMguide S1 | Optional S5 reflection/model-selection block | Structural constraints can be imposed in sparse Galerkin regression to improve physically admissible reduced dynamics | Context for why structural constraints are scientifically meaningful | Optional candidate; local-bibliography/context mapped, full text unverified | Net-new; **distinct** from `LoiseauEtAl2018SparseROM` (JFM 844, DOI 10.1017/jfm.2018.147) | Different task and object; it does not justify the present reflection map or closed-loop result |
+
+## 5. Exact duplicate map and identity collisions
+
+| Local key | Identity disposition | Current R5 mapping / action |
+|---|---|---|
+| `pendry2006controlling` | Exact duplicate | Map to `PendryEtAl2006Controlling`; no new count |
+| `rabault2019artificial` | Exact duplicate | Map to `RabaultEtAl2019DRLFlowControl`; no new count |
+| `Gautier2015` | Exact duplicate | Map to `GautierEtAl2015ClosedLoop`; no new count |
+| `Deng2020` | Exact duplicate | Map to `DengEtAl2020PinballBifurcations`; no new count |
+| `CornejoMaceda2021` | Exact duplicate | Map to `CornejoMacedaEtAl2021PinballControl`; no new count |
+| `Boree2003` | Exact duplicate | Map to `Boree2003ExtendedPOD`; no new count |
+| `Urzhumov2012` | **Identity collision, not duplicate**: legacy key stores Urzhumov & Smith (2011), “Fluid flow control with transformation media,” PRL 107, 074501 | Do not map to `UrzhumovSmith2012ActiveCloaks`; prohibit old key; 2011 identity remains outside this wave |
+| `Loiseau2018Galerkin` | **Distinct identity**: Loiseau & Brunton (2018), JFM 838, DOI 10.1017/jfm.2017.823 | Optional net-new candidate; do not merge with `LoiseauEtAl2018SparseROM`, JFM 844, DOI 10.1017/jfm.2018.147 |
+
+## 6. Union arithmetic and projected article effect
+
+- Current R5 unique references: 16.
+- High-confidence Confirmation net-new: 12 (`Leonhardt`, `Miller`, `Vasquez`, `Lin`, `Chen`, `Ren`, `Deng 2021`, `Feng 2023`, `Beem`, `Sakoe`, `Müller`, and `Feng & Wang 2010`).
+- High-confidence legacy net-new: 3 (`Li 2017`, `Brunton 2016`, `Cranmer 2023`).
+- High-confidence union: **15 net-new**, projected **31 total unique references**.
+- Optional Confirmation net-new: 3 (`Monticone`, `Zou`, `Brunton 2020`).
+- Optional legacy net-new: 1 (`Loiseau & Brunton 2018`).
+- Full local ceiling: **19 net-new**, projected **35 total unique references**.
+- Duplicate identities add zero. The 2011 Urzhumov collision is neither merged nor counted as a candidate.
+- Projected word uplift: **+535–815**, taking the current uniform body count from **6,303 to 6,838–7,118**. This projection excludes citation commands and assumes the current counting method; it must be recomputed after any authorized prose insertion.
+
+## 7. Undermind gap-search topics — separate from local admission
+
+No Undermind search was run in this local-source audit. The next literature wave should search the existing DynamisLab workspace/library first, then use bounded full-PDF reads for these gaps:
+
+1. A modern, authoritative AFC review that distinguishes integrated-force, separation/shedding and distributed wake/field objectives without claiming that all conventional AFC is local or scalar.
+2. Finite-Re active hydrodynamic cloaking/stealth studies after Urzhumov and Ren, with exact observer metric, configuration and claim ceiling.
+3. Active illusion/target-signature control in nonlinear fluid wakes, specifically to determine whether any prior work supports only conceptual framing; priority language remains prohibited.
+4. Closed-loop symbolic policy distillation for delayed PDE/flow systems, focusing on state-distribution shift and deployment-based validation rather than offline action fit.
+5. Reflection/equivariance constraints in learned or symbolic flow controllers, clearly separated from plant symmetry and from the present imposed surrogate map.
+6. Action-conditioned/correlation-oriented flow decompositions and their noncausal interpretation, without importing OID claims or treating CCD as validation.
+7. DTW use in fluid/wake comparison with explicit multichannel normalization and path constraints; the search cannot retroactively verify this project's metric implementation.
+8. Sparse wake sensing and observer placement precedents that avoid an optimality claim for the present three-probe layout.
+
+Search results remain external-proposition candidates only. They cannot verify internal values, configuration bindings, target roles, artifact provenance or claim readiness.
+
+## 8. Hard quarantine
+
+The expansion pass must not introduce or imply:
+
+- positive Illusion symbolic regression, a reusable Illusion law, or revival of historical negative Illusion SR;
+- transfer of the Kármán SR law to Vortex, Erase, Illusion, other Reynolds numbers or other configurations;
+- any OID result or OID-based claim;
+- CCD causality, mechanistic proof, superiority to POD, independent confirmation, validation or triangulation;
+- inherited old Reynolds-number conversions, legacy/current plant continuity or pooled configuration chains;
+- old thesis/report numbers, percentages, thresholds, sensor coordinates, training settings or experiment facts as current R5 evidence;
+- experimental validation, CFD/experiment agreement, PIV, hardware readiness or any S9 scientific content;
+- first, novel, unprecedented, robust, optimal, universal or other priority wording.
+
+Literature never verifies DynamisLab results, numbers, acquisitions, artifacts, mechanisms, readiness or priority.
+
+## 9. Acceptance checks and closeout
+
+- **Files read:** both local bibliographies; Confirmation Chapters 1–5; legacy `FLMguide.tex`; current `R5_REFERENCES.bib`; current S1–S8 and S10; citation ledger; live length/share ledger; final audit.
+- **Files changed:** only `../00_CITATION_CANDIDATE_LEDGER.md` and this audit. No section TeX or bibliography was edited.
+- **State preservation:** current 16/16 citation integration and final-audit `CONDITIONAL PASS` remain the active disposition. Expansion records are provisional.
+- **Tools/evidence:** Nowledge was searched with three narrow R5/citation/collision queries and used for navigation only; repository files controlled the audit. Working Memory, thread recall and `mem_fs` were not used. Undermind and SciVerse were not used because this unit maps local sources and explicitly separates later gap searches. No PDF was read, so every new identity remains `local-bibliography/context mapped, full text unverified`.
+- **Computation:** only set/count/range arithmetic and file consistency checks are authorized here; no CFD, training, evaluator, rendering, artifact generation or new scientific analysis was performed.
+
+
+## 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: `CITATION_EXPANSION_LOCAL_FULLTEXT_VERIFICATION.md`.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/CITATION_EXPANSION_UNDERMIND_GAP_AUDIT.md

@@ -0,0 +1,33 @@
+# Citation Expansion Undermind Gap Audit
+
+**Project / unit:** DynamisLab JFM post-R5 / existing-workspace gap curation.  
+**Owned outputs:** this audit and `../00_CITATION_CANDIDATE_LEDGER.md`.  
+**Excluded:** section TeX, `R5_REFERENCES.bib`, final disposition, new search/deep search, and internal verification.
+
+## Routing and evidence
+
+Completed mandatory routing: `get_orientation`; `list_workspaces` (confirmed DynamisLab id `4383d145-2c80-4991-8a9c-7ac2dd683438`); root and targeted `inspect_folders`, `inspect_files`, `inspect_deep_searches`; folders `/JFM References/01`--`/07`; note `/Wake-field L2 evaluation and downstream sensor rationale`; and six named completed searches. No `search_papers` or `launch_deep_search` was needed. Local citation controls, current S1--S8/S10, bibliography and final audit were read.
+
+Every selected paper was checked with `get_paper_info(show_doi=true)` and had a PDF. One bounded `read_pdfs` batch asked a separate self-contained question per paper for only its proposed proposition and page/section. The ledger table is the controlling row-level identity, DOI, proposition, passage, topic and state record. All 18 selected rows are **`full-PDF`**. Existing search reports and the sensor note were used only as **`full-report/mapped`** routing evidence. `Mau01` and `Bon94` remained **`mapped/inaccessible`** and support no wording; no snippet evidence was used.
+
+## Section-role determination
+
+- **S1 (2):** `Fan20b` and `Li21` add bounded experimental/numerical and physics-informed sparse-sensing RL lineage without efficacy, mechanism or priority claims.
+- **S2 (5):** `Mar24b`, `Jin21`, `Gon19`, `Nai20b`, `Wan23b` cover complementary actuation/wake coordinates, estimation-vs-feedback placement, periodic sparse phase information, force observables and temporal feature lifting. None proves present sensor optimality or observability.
+- **S5 (4):** `Deb15`, `Cas22`, `Ren19`, `Zol25` distinguish direct closed-loop symbolic-law search and jointly learned model-based sparse RL from post-hoc regression of a fixed PPO policy. None validates the present law or makes it a governing equation.
+- **S6 (3):** `Tad10`, `Jor07`, `Dis18` supply mean-field and correlation/observable-ranked ancestry only. OID is not imported; wording remains descriptive/noncausal.
+- **S8 (4):** `Mit01`, `Cha11b`, `Sie20`, `Wat96` sharpen Re/gap/rotation/dimensionality/branch and viscous-matching boundaries. They do not explain the project endpoint or repair the rejected circulation closure.
+
+## Duplicates, rejections and arithmetic
+
+Exact current-R5 duplicates are `Rai20`, `Mac20`, `Loi17`, `Bor03`, `Sch12`, `Par20`, `Rab18b`, `Ued03`; local-wave identity collisions are `Li16` and `Loi16`. `Bru16` was redundant with local SINDy plus sharper `Zol25`; `Ish17` lacked a retained sentence job; `Mau01`/`Bon94` were inaccessible and superseded by full-PDF lineage; extra inviscid multiple-cylinder papers were redundant with current Crowdy.
+
+The evidence-backed union is 16 current + 15 frozen local + 18 verified Undermind = **49**. Recommend **46--49** after sentence-level pruning. Estimated Undermind uplift is **+435--655 words**; combined post-R5 uplift **+970--1,470**, yielding approximately **7,273--7,773** words before uniform recount. These are planning ranges, not quotas.
+
+## Quarantine and disposition
+
+No OID, PCD causality, CCD causality/validation/superiority/triangulation, positive Illusion SR, cross-case SR transfer, old configuration chain, project-number/artifact verification, experiment validation or priority wording is permitted. Literature supports external propositions only. No TeX, bibliography or final audit was edited; current 16/16 integration and `CONDITIONAL PASS` remain active.
+
+## Tool closeout
+
+Nowledge: three narrow `memory_search` calls, navigation only; no Working Memory, thread recall or `mem_fs`. Undermind: `get_orientation`, `list_workspaces`, `inspect_folders`, `inspect_files`, `inspect_deep_searches`, `get_paper_info`, `read_pdfs`; no workspace mutation. No CFD, training, evaluator, rendering, artifact generation or dependency change.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/GLOBAL_PASS3_LENGTH_DISPLAY_AUDIT.md

@@ -0,0 +1,39 @@
+# Global Pass 3 — Post-citation-integration uniform length and display audit
+
+**Project / round / unit:** DynamisLab JFM / post-R5 citation integration / decisive global recount.  
+**Date:** 2026-08-11.  
+**Disposition:** **PASS as a static length/display diagnostic.** This pass grants no science, source-state, artifact or runtime-compile approval.
+
+## Uniform method and supersession
+
+The audit reran exactly the established Python 3 source-text method over all ten `sections/*.tex` files: remove unescaped comments and headings; remove complete display-math bodies, inline-math spans and citation commands; remove remaining TeX commands/braces; then count lexical alphanumeric tokens with internal hyphens or apostrophes. S9 `AUTHOR FIELD` instructions are classified separately as control tokens. This is a reproducible source count, not a typeset word-equivalent.
+
+This refresh supersedes the former 6,303-word pre-expansion snapshot. Current arithmetic is mirrored authoritatively in `../00_LIVE_LENGTH_SHARE_LEDGER.md`.
+
+## Final current recount
+
+| Section | Uniform substantive words | Share |
+|---|---:|---:|
+| S1 | 1,123 | 15.11% |
+| S2 | 1,612 | 21.68% |
+| S3 | 578 | 7.78% |
+| S4 | 793 | 10.67% |
+| S5 | 958 | 12.89% |
+| S6 | 745 | 10.02% |
+| S7 | 499 | 6.71% |
+| S8 | 591 | 7.95% |
+| S9 | 0 substantive; 272 control | excluded |
+| S10 | 535 | 7.20% |
+| **Total** | **7,434** | **100%** |
+
+S4–S6 contain **2,496 words / 33.58%**. The provisional 42%/44% diagnostics correspond to 3,122/3,271 words, gaps of 626/775. This does not authorize padding. Their five displayed equations, fourteen nonblank displayed rows and 82 inline-math spans make prose share an incomplete proxy for typeset emphasis.
+
+## Current burden and branch
+
+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)**. There are no figure or table environments. Citation mapping is 49/49 against the 49-entry bibliography, with zero undefined keys, uncited entries, duplicate keys or duplicate normalized DOIs.
+
+The active omission branch has 7,434 substantive words and excludes 272 S9 control tokens. Author admission of 150–300 substantive S9 words would yield 7,584–7,734 words and an unchanged-centre share of 32.91–32.27%.
+
+## Closeout
+
+All ten current section paths were counted; totals reconcile to 7,434 and S4–S6 to 2,496. All 33 post-R5 keys are cited; none is a prune candidate merely for being uncited. No section prose, bibliography, literature research, scientific computation, evaluator, CFD, training, rendering or artifact generation was used. Repository source state is `read/static-counted`, not independently artifact-verified.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/GLOBAL_STRUCTURE_LOGIC_PASS_01.md

@@ -0,0 +1,127 @@
+# Round-5 global optimization audit — Pass 1: structure and logic
+
+**Date:** 2026-08-11  
+**Scope:** current S1–S10 TeX, all ten section audits, both existing wave audits, R5 registry/ledgers/packets/rules, and current R4 section/global controls.  
+**Deliverable:** findings and coordinator-approved deltas for the original persistent writers.  
+**Exclusions:** no section-TeX edit; no literature search/read; no new scientific analysis, evaluator, CFD, training, rendering, artifact generation, bibliography construction or final-master assembly.
+
+## Determination
+
+**Disposition: CONDITIONAL PASS — structure/logic.** The frozen order, 65 paragraph jobs, three-level claim ladder, section ownership, preview/detail splits and scientific bridges are present. The article has no structural-science blocker outside the already declared Section-9 branch. Pass 1 does **not** close evidence, citation, title, equation-source, display, provenance, author-field or production gates, and it does not grant any section a final pass.
+
+The manuscript-wide logic is intact:
+
+`declared-reference gap → configurations/controller/evaluation → asymmetric capability progression → tested Kármán performance → executable action reduction → distinct mean/residual field organization → changed-contract non-transfer → separate steady context and failed closure → author-owned experiment branch → bounded synthesis`.
+
+## R4 jobs, ladder and ownership
+
+- **Jobs:** all **65/65** inventory jobs have either reader-facing prose or the authorized S9 placeholder. S1 5; S2 5 frozen jobs represented by 11 prose paragraphs; S3 6; S4 7; S5 5; S6 6; S7 5; S8 5; S9 3 author fields; S10 5. No missing or extra scientific job was found.
+- **Ladder:** S1 defines the three levels; S3 previews breadth and selects Kármán for depth; S4 establishes the sole independent field-level result; S5–S6 support only the bounded Kármán/cloak interpretation; S7–S8 deny transfer; S10 closes in the same order. No section raises the ladder.
+- **Ownership:** S2 owns reusable definitions; S3 owns previews; S4 owns exact PPO performance; S5 owns SR law/deletion; S6 owns corrected four-role means/residual/CCD; S7 owns changed-task detail; S8 owns steady/theory; S9 owns author experiment supply; S10 synthesizes without new evidence. Equation ownership also passes: S2 reusable definitions, S5 reflection/law, S6 scalar/residual, S8 outer-reference/incompatibility, none in S1/S3/S4/S7/S9/S10.
+- **Near-collisions:** S4 PPO, S5 standardized SR and S6 corrected four-role errors remain explicitly different acquisitions/estimands. SR terms are not identified with CCD coordinates. Steady rotation is contextual, not amplitude agreement. No pooling, replication, agreement, validation or triangulation appears.
+
+## Bridges and deduplication
+
+| Interface | Finding | Status / owner delta |
+|---|---|---|
+| S1→S2 | Contribution ladder hands directly to configuration, role, Reynolds and evaluation definitions. | PASS. |
+| S2→S3 | Existing wave record passes definitions-to-viewing handoff and preview/detail ownership. | PASS; no delta. |
+| S3→S4 | S3 retains rounded `approximately 82%`; S4 owns exact values, region, phase construction, DTW and actions. | PASS. |
+| S4→S5 | Independent field result becomes the executable-policy-compression question without importing SR. | PASS. |
+| S5→S6 | Tested persistent/smaller action organization motivates, but does not predict, the distinct field estimands. | PASS; existing conditional S4–S6 record controls open source/artifact gates. |
+| S6→S7 | Exact Kármán interpretation closes with a case-local non-transfer boundary before changed contracts. | PASS. |
+| S7→S8 | Separate steady endpoint is posed as context only, with no continuation, validation or common arithmetic. | PASS. |
+| S8→S9 | Provenance-conditional, evidence-neutral morphology question; experiment cannot repair quantitative/mechanistic gates. | PASS. |
+| S9→S10 | No experimental result is exported. S10 currently proceeds from retained S1–S8 only. | PASS under placeholder/omit branch; author decision required below. |
+| S10 close | Restates the ladder without values, equations, citations, display calls, Erase result or unfinished-route inventory. | PASS after the narrow S1 terminology correction below. |
+
+Deduplication is controlled. Repeated configuration and role names are local decoding, not repeated methods. The Kármán `approximately 82%` appears once as S3 preview and exact values once in S4. The persistent-plus-smaller-correction organization appears in S1 as the ladder, in S5/S6 at owner depth, in S8 as explicitly non-equivalent context and in S10 as the one required synthesis; this is functional recurrence rather than duplicate evidence. No deletion is approved in Pass 1.
+
+## Exact S1 named-task correction
+
+**Finding:** S1-P05 currently says `Vortex, Illusion, Erase and steady material do not inherit that interpretation`, while S1-P05's declared level 1 is `learned wake-signature capability across named tasks`. Current retained S7/S10 status is narrower: Vortex and Erase role fields are morphology/execution material without a common efficacy result; the positive named non-zero-target capability is Illusion PPO. The sentence can invite the unsupported reading that all named items belong equally to level 1.
+
+**Approved Unit-A delta (exact replacement; no other S1 edit):**
+
+Replace:
+
+> `Kármán cloaking receives depth because it alone retains the aligned policy, target, controlled, physical-zero, action and field evidence needed for all three levels; Vortex, Illusion, Erase and steady material do not inherit that interpretation.`
+
+with:
+
+> `Kármán cloaking receives depth because it alone retains the aligned policy, target, controlled, physical-zero, action and field evidence needed for all three levels; the event-relative Vortex acquisition and Erase role fields remain morphology evidence, Illusion retains only its named non-zero-target PPO capability, and none of these materials or the separate steady chain inherits the Kármán interpretation.`
+
+This correction is terminology/status normalization only. It adds no result, value or evidence and leaves the five-paragraph architecture unchanged. Unit A writer `5ddac128-c9c7-4b83-8fb9-aed05023a9b0` owns integration after coordinator approval.
+
+## S7–S9 controlling record and Section-9 branch
+
+This sidecar designates the existing records as the controlling S7–S9 interface chain; no competing wave file is required:
+
+1. `00_AGENT_REGISTRY.md`, Unit-D row: stage `S7–S9-interface-pass` and retained S9 boundary.
+2. `audits/SECTION_07_AUDIT.md`, especially the S7→S8 non-transfer finding.
+3. `audits/SECTION_08_AUDIT.md`, especially the provenance-conditional S8→S9 bridge.
+4. `audits/SECTION_09_AUDIT.md`: author-interface, zero substantive manuscript words, no result export.
+
+**Required author/coordinator branch decision before evidence/data and length/display global passes can close:**
+
+- **Branch A — omit S9 from the assembled manuscript (current safe fallback).** Remove the placeholder from manuscript assembly later; retain its audit as provenance/control. S8 should bridge directly to S10 during final assembly only if a transition is needed. No S9 allowance is redistributed automatically.
+- **Branch B — retain an author-written qualitative S9.** Frank supplies apparatus/execution facts, trial/media provenance, permissions, comparator-led morphology and confounds; then a targeted provenance audit and explicit reopening must pass before prose/media admission. Only after that decision may a bounded S9 observation be considered for S10.
+- **Branch C — retain author fields temporarily in the R5 working bundle.** This is acceptable as workflow state but is **not** a manuscript-content decision and cannot be called a section prose pass.
+
+Until one branch is selected, normalize S9 everywhere to `author-interface / placeholder-control complete / substantive prose and result import blocked`; do not use `section-pass`, `experiment complete`, or count its 275 tokens as body prose.
+
+## Section and interface status normalization
+
+Use the following controlling R5 statuses after Pass 1. These do not alter evidence/artifact gates.
+
+| Unit / section | Normalized status | Required record delta |
+|---|---|---|
+| S1 | `section-audit; Pass-1 correction approved, not integrated` | Unit A applies exact sentence replacement; then mark `global-structure-logic-conditional`. |
+| S2 | `S2–S3-interface-pass; local source/display/bibliography gates open` | No prose delta. Do not call section-pass. |
+| S3–S4 | `section-audit; production gates open` | Registry already matches. S3/S4 audits should not be promoted by S2 interface pass. |
+| S5–S6 | `S4–S6-interface-conditional` | Registry controls. Section audits' `integrated rewrite / conditional wave audit` is consistent. |
+| S7–S8 | `S7–S9-interface-pass; section prose audited` | Their local audit label `section-pass` is narrower than final R5 pass; interpret it as section-prose pass only. Registry remains controlling. |
+| S9 | `author-interface; placeholder-control complete; substantive prose/result blocked` | Author branch required; never manuscript-prose pass. |
+| S10 | `section-audit; global S1/S10 interface conditional` | No prose delta after S1 correction; imported results otherwise pass. |
+| Global | `structure/logic CONDITIONAL PASS` | Evidence/data, length/display, terminology/style/TeX and final audit remain unrun. |
+
+Ledger inconsistencies are bookkeeping defects, not prose defects: S3 is variously 627/637 words; S5–S6 appear as 614/608, 621/625 and a 2,070 S4–S6 total that does not arithmetically match every row. Normalize these only in the later length/display pass using one declared counting script/method; do not use Pass 1 to change prose or infer shares.
+
+## Title and bibliography gates
+
+- **Article title:** the R4 provisional title `Interpreting learned hydrodynamic cloaking and illusion in a fluidic pinball` remains provisional (`discuss`). Pass 1 finds it consistent with the declared-reference ladder and positive PPO Illusion scope, but it is not final. Final title selection belongs to the author and triggers the recorded terminology/claim audit. Do not place it in a submission master yet.
+- **Section titles:** current S1–S10 titles are structurally coherent. S9's `Qualitative open-loop experiment` title is conditional on selecting the retain-S9 branch; if S9 is omitted, no orphan subsection title enters assembly. No title change is approved here.
+- **Bibliography:** current TeX contains no fabricated `\cite` keys, which is correct for the present gate. The citation candidate ledger records identities/DOIs/read states but is not a bibliography. Before final TeX preflight, verify each retained external proposition, exact source state and repository-approved key; then insert citations only through the owning original writer or coordinator-approved bibliography pass. S10 retains no reader-facing citation. Bibliography remains **OPEN manuscript-wide**, especially S1 framing and S2 method context.
+
+## Approved deltas for original writers
+
+1. **Unit A / S1 — approved and required:** apply only the exact named-task correction above; update S1 audit and registry status to show integration. Do not add citations, values, examples or length padding.
+2. **Unit E / S9–S10 — decision-dependent:** after Frank/coordinator selects S9 Branch A or B, resume writer `b238cc11-c57b-481f-922d-ae7c410171c0`. Branch A updates controls/assembly handoff only and leaves S10 prose unchanged unless the coordinator specifically approves a neutral transition. Branch B requires author supply plus targeted provenance audit before any TeX result or S10 import.
+3. **Unit D / S7–S8 — no prose delta:** retain current text. Registry plus S7/S8/S9 audits are the designated S7–S9 control. Update only status wording if the coordinator wants local audits to say `section-prose pass` rather than final `section-pass`.
+4. **Units B/C — no prose delta:** retain current S3–S6 text and all ownership/estimand boundaries. Defer word-count reconciliation to global length/display and source/artifact gates to evidence/data/production owners.
+5. **Global controls — approved bookkeeping delta:** after original-writer integration/branch decision, update the registry, claim ledger, paragraph inventory and live length ledger from this sidecar. Do not mark `00_R5_FINAL_AUDIT.md` as run and do not check the structure/logic box until the S1 correction is integrated and the S9 branch is recorded.
+
+## Gate checklist after Pass 1
+
+- [x] Frozen assembled order and all 65 jobs accounted for.
+- [x] Three-level claim ladder preserved and owned.
+- [x] Preview/detail and equation ownership pass.
+- [x] Scientific bridges S1–S10 pass under the S9 placeholder/omit boundary.
+- [x] No harmful evidence duplication or estimand pooling found.
+- [x] Exact S1 named Kármán/Vortex/Illusion/Erase status correction specified.
+- [x] Existing registry + S7/S8/S9 audits designated as controlling S7–S9 record.
+- [ ] S1 correction integrated by original Unit-A writer.
+- [ ] S9 omit-versus-author-written branch selected and recorded.
+- [ ] Section/interface statuses propagated consistently through registry/ledgers.
+- [ ] Final title selected or provisional status explicitly retained for assembly.
+- [ ] Repository bibliography keys verified and citations integrated.
+- [ ] Evidence/data global pass run.
+- [ ] Length/display global pass run with one reconciled counting method.
+- [ ] Terminology/style/TeX preflight and final audit run.
+
+## Knowledge and evidence-use closeout
+
+- **Nowledge:** three narrow `DynamisLab JFM` memory searches used for R5 global controls, S1 named-task status and S7–S9 interface/branch navigation. Repository controls overrode memory.
+- **Repository:** all current S1–S10 TeX, ten section audits, two existing wave audits, R5 registry/ledgers/context packets/rules, and current R4 global and section controls were read for architecture/status. `read` here is not independent scientific or artifact verification.
+- **Literature:** no search and no PDF read in this pass; only the repository-recorded citation states were inspected as control metadata.
+- **Not used:** thread recall, `mem_fs`, Working Memory, durable-memory write, scripts for scientific analysis, evaluator execution, CFD, training, rendering, artifact generation or final-master work.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/SECTION_01_AUDIT.md

@@ -0,0 +1,39 @@
+# Section 1 audit — Reader-facing citation compression
+
+**Unit:** A | **Writer ID:** `5ddac128-c9c7-4b83-8fb9-aed05023a9b0` | **Stage:** `citation-compression` | **Output:** `../sections/SECTION_01_INTRODUCTION.tex`
+
+## One-to-one paragraph cards
+
+Counts exclude the section heading and citation commands/keys.
+
+| ID | Reader-facing job | Citation role and inference limit | Actual words | Status |
+|---|---|---|---:|---|
+| S01-P01 | Define cloak/illusion; separate transformation, active-source, non-zero illusion, free-flow hydrodynamic and finite-$Re$ boundaries | Source roles are split across clear sentences; local candidates remain broad mapped propositions; no linear-to-viscous transfer or priority | 197 | pass |
+| S01-P02 | Select pinball and distinguish force/AFC context from target-relative field testing | Deng wording is title-level; Feng is force control only; force-versus-field caveat appears once | 157 | pass |
+| S01-P03 | Motivate DRL lineage and flow/task-dependent sparse sensing | Ren appears only here; Fan/Li--Zhang retain bounded roles; no PPO algorithm definition or Beem analogy | 167 | pass |
+| S01-P04 | Preview deployed policy-map reduction and distinguish GP/SINDy/PySR compactly before S5 | S5 owns taxonomy detail; PySR is implementation identity; no governing-equation or offline-validation claim; CCD remains noncausal | 174 | pass |
+| S01-P05 | State execution scope, Kármán-only field proximity, Erase/steady boundaries and S2 bridge | Unchanged; literature supports no internal result | 208 | pass |
+
+**Total:** **903 publishable prose words**, down **220** from the 1,123-word citation-expansion version and at the requested compression boundary.
+
+## Citation inventory
+
+**Current S1:** **25 unique keys / 26 occurrences**.
+
+**Preserved keys:** `PendryEtAl2006Controlling`, `Leonhardt2006OpticalMapping`, `Miller2006PerfectActiveCloaking`, `VasquezEtAl2009ActiveExterior`, `MaEtAl2013ActiveCloaking`, `LinEtAl2021ActiveAcousticIllusions`, `ChenEtAl2024MultilayerHydrodynamicCloak`, `UrzhumovSmith2012ActiveCloaks`, `DengEtAl2020PinballBifurcations`, `CornejoMacedaEtAl2021PinballControl`, `RaibaudoEtAl2020PinballControl`, `DengEtAl2021PinballForceModel`, `FengWang2010SyntheticJetCylinder`, `FengEtAl2023PinballForceControl`, `RabaultEtAl2019DRLFlowControl`, `ParisEtAl2021RobustFlowControl`, `FanEtAl2020BluffBodyRL`, `RenEtAl2021HydrodynamicStealthDRL`, `LiZhang2022ConfinedWakeRL`, `GautierEtAl2015ClosedLoop`, `LiEtAl2017GeneticProgrammingControl`, `BruntonEtAl2016SINDy`, `Cranmer2023PySR`, `LoiseauEtAl2018SparseROM`, `LiEtAl2022FlowEstimation`.
+
+**Became unused in S1:** `SchulmanEtAl2017PPO` (S2 owns the algorithm definition) and `BeemTriantafyllou2015WakeSensing` (retained only in S2). No bibliography entry was deleted.
+
+Mapped/full-text status is unchanged and not adjudicated in this pass. All propositions remain broad, with no new detail or number.
+
+## Compression and quarantine audit
+
+- Duplicate Ren use removed; the hydrodynamic-stealth precedent remains only in P3.
+- Four-key active-cloak compound sentence split into active-source and non-zero-illusion roles.
+- Chen now reads exactly at title level: `a multilayer hydrodynamic cloak in free flow has been reported`.
+- Deng now reads: `Transient and post-transient pinball-force dynamics have been treated with reduced modelling`.
+- Feng says force control, not “control and tracking”.
+- P4 compresses GP/SINDy/PySR and points detail to S5; PySR is named as implementation only.
+- Hard quarantine remains intact: no positive Illusion SR, cross-case SR transfer, OID, causal/validating CCD, old chain/number, experiment validation or priority wording.
+
+**Disposition:** `citation-compression`; five jobs retained and final/global disposition unchanged.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/SECTION_02_AUDIT.md

@@ -0,0 +1,43 @@
+# Section 2 audit — Reader-facing citation compression
+
+**Unit:** A | **Writer ID:** `5ddac128-c9c7-4b83-8fb9-aed05023a9b0` | **Lifecycle stage:** `citation-compression` | **Output:** `../sections/SECTION_02_PROBLEM_METHOD_EVALUATION.tex`
+
+## One-to-one paragraph cards
+
+Thirteen cards map to the reader-facing prose paragraphs; equation introductions and their following prose form one job. Displayed equations and citation commands/keys are excluded.
+
+| ID | Reader-facing job | Citation role and inference limit | Actual words | Status |
+|---|---|---|---:|---|
+| S02-P01a | Pinball, coordinates, confinement and separate configurations | No plant equivalence or cross-configuration pooling | 153 | pass |
+| S02-P01b | Scaling, $Re_D/U_0$, LBM and implementation split | No inherited conversion or implementation equivalence | 159 | pass-with-gates |
+| S02-P02a | Action scaling and compact force-context bridge | Deng/Feng motivate force as compact signal only; field caveat stated once | 161 | pass |
+| S02-P02b | Force/probe observations and complementary measurement roles | Centre/off-centre language is design intent; Nair/Marra roles bounded; Beem retained once here | 128 | pass |
+| S02-P02c | Placement/lag, periodic sensing and temporal-history limits | Gong/Jin/Wang retain distinct bounded roles; no present optimization/observability claim | 154 | pass |
+| S02-P02d | PPO and individual-policy reproduction | S2 retains `SchulmanEtAl2017PPO`; no convergence guarantee | 101 | pass-with-gates |
+| S02-P03a | Acquisition roles and cloak/illusion targets | Target and physical zero remain non-interchangeable | 113 | pass |
+| S02-P03b | Force-plus-sensor objective and field boundary | Reward remains signal-space only | 103 | pass-with-gates |
+| S02-P04a | Concise DTW genealogy, conventions and physical limits | Generic sources define dynamic programming only; project conventions remain explicit | 104 | pass |
+| S02-P04b | Registered ROI, mean-field error and geometry guard | No solver-exact common-fluid-mask claim | 103 | pass-with-gates |
+| S02-P04c | Phase/event/mean/steady measures and DTW/field separation | No pooled score or cross-measure transfer | 134 | pass |
+| S02-P05a | K2/S2 observable qualification | K2 remains `partial pass`; no control/cloaking qualification | 110 | pass-with-gates |
+| S02-P05b | Definitions-to-results bridge | No performance preview or transfer | 31 | pass |
+
+**Total:** **1,554 publishable prose words**, down **140** from the 1,694-word citation-expansion version and at the requested compression boundary.
+
+## Citation inventory
+
+**Current S2:** **14 unique keys / 14 occurrences**; no S2 citation became unused.
+
+Keys retained: `DengEtAl2021PinballForceModel`, `FengEtAl2023PinballForceControl`, `NairEtAl2021PhaseFlowControl`, `MarraEtAl2024ActuationManifold`, `RaibaudoEtAl2020PinballControl`, `CornejoMacedaEtAl2021PinballControl`, `LiEtAl2022FlowEstimation`, `BeemTriantafyllou2015WakeSensing`, `GongEtAl2020VortexEstimation`, `JinEtAl2022SensorActuatorPlacement`, `WangEtAl2024DynamicFeatureDRL`, `SchulmanEtAl2017PPO`, `SakoeChiba1978DynamicProgramming`, `Muller2015FundamentalsMusicProcessing`.
+
+Mapped/full-text status is unchanged and not adjudicated. No bibliography entry was deleted.
+
+## Compression and quarantine audit
+
+- Probe wording now states design intent: the centre/off-centre probes are `intended to sample` phase, deflection and cross-wake motion.
+- Force/field non-equivalence is stated once in the action bridge and once where the online objective is formally bounded; duplicated literature caveats were removed.
+- DTW was reduced from 225 to 104 words while retaining dynamic-programming genealogy, limited local time reparameterization, declared endpoint/window/step constraints, distance/similarity and native/normalized conventions, and the non-physical-delay/non-field-equivalence warning.
+- Project facts, configurations, equations, $U_0/Re_D$, roles, spatial metrics and qualification statements are unchanged.
+- Hard quarantine remains intact: no old chain, positive Illusion SR, cross-case SR, OID, causal/validating CCD, experiment validation, old internal value or priority claim.
+
+**Disposition:** `citation-compression`; thirteen prose cards/five frozen jobs retained and final/global disposition unchanged.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/SECTION_03_AUDIT.md

@@ -0,0 +1,46 @@
+# Section 3 audit — Unit B first lifecycle
+
+**Writer:** `a7e7df0b-a529-47e3-85b5-95683f9e7e24`  
+**Output:** `sections/SECTION_03_CAPABILITY_PROGRESSION.tex`  
+**Stage at first draft:** `first-draft`; literature diagnosis and integrated rewrite recorded below.
+
+## Evidence cards
+
+| ID | Job and claim ceiling | Comparator / coordinate / authority | Display dependency and local caveat | Fallback | Status |
+|---|---|---|---|---|---|
+| S03-P01 | Progression contract; four increasing task complexities, no pooled efficacy | steady endpoint; periodic Kármán phase; Vortex event coordinate; Illusion literal phase; R4 S3 packet and global ledger | no future-display narration in body; accepted interface and all exact panel/provenance/caption/art gates remain open here | prose progression | drafted; 89 words |
+| S03-P02 | Qualitative steady seed in two velocity components | target vs controlled endpoint vs stationary, $x/D=10$; steady profile CSV/manifest and S3-1 Plan | endpoint only; S8 owns spin, settling, metric and theory | component-profile prose | drafted; 96 words |
+| S03-P03 | Rounded Kármán preview and S4 ownership | target vs frozen PPO vs physical zero; complete-cycle mean and separate phase-8; R4 S3/S4 and wake-L2 authority | about 82% dimensionless reduction only; errors historical-$U_0$ normalized; $U_0/U_{ref}$ unbound; reward/visual/L2 distinct | qualitative role ordering | drafted; 109 words |
+| S03-P04 | Vortex morphology at one event coordinate | target/controlled/zero at literal offset; S3-2 Plan and mapped event fields | exact offset/provenance and evaluator remain open; body makes no scalar, evolution, suppression or tracking claim | morphology only | drafted; 78 words |
+| S03-P05 | Executed non-zero-target PPO case; morphology consistent with retargeting | same acquisition: $0.75L_0$ target / seed-43 PPO controlled / physical zero at literal phase; current V5 role metadata and plot manifest | no field scalar: no approach-to-target or improvement-over-zero claim; no positive Illusion SR | exact three-role morphology | drafted; 92 words |
+| S03-P06 | Why Kármán receives depth | evidence-intersection comparison, R4 S3/S4 ownership ledger | not representative or transferable; exact detail remains S4-owned | evidence-based bridge | drafted; 102 words |
+
+## Contract and evidence audit
+
+- Six paragraph jobs remain in frozen order. The body now reads as a physical progression rather than a future figure plan; S3 retains only the rounded Kármán ordering/reduction and leaves exact errors, ROI, phasing, DTW and actions to S4.
+- The accepted S3-1/S3-2 interface remains an audit-level production control. Crop, rendering, source-payload, fallback and readiness wording was removed from the body and retained in this audit.
+- Illusion role authority is resolved: the current V5 plot manifest groups target `v5/ill_075L/target`, seed-43 controlled `v5/ill_075L_seed43/controlled` and physical zero `v5/ill_075L/zero` as `ill_075L_seed43`; all three role metadata files bind case `ill_075L`, 750 warm-up steps, 250 collected boundaries, sample interval 1100, cycle window 18, the same configuration hash `2492cc49...7539`, target hash `dcaa97c4...f7b` and common phase procedure. They are separate role trajectories within one retained same-case acquisition contract, not one trajectory. S3 and S7-compatible ceiling: **the PPO policy was executed toward the non-zero target and the controlled morphology is consistent with retargeting; without a scalar field comparison, do not claim approach to target, improvement over physical zero, full-state matching or positive Illusion SR.** This ceiling is authoritative for S3 and recommended to the S7 owner; S7 was not edited.
+- Internal statements bind the current R4 packets, Figure Plans, steady manifest/profile source and S4 wake-L2 authority. Repository reads verify paths and recorded values; raw role arrays, final hashes and production crops were not independently audited in this lifecycle.
+- No citation, label or reference key is present. A repository search found citation commands in the draft guide but no verified bibliography database or key-definition authority for these section files. Exact request: the bibliography owner must bind intended S3 citations to existing repository key definitions before any `\cite` command is inserted. Literature remains rhetoric-only.
+- Evidence/data reconciliation retains the rounded approximately 82% preview as a dimensionless reduction while stating that its underlying field errors are historical-$U_0$ normalized and that $U_0/U_{ref}$ remains unbound.
+- Coordinator narrow review revised S03-P03 comparator-first: the three roles form the declared comparison, and controlled is closer than physical zero under each named estimand; the rounded approximately 82% preview and all caveats are unchanged.
+- Approved S2--S3 interface findings imported from `audits/SECTION_02_AUDIT.md#s2s3-interface-audit`: exact configuration and physical-zero terminology are retained; definitions are imported without bare inherited Reynolds labels, preview/detail transfer or signal/field pooling. No competing interface sidecar was created.
+
+## Bounded literature diagnosis and integrated changes
+
+The complete first drafts were diagnosed with five closest papers already present in the DynamisLab workspace/library; metadata, DOI and PDF availability were verified before reading. Each was read at `full-PDF` level with a section-specific, rhetoric-only question: Rabault et al., *Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control* (JFM 865, DOI `10.1017/jfm.2019.62`); Paris et al., *Robust flow control and optimal sensor placement using deep reinforcement learning* (JFM 913, DOI `10.1017/jfm.2020.1170`); Xia et al., *Active flow control for bluff body drag reduction using reinforcement learning with partial measurements* (JFM 981, DOI `10.1017/jfm.2024.69`); Wang et al., *Dynamic feature-based deep reinforcement learning for flow control of circular cylinder with sparse surface pressure sensing* (JFM 988, DOI `10.1017/jfm.2024.333`); and Cornejo Maceda et al., *Stabilization of the fluidic pinball with gradient-enriched machine learning control* (JFM 917, DOI `10.1017/jfm.2021.301`).
+
+Cross-paper diagnosis: close results sections define the reference/baseline before interpretation, separate aggregated learning records from a representative deterministic deployment, move from the principal comparison to fields and then subordinate sensor/action diagnostics, let captions decode roles/windows/encodings, and place realization or envelope limits beside the affected result. The papers differ in scalar-first versus field-first ordering; no universal order was inferred.
+
+Reader-facing sprint changes: recast the opening as a four-step physical progression; replaced preview, crop, rendering, payload, audit and fallback narration with direct role comparisons; resolved the exact three-role Illusion contract and applied the no-scalar morphology ceiling; retained endpoint, event-coordinate and target/control limits locally; and preserved S4 ownership of exact Kármán errors, ROI, phasing, DTW and actions. Production controls and fallbacks remain below. No literature-derived proposition, new fact, citation or artifact-status change was introduced.
+
+## Gates and readiness
+
+- **OPEN:** S3-1 final crop/typography; S3-2 exact Vortex offset, exact leaf/hashes, units, scales, crops, caption and render;  publication readiness.
+- **CLOSED only:** author choice of the two-figure interface.
+- **Fallbacks:** profile-only S3-1; one source-bound Vortex role comparison with no scalar; Illusion target/control only if zero fails; no compensatory efficacy prose.
+- **Disposition:** reader-facing rewrite complete under unchanged artifact gates; not a section pass and not artifact-ready. Uniform Pass-3-method count: 566 words across six paragraph jobs (89/96/109/78/92/102). Shared ledgers were not edited in this sprint.
+
+## Knowledge and computation closeout
+
+Three narrow Nowledge searches were used for navigation/history and then checked against repository controls. Working Memory, thread recall and `mem_fs` were not used. A durable-memory write was attempted at closeout but blocked as an unauthorized external side effect; the repository audit/registry remains the record. No CFD, training, event evaluation, rendering, artifact generation or new analysis was run.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/SECTION_04_AUDIT.md

@@ -0,0 +1,48 @@
+# Section 4 audit — Unit B first lifecycle
+
+**Writer:** `a7e7df0b-a529-47e3-85b5-95683f9e7e24`  
+**Output:** `sections/SECTION_04_KARMAN_RESULTS.tex`  
+**Stage at first draft:** `first-draft`; literature diagnosis and integrated rewrite recorded below.
+
+## Evidence cards
+
+| ID | Job and claim ceiling | Comparator / estimand / authority | Display dependency and local caveat | Fallback | Status |
+|---|---|---|---|---|---|
+| S04-P01 | Finite tested condition set and metric separation | uniform/free-slip $Re_D$ and disturbance series; latest eval table/summary | non-reference $n=1$; outer radii change learning rate; no continuous trend | move disturbance group to supplement | drafted; 114 words |
+| S04-P02 | Five-history recurrence and exact checkpoints | seeds 41--45, full 500 iterations; meta-backed latest summary | no uncertainty, convergence, stability or broad robustness | retain exact histories in prose | drafted; 86 words |
+| S04-P03 | Sampled-search ceiling and reward/field separation | complete histories plus online objective authority | selected evaluation is not sixth realization; reward is not field proximity | compact boundary | drafted; 78 words |
+| S04-P04 | Configuration/evidence-chain separation | uniform/free-slip survey vs parabolic/no-slip field authority | no common reader Reynolds label or chain pooling | direct field question | drafted; 62 words |
+| S04-P05 | Reference field roles and shared-comparator lineage | parabolic/no-slip target/frozen PPO/physical zero; S4/S5 manifests and hashes | target/zero shared across S4/S5; PPO/SR controlled distinct; no independent agreement | prose role ordering | drafted; 89 words |
+| S04-P06 | Complete-cycle mean proximity | $E_{mean}^{(U_0)}$ 0.085771/0.475206; registered $20D\times10D$ ROI; wake-L2 authority | geometry guard; no persisted exact mask; no pointwise/full-domain identity | scalar comparison | drafted; 90 words |
+| S04-P07 | Eight-slot phase proximity | $E_{phase8}^{(U_0)}$ 0.112680/0.630878; independent role phasing | nearest snapshots, no ensembles/global minimization; not replication; $U_0/U_{ref}$ unbound | scalar comparison | drafted; 91 words |
+| S04-P08 | Complementary signal/action organization | DTW 0.932899/0.625127; indices 96--145; historical $tU_0/D=38.4$--58.0 | DTW not full-field equivalence; target has no action; spectrum remains exact-window gated | omit spectrum | drafted; 98 words |
+| S04-P09 | Bounded synthesis and compact-law bridge | tested-range, field, signal and action authorities | no mechanism/optimality/universal law; no SR/S6 validation or triangulation | compact unresolved question | drafted; 85 words |
+
+## Contract and evidence audit
+
+- Nine reader-facing paragraphs implement the seven frozen argumentative jobs by splitting the former condition/history and field/error blocks: tested conditions; five-history recurrence; sampled-search/reward ceiling; configuration hinge; field roles/lineage; mean L2; phase L2; signal/action evidence; compact-law bridge. No ownership or claim ceiling changed.
+- Reader-facing configuration labels remain distinct. The parabolic-source reference case has no reader-facing Reynolds value.
+- Reward, target-normalized DTW and spatial L2 remain different quantities. Mean and phase errors remain separate estimands; S4 values are not compared with S6 corrected errors.
+- Targeted S4/S5 lineage reconciliation used current named manifests/metadata/hashes only, without metric rerun. Both plotting manifests bind the same exact target path `.../legacy/karman_re100/target/phase_fields.npz` and physical-zero path `.../legacy/karman_re100/zero/phase_fields.npz`; the S5 manifest records hashes `02555a5e...53ec` and `45620f98...8ac`, respectively. S4 PPO controlled (`.../controlled/phase_fields.npz`, S5-manifest hash `ab376c1f...b606`) and S5 SR controlled (`.../sr/phase_fields.npz`, hash `4c483b90...9de`) are distinct acquisitions. Thus coincident physical-zero errors arise from shared comparator artifacts, not independent agreement; controlled errors remain non-equivalent acquisition-specific estimands.
+- The wake-L2 authority defines both errors as $U_0$-normalized and records historical source time. S4 now labels $E_{mean}^{(U_0)}$, $E_{phase8}^{(U_0)}$ and $tU_0/D$ explicitly. $U_0/U_{ref}$ remains unbound; dimensionless zero-relative reductions are retained.
+- Exact learning values and confounders were checked against `latest_results_summary.json`, `latest_eval_summary.csv` and the retained summary CSV. Direct inspection of all five complete histories verifies low initial rewards at iteration 1 (0.0552--0.1363) relative to their authoritative 0.9160--0.9412 best checkpoints, so the bounded phrase `progress from low reward to high-performing checkpoints` is retained. Role semantics and DTW values were checked against current metadata/JSON. The time window was checked against `phase_alignment.json`.
+- Coordinator narrow review replaced `independent role acquisitions` with `separately retained role acquisitions` to avoid confusion with independent validation; the acquisition and configuration boundaries are unchanged.
+- No citation, label or reference key is present. A repository search found citation commands in the draft guide but no verified bibliography database or key-definition authority for these section files. Exact request: the bibliography owner must bind intended S4 citations to existing repository key definitions before any `\cite` command is inserted. Literature remains rhetoric-only.
+
+## Bounded literature diagnosis and integrated changes
+
+The complete first drafts were diagnosed with five closest papers already present in the DynamisLab workspace/library; metadata, DOI and PDF availability were verified before reading. Each was read at `full-PDF` level with a section-specific, rhetoric-only question: Rabault et al., *Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control* (JFM 865, DOI `10.1017/jfm.2019.62`); Paris et al., *Robust flow control and optimal sensor placement using deep reinforcement learning* (JFM 913, DOI `10.1017/jfm.2020.1170`); Xia et al., *Active flow control for bluff body drag reduction using reinforcement learning with partial measurements* (JFM 981, DOI `10.1017/jfm.2024.69`); Wang et al., *Dynamic feature-based deep reinforcement learning for flow control of circular cylinder with sparse surface pressure sensing* (JFM 988, DOI `10.1017/jfm.2024.333`); and Cornejo Maceda et al., *Stabilization of the fluidic pinball with gradient-enriched machine learning control* (JFM 917, DOI `10.1017/jfm.2021.301`).
+
+Cross-paper diagnosis: close results sections define the reference/baseline before interpretation, separate aggregated learning records from a representative deterministic deployment, move from the principal comparison to fields and then subordinate sensor/action diagnostics, let captions decode roles/windows/encodings, and place realization or envelope limits beside the affected result. The papers differ in scalar-first versus field-first ordering; no universal order was inferred.
+
+Reader-facing sprint changes: separated condition sampling, five-history recurrence and reward/field inference; translated retained/historical-source/estimand/evidence-chain wording into training, trajectory, field, signal and simulation-scaling language; consolidated repeated limitations at the affected result; separated configuration ownership from the deep field comparison; split mean, phase and signal/action results into distinct estimand-led paragraphs; and replaced production narration with direct physical interpretation plus local mask, phasing, acquisition and no-target-action limits. Historical $U_0$ notation and S4/S5 shared-comparator lineage remain explicit. No literature-derived proposition, new fact, citation or artifact-status change was introduced.
+
+## Gates and readiness
+
+- **OPEN:** S4-1 production redraw and Reynolds-label metadata audit; S4-2 unit/crop/scale/caption audit (target/zero and controlled lineage hashes are now bound); exact shared spectrum window; action-unit choice; zero sensor portrait; publication readiness. Production, audit, fallback and source-payload wording is retained here rather than in the body.
+- **Fallbacks:** move disturbance group to supplement; omit spectrum first and DTW from display second; retain field comparison as prose/scalars if visual alignment fails.
+- **Disposition:** reader-facing rewrite complete with exact source-bound values and unchanged production gates; not artifact-ready or a final section pass. Uniform Pass-3-method count: 793 words across nine paragraphs (114/86/78/62/89/90/91/98/85). Shared ledgers were not edited in this sprint.
+
+## Knowledge and computation closeout
+
+Three narrow Nowledge searches were used for navigation/history and then checked against repository authorities. Working Memory, thread recall and `mem_fs` were not used. A durable-memory write was attempted at closeout but blocked as an unauthorized external side effect; the repository audit/registry remains the record. No CFD, training, spectrum computation, field redraw, artifact generation or new analysis was run.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/SECTION_05_AUDIT.md

@@ -0,0 +1,39 @@
+# Section 5 Round-5 audit
+
+**Writer:** `cf2466a8-c3c8-44e9-a926-e7d826a8816a`  
+**Stage:** integrated rewrite / conditional wave audit. **Output:** `sections/SECTION_05_KARMAN_SR_LAW.tex`. **Wave audit:** `WAVE_S04_S06_INTERFACE_AUDIT.md`.
+
+## Paragraph evidence cards
+
+- **S05-P01a — reduction question.** Claim: the persistent rear-rotation hierarchy motivates an executable surrogate for the fixed PPO policy. Authority: R4 S5 §§2,4,9 and `src/SR_analysis/{CLAIMS.md,results/README.md}` (`read`). Boundary: no flow model, mechanism or shared S4/S6 comparison.
+- **S05-P01b — external method distinction.** Claim: direct genetic-programming controller search evaluates explicit laws in the control loop, whereas SINDy identifies sparse dynamics and SINDy-RL jointly learns sparse models with a policy. External support: `GautierEtAl2015ClosedLoop` and full-PDF `DebienEtAl2016GeneticProgrammingRamp`, `RenEtAl2019MachineLearningVIV`, `CastellanosEtAl2022FewSensorMLControl`, `ZolmanEtAl2025SINDyRL`; `BruntonEtAl2016SINDy` is local mapped and supports only the broad dynamical-identification identity. Boundary: these are different mathematical tasks and do not validate the present law.
+- **S05-P01c — post-hoc policy map.** Claim: the current SR fit is post-hoc regression of a fixed PPO state--action map, with post-state `i` predicting action `i+1`; deployment, not offline fit, tests control. `Cranmer2023PySR` supplies local mapped software/search-method identity only. Authority for internal alignment and role: R4 S5 and current SR package (`read`). Boundary: no novelty, priority, governing-equation or literature-validation claim.
+- **S05-P02 — imposed map.** Claim: $Gx$ is the centreline-reflected measured state, with exact sensor, velocity, force and action parity; the imposed map creates an odd front head and shared rear generator. Authority: R4 S5 §§4,6 and canonical formula authority (`read`). Boundary: imposed after PPO collection; no PPO/plant symmetry proof.
+- **S05-P03a — executable law.** Claim: native signed-force conventions and $C_{d,i}=2f_{x,i}/(\rho U_0^2D)$, $C_{l,i}=2f_{y,i}/(\rho U_0^2D)$ with retained lattice $\rho=1$ bind the coefficient interpretation. Authority: `SR_analysis/utils/{feature_builder.py,g_operator.py}`, canonical formula artifacts, `CLAIMS.md` and `results/README.md` (`read`). Equation licensed.
+- **S05-P03b — action units.** Claim: historical $\alpha=\omega/U_0$ and the general surface-speed conversion are retained. Gate: $U_0=U_{ref}$ unresolved. Fallback: no $s=\alpha R$ simplification.
+- **S05-P04a — deployment selection.** Claim: simplicity/offline fit nominates a map, while execution on surrogate-visited states selects its control meaning. Authority: current SR package (`read`). Literature positions the distinction only.
+- **S05-P04b — trajectory metric.** Claim: exact 480-settling/160-recorded-boundary result and rolling native-DTW contract: 30-boundary window, one circular lag, six channel scores averaged. Authority: role-local `dtw_summary.json`/`metadata.json`, SR `PIPELINE.md`, claims and acquisition README (`read`). Boundary: lag is comparison metadata, not delay.
+- **S05-P04c — spatial metrics.** Claim: exact SR/zero $E_{\mathrm{mean}}$ and $E_{\mathrm{phase8}}$ use historical-$U_0$-normalized two-component ROI RMS, with complete-cycle means versus eight nearest-boundary phase slots; target, SR and zero share same-case comparator lineage. Authority: `drl_pinball/data/reproduction/sr_wake_l2_summary.json` (`read`). Boundary: independently acquired S4 PPO-controlled errors are not repeat measurements.
+- **S05-P05a — deletion measurement.** Claim: exact ranking under the same 40-step, six-channel, shared-lag native-DTW parent/deletion contract. Authority: `term_deletion.{md,csv}` (`read`).
+- **S05-P05b — deletion interpretation.** Claim: the rear constant has the largest listed degradation in each tested case, but each deletion changes the subsequent trajectory. Boundary: parent-relative and noncausal; no necessity or mechanism.
+
+## Contract audit
+
+Five of five frozen jobs drafted across eleven one-to-one paragraph cards. Positive SR is Kármán/cloak-only; no Illusion-positive result. Symmetry is imposed, deletion is parent-relative/noncausal, and S4/S5/S6 estimands remain separated. Reader-facing citations use only R5-owned keys; no label or reference is fabricated. Required equations are source-bound, and `operatorname{odd}` is defined locally by the displayed antisymmetrization. Figure, action-extraction, target reconciliation and historical velocity-reference gates remain open; optional deployment envelope omitted.
+
+Literature diagnosis used existing DynamisLab full PDFs only; details and integrated fixes are recorded in the citation ledger. It supported ordering and caveat placement, not internal facts.
+
+## Length, share and production fields
+
+- **Uniform substantive words:** **899** by the established global count; 5/5 jobs retained.
+- **Allocation/share:** 15% allocation reserved; actual article share unresolved until the complete article. S4–S6 jointly target 44% with an ordinary 42% floor. No prose is added to consume an illustrative allocation.
+- **Bibliography/citation inventory:** 7 active S5 keys, all resolving in `../R5_REFERENCES.bib`: `GautierEtAl2015ClosedLoop`, `DebienEtAl2016GeneticProgrammingRamp`, `RenEtAl2019MachineLearningVIV`, `CastellanosEtAl2022FewSensorMLControl`, `BruntonEtAl2016SINDy`, `ZolmanEtAl2025SINDyRL`, and `Cranmer2023PySR`. `CornejoMacedaEtAl2021PinballControl` remains in the bibliography but is no longer cited in S5 after the method distinction rewrite.
+- **Displays:** labels, callouts, caption wording, action extraction and production provenance remain deferred/gated; optional deployment envelope remains omitted.
+- **Wave disposition:** conditional; see `WAVE_S04_S06_INTERFACE_AUDIT.md`.
+
+## Reader-facing sprint audit
+
+The rewrite retains five evidence jobs in a more readable progression: physical action hierarchy → surrogate/execution question → imposed symmetry → executable law and units → closed-loop evidence → parent-relative deletion and field bridge. Body prose contains no redraw, recomposition, gate or publication-readiness language; those dependencies remain in this audit. Dense execution and deletion material is split so that the measurement contract precedes its interpretation; no internal value, case, mechanism or claim was added. Current uniform recount is **899** substantive words by the established global method.
+## Citation integration closeout
+
+S5 adds six active keys to the prior `GautierEtAl2015ClosedLoop` citation: local mapped `BruntonEtAl2016SINDy` and `Cranmer2023PySR` support broad method identities only; full-PDF `DebienEtAl2016GeneticProgrammingRamp`, `RenEtAl2019MachineLearningVIV`, `CastellanosEtAl2022FewSensorMLControl` and `ZolmanEtAl2025SINDyRL` support the more specific direct-controller, executed-behaviour and joint model-based RL distinctions recorded in the citation ledger. No citation supports an internal coefficient, alignment, value, deletion ranking or performance claim. `CornejoMacedaEtAl2021PinballControl` was removed from S5 because the expanded sentence now has narrower method owners; its bibliography record is unchanged. `LiEtAl2017GeneticProgrammingControl` is now uncited pending global bibliography pruning because the full-PDF direct-controller sources own the sentence. Optional constrained Galerkin was not imported.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/SECTION_06_AUDIT.md

@@ -0,0 +1,42 @@
+# Section 6 Round-5 audit
+
+**Writer:** `cf2466a8-c3c8-44e9-a926-e7d826a8816a`  
+**Stage:** integrated rewrite / conditional wave audit. **Output:** `sections/SECTION_06_KARMAN_FIELD_CCD.tex`. **Wave audit:** `WAVE_S04_S06_INTERFACE_AUDIT.md`.
+
+## Paragraph evidence cards
+
+- **S06-P01 — four-role means.** Claim: exact target/zero/constant/PPO-controlled mean comparison with labels $T$, $0$, $C$ and $D$ defined locally, with the target consistently represented by its retained mean $\overline{\boldsymbol{u}}_T$. Authority: `CCD_analysis/{FINAL_RESULTS.md,MATHEMATICAL_DERIVATION.md,karman_dynamic/CORRECTED_MATH_CONTRACT.md}` (`read`). **Exact retained four-role mean-window binding: BOUND.** Target, zero, constant-mean and PPO-controlled roles each use half-open `retained_slice: [480,840]`, 360 boundaries. Exact pointers: current authority parent paths in `/home/frank14f/optane/DynamisLab/ccd/karman-dynamic/canonical/results/roi-mean/summary.json` → each `sources/{target,zero,constant_mean,drl}/campaign.json`; computation is literal in `src/CCD_analysis/karman_dynamic/dynamic_increment.py::_dense_statistics` (`ux[start:stop]`, `uy[start:stop]`). The ROI result manifest SHA256 is `f7f6141042678e52314a4d8f76d589108da8e844c8ce504f8f523a6f88a17e73` in `src/CCD_analysis/artifact_catalog.json` and the validated publication checkpoint. Metric: inclusive `34<=x/D<=54`, `|y/D|<=5`, exact common mask, 80,200 points, coordinate weighting. Caveat: distinct from S4/S5.
+- **S06-P02 — signed fields.** Claim: the four retained means lead to a component-level physical reading of `zero-target` deficit and largely opposing `PPO-controlled minus zero` addition, explicitly separated from the two-component scalar metric. Authority: R4 S6 and Figure Plan bound stored means (`read/mapped`). Boundary: the component view does not establish momentum restoration. Composite OPEN; fallback prose plus existing zero-target.
+- **S06-P03 — scalar accounting.** Claim: `E0=0.4694352273`, `EC=0.1110140739`, `ED=0.0857787860`, shares 93.42/6.58, followed immediately by denominator and non-energy/non-residual-norm explanation. Authority: `FINAL_RESULTS.md`, derivation and results index (`read`). Caveat: scalar error differences, not energy, variance, physical/modal contribution or causality. Equation licensed.
+- **S06-P04 — residual.** Claim: the two-component velocity notation $\boldsymbol{u}_r(k)$ is used consistently, and a separately centred 19-cycle by 10-bin non-paired residual $\boldsymbol{u}^{\mathrm{res}}_{c,b}(k)$ is constructed as a field object distinct from scalar error differences. Authority: corrected math contract and derivation (`read`). External context: full-PDF `TadmorEtAl2010CylinderMeanField` supports only the mean/shift-mode distinction; the body explicitly denies identity with the present subtraction. Caveat: no pairing, lag, interpolation, whitening, preprojection or counterfactual. Residual equation licensed.
+- **S06-P05 — CCD/POD.** Claim: canonical correlation decomposition (CCD) is expanded at first use; exact action-coordinate preprocessing leads to rank-3 action-correlated correction-field modes; POD then supplies the identical-object field-only boundary. Authority: `FINAL_RESULTS.md`, derivation, `karman_dynamic/{interpretation.py,mode_diagnostics.py}`, real-CCD schema and figure/data contract (`read`). Action convention: native counter-clockwise-positive front/upper/lower effective actions in solver angular-velocity command units; orthonormal rear sum/difference coordinates; physical mean removal followed by empirical centring; no standardization/whitening. Caveat: descriptive/noncausal; action association only; POD detail supplemental. The body-level “essentially the same tested rank-3 field subspace” sentence is manuscript-licensed by current `FINAL_RESULTS.md`, `CLAIMS.md`, `RESULTS_INDEX.md` and `WRITING_HANDOFF.md`, not merely an internal diagnostic. External ancestry: full-PDF `JordanEtAl2007ObservableJetModes` and `DiscettiEtAl2018CorrelatedFieldEstimation` support observable-ranked and point--field correlation propositions; current full-PDF Borée/Loiseau/Schlegel sources retain EPOD, observable-associated and POD boundaries. Composite OPEN.
+- **S06-P06 — synthesis.** Claim: mean-dominant ledger plus a residual associated with the smaller constant-to-PPO-controlled scalar increment and organized through descriptive action association, case-specific. The 6.58% scalar share is not a residual-norm statement. Authority: R4 S6 frozen synthesis (`read`). Caveat: no SR-coordinate identity, triangulation, transfer or mechanism.
+
+## Contract audit
+
+Six of six jobs drafted. Four-role order, signed subtraction order, scalar semantics, non-pairing, fixed CCD term and POD boundary pass. Internal complementarity values `-0.9961/0.1662` are not printed. CCD operator omitted from body because the residual and conceptual boundary suffice. Main display dependencies remain in this audit only; cross-cloak audit omitted from body and sprint. Reader-facing citations use only R5-owned keys; no label or reference is fabricated.
+
+Literature diagnosis used existing DynamisLab full PDFs only; details and integrated fixes are recorded in the citation ledger. Literature did not verify any internal value or promote artwork.
+
+## Length, share and production fields
+
+- **Uniform substantive words:** **747** by the established global count; 6/6 jobs retained.
+- **Allocation/share:** 16% allocation reserved; actual article share unresolved until the complete article. S4–S6 jointly target 44% with an ordinary 42% floor. No prose is added to consume an illustrative allocation.
+- **Source binding:** exact target/zero/constant/PPO-controlled mean windows are **BOUND** to the same 360 role-local boundaries, half-open slice `[480,840)`, by the current ROI result parent manifests and literal mean implementation. No value was altered or inferred.
+- **Bibliography/citation inventory:** all 6 active S6 keys resolve in `../R5_REFERENCES.bib`: `TadmorEtAl2010CylinderMeanField`, `Boree2003ExtendedPOD`, `JordanEtAl2007ObservableJetModes`, `DiscettiEtAl2018CorrelatedFieldEstimation`, `LoiseauEtAl2018SparseROM`, and `SchlegelEtAl2012LeastOrder`.
+- **Displays:** labels, callouts, caption wording, exact retained-window provenance and composite production remain deferred/gated.
+- **Wave disposition:** conditional; see `WAVE_S04_S06_INTERFACE_AUDIT.md`.
+
+## Approved evidence/data delta — S10 handoff
+
+- S6 now uses $\overline{\boldsymbol{u}}_T$ consistently in the mean-error equation.
+- “Smaller residual” is prohibited: the 6.58% value describes the constant-to-PPO-controlled share of the scalar mean-field error reduction, not a residual norm.
+- **Exact minimal Unit-E recommendation for S10-D02 (do not import as new evidence):** replace `while the smaller residual is organized descriptively through action coordinates as` with `while the separately centred residual associated with the smaller constant-to-PPO-controlled scalar increment is organized descriptively through action coordinates as`. Retain the following fixed term `\emph{action-correlated correction-field modes}` and all existing acquisition/estimand and noncausal caveats.
+- The exact four-role mean window remains authority-bound to 360 role-local retained boundaries, half-open slice `[480,840)`, under the source chain recorded in S06-P01.
+
+## Reader-facing sprint audit
+
+The rewrite retains six evidence jobs as distinct readable paragraphs: four-role construction/metric → signed mean fields → scalar accounting → residual definition → CCD/action association → matched POD boundary and synthesis. The CCD and POD material is split into construction, association and field-only comparison, while keeping the `[480,840)`/360-boundary window, exact action coordinates, target-mean notation and noncausal ceilings. Body prose contains no redraw, recomposition, gate or publication-readiness language; those dependencies remain here. Current uniform recount is **747** substantive words by the established global method.
+## Citation integration closeout
+
+S6 adds `TadmorEtAl2010CylinderMeanField` for the bounded mean/shift-mode context and `JordanEtAl2007ObservableJetModes` plus `DiscettiEtAl2018CorrelatedFieldEstimation` for observable-ranked and point--field correlation ancestry. All three are full-PDF propositions recorded in the Undermind expansion audit. Existing `Boree2003ExtendedPOD`, `LoiseauEtAl2018SparseROM` and `SchlegelEtAl2012LeastOrder` retain the EPOD, observable-associated and POD/least-order boundaries. No citation supports the four-flow errors, percentages, residual construction, rank-3 project result, subspace comparison, causality, superiority or triangulation. The fixed term `\emph{action-correlated correction-field modes}` is unchanged; OID and PCD are not imported.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/SECTION_07_AUDIT.md

@@ -0,0 +1,46 @@
+# Section 7 Round-5 audit — Unit D
+
+**Output:** `sections/SECTION_07_CROSS_CASE_EXTENSIONS.tex`  
+**Writer agent ID:** `da7067b9-38f7-4b50-bc3b-e128f9f08a2f`  
+**Stage:** `section-pass` after first draft, bounded literature diagnosis and integrated rewrite  
+**Counting method:** regex count of prose after removing the section heading and displayed-equation environments; TeX commands/symbols are not counted as words.  
+**Actual body words:** **502**; provisional S7 envelope **640–800**; illustrative midpoint **696**; section has **5/5 jobs**. No authorized article total exists, so an actual article share cannot yet be computed; the controlling provisional share remains **8%**.
+
+## Evidence cards
+
+| ID | Job and licensed claim | Comparator / definition | Internal authority and state | Metric / display dependency | External diagnosis | Adjacent caveat / fallback | Status |
+|---|---|---|---|---|---|---|---|
+| S07-P01 | Change clock/target definitions without importing Kármán depth | periodic phase vs event-relative offset; background vs non-zero or clean target; local role triplets | R4 S7 packet and global ledger `read` | no common metric is formed | Den20 full-PDF supports regime-clock separation only | role acquisitions, clocks and estimands remain case-local | pass |
+| S07-P02 | Vortex role identity and morphology at central Taylor incidence, offset `+10`; reactive/local-information hypothesis only | target / frozen-PPO controlled / physical-zero | role metadata `read`; event-field paths and evaluator exclusion `path-confirmed`; raw arrays not recomputed | no scalar field evaluation reported; morphology only | Den20 full-PDF supports placing failure beside visual comparison | no scalar, controlled-below-zero, arbitrary-family, observability or memorylessness claim | pass |
+| S07-P03 | bounded `0.75` target-size seed-43 PPO retargeting statement; negative SR boundary | declared non-zero target; exact role list intentionally omitted pending concurrent authority reconciliation | R4 S7 packet and Figure Plan `read`; retained sources `mapped`, no R5 recomputation | policy executed toward declared target; morphology consistent with retargeting; no scalar change in target-relative error reported | Mac20 full-PDF supports controller/mean-field boundaries only | no positive Illusion SR or Kármán-surrogate transfer | pass |
+| S07-P04 | Erase has clean-target, controlled and physical-zero role material only | one late clean target vs eight controlled/zero phase-referenced fields | role metadata `read`; evaluator exclusion and source paths verified | no scalar field evaluation; morphology only; temporal estimands remain distinct | Den20 full-PDF supports local temporal-object definitions | rolling reward is not independent clean-field efficacy; no scalar or success wording | pass |
+| S07-P05 | synthesize available material and enforce strict non-transfer | distinct configurations, acquisitions and estimands | global ledger, context packet and R4 S7 exclusions `read` | no cross-case metric or transferred interpretation | cross-paper pattern: retain local negative routes and limitations | no common efficacy, robustness, mechanism, SR/CCD transfer or universality | pass |
+
+## First-draft literature diagnosis and integrated fixes
+
+After the complete first draft, five close existing-workspace papers were read as full PDFs for Unit D; S7 used the relevant diagnoses from Deng et al. and Cornejo Maceda et al. The clear contract-preserving fixes retained in the integrated text were: define the event-relative clock before interpreting the offset; place the no-scalar boundary immediately after the Vortex morphology statement; renew roles and target semantics for Illusion and Erase; and keep the negative controller/evaluator routes local rather than in a detached limitations paragraph. Literature supplied rhetoric only and did not verify any project acquisition, value or artifact. No citation command or bibliography key was inserted because no repository-approved key was established.
+
+## Reader-facing sprint revision
+
+The body now uses physical cases, trajectories and fields as sentence subjects. A density follow-up retained only authority-licensed physical explanation: the common Vortex event coordinate aligns incoming-structure/pinball/downstream-wake morphology by role; Illusion compares against a declared non-zero wake pattern rather than undisturbed inflow; and Erase snapshot averaging would remove phase variation whereas snapshot-wise aggregation would retain it. No more specific shape, attenuation or recovery observation is licensed by the packet, so none was added. Evaluator, provenance, artwork, production and gate-management language was removed from reader-facing prose. Missing evidence is expressed scientifically: no scalar field evaluation is reported for Vortex or Erase, so only morphology is considered. Illusion retains only the bounded non-zero-target execution/retargeting-morphology statement and all non-transfer boundaries remain adjacent. The existing bibliography was checked: `CornejoMaceda2021` is verified, but no external citation was needed for the revised internal case statements; the remaining diagnosed papers still require approved bibliography keys before any optional citation is inserted.
+
+## Final high-value polish
+
+Illusion now uses the bounded statement required during concurrent role reconciliation: the seed-43 PPO policy was executed toward a declared non-zero target and its morphology is consistent with retargeting; no scalar change in target-relative error is reported. The exact Illusion role list is intentionally absent from prose, so a later Unit-B authority resolution cannot conflict with S7 wording. `PPO-controlled` is standardized across Vortex, Illusion and Erase. Acquisition taxonomy and repeated caveat inventories were compressed into physical field and temporal-comparison language.
+
+## Contract and interface audit
+
+- Five paragraph jobs are present. The inventory's compact mapping (Vortex combined into one paragraph; separate Illusion, Erase and envelope paragraphs) preserves the frozen five-job architecture.
+- S7 imports no Kármán number, SR law, CCD result or mechanism. `Kármán` appears only as the explicit non-transfer comparator.
+- Vortex offset `+10` is called event-relative acquisition index, never phase or physical time.
+- Vortex and Erase contain no efficacy scalar, reduction, success, restoration or controlled-below-zero claim.
+- Illusion retains only PPO execution toward a declared non-zero target with morphology consistent with retargeting; no scalar target-relative error change or positive Illusion SR transfer is claimed.
+- S7→S8 bridge asks whether a separately acquired endpoint supplies context while prohibiting continuation, validation and numerical connection.
+- The synthesis now uses the cases/PPO policies as concrete subjects; it does not attribute problem posing or efficacy to the acquisitions.
+- No `\cite`, `\ref`, `\label`, caption, figure environment or invented key is present.
+
+## Gates, fallbacks and closeout
+
+**Open/blocked:** versioned Vortex and Erase evaluators binding exact target, roles, ROI/mask, normalization, window and provenance; Illusion exact scalar binding; all panel source/provenance/crop/scale/caption/redraw/publication readiness.  
+**Fallbacks retained:** Vortex and Erase physical-variable target/controlled/physical-zero role fields only; Illusion role fields without an unbound scalar.  
+**Not used:** CFD, training, evaluator execution, field processing, artifact generation, new analysis, final-master assembly, thread recall, `mem_fs`, Working Memory or durable-memory write.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/SECTION_08_AUDIT.md

@@ -0,0 +1,54 @@
+# Section 8 Round-5 audit — Unit D
+
+**Output:** `sections/SECTION_08_STEADY_THEORY.tex`  
+**Writer agent ID:** `da7067b9-38f7-4b50-bc3b-e128f9f08a2f`  
+**Stage:** `section-pass` after first draft, bounded literature diagnosis and integrated rewrite  
+**Counting method:** established global uniform source-text count: remove comments/headings, displayed equations, inline math, citation commands and remaining TeX commands/braces; count lexical alphanumeric tokens.  
+**Uniform substantive words:** **591**; provisional S8 envelope **480–600**; illustrative midpoint **522**; section has **5/5 jobs**. No authorized article total exists, so an actual article share cannot yet be computed; the controlling provisional share remains **6%**.
+
+## Evidence cards
+
+| ID | Job and licensed claim | Comparator / definition | Internal authority and state | Metric / equation / display dependency | External diagnosis | Adjacent caveat / fallback | Status |
+|---|---|---|---|---|---|---|---|
+| S08-P01 | separately acquired steady endpoint may provide context | parabolic/no-slip periodic chain vs uniform/free-slip steady chain | context packet, global ledger, R4 S8 packet `read` | imports only ranked persistent-plus-correction organization | Mac20 full-PDF supports branch/controller localization | no continuation, amplitude match, replication or validation | pass |
+| S08-P02 | settling-qualified low-deficit endpoint at accepted action | stationary unsteady reference; reverse `s=5.2` non-amplitude-matched endpoint | steady README/RESULTS and figure manifest `read`; endpoint hashes mapped | at $x/D=10$, $\max_{y\in\mathcal C}|\boldsymbol u_{ctl}-\boldsymbol u_{in}|/U_\infty=0.02317$ over the common-fluid cross-section; two-component Euclidean norm, no quadrature weighting; exact provenance `0.0231672176752314` | Cha11b and Mac20 full-PDF support condition-first comparison and local stability limits | endpoint, not optimum/stable branch/universal law | pass |
+| S08-P03 | wall velocity is kinematic fact; field sequence motivates viscous hypothesis | accepted vs stationary/reverse source fields | wall law and current steady package `read`; no recomputation | endpoint observations only; causal links require intervention/budget tests | Cha11b and Ued03 full-PDF support direction/no-slip first and explicit validity boundaries | no amplitude-matched intervention, closed budget, force ownership or stability analysis | pass |
+| S08-P04 | prescribed-circulation Hodge/period family organizes an outer field and fails closure by no-slip incompatibility and opposite sign | $g=0$, manuscript $g_{opt}=6.148$, $g_{fit}=-12.20$; exact provenance `6.1479442395622`, `-12.200554143243473` | RESULTS, DERIVATION selections, figure README/manifest and R4 packet `read`; source payloads mapped | main Hodge/period display plus adjacent no-slip and sign tests | Crowdy, Ueda, Mittal, Chan, Sierra and Watson full-PDF support the outer-model assumptions, rotating no-slip boundary, parameter/branch dependence and viscous-matching limitation | prescribed periods are not selected by spin; not finite-Re mechanism validation | pass |
+| S08-P05 | synthesis is compatible but non-equivalent and organizational | periodic persistent-plus-correction vs separate endpoint/hypothesis/failed closure | R4 S8 imports/exports and S7 non-transfer boundary `read` | prose joins only compatible, non-equivalent observations | cross-paper pattern: preserve branch identity and dedicated proof requirements | active S9 omission branch; section ends on missing viscous tests and rejected closure | pass |
+
+## First-draft literature diagnosis and integrated fixes
+
+After the complete first draft, five close papers from the existing DynamisLab workspace were read in full-PDF state: Deng et al. (JFM 918 A4), Cornejo Maceda et al. (JFM 917 A42), Chan et al. (JFM 679), Ueda et al. (JFM 495), and Crowdy (EJMB/F 25). The integrated rewrite retained only clear contract-preserving practices: define action signs and nondimensional rate before endpoint interpretation; distinguish endpoint fields from stability or mechanism; front-load irrotationality, impermeability and prescribed-period meaning; place one-circle no-slip incompatibility and the project sign failure in the same paragraph as the outer solution; and end with dedicated missing tests. Literature did not establish any project number, source status, mechanism or artifact readiness. At that lifecycle stage no key was inserted; the citation-integration closeout below supersedes this bibliography status without changing the diagnosis.
+
+## Reader-facing sprint revision
+
+The body now introduces the prescribed-circulation family through a physical closure question: whether circulation freedom left by impermeability can explain the observed finite-Re organization. It then separates what the outer family can do (reorganize streamlines) from what it cannot do (impose rotating no slip), before using the opposite fitted and objective signs to reject closure. Figure-production, provenance, artwork and gate-management prose was removed from the section; missing mechanism evidence is stated as the intervention, diagnostic and budget tests required to isolate the proposed viscous links. The internal endpoint values remain uncited. The external outer-family and rotating-no-slip propositions now use the approved Crowdy and Ueda R5 keys recorded below.
+
+## Final high-value polish
+
+The active S9 omission branch removes the experiment bridge from S8. The final paragraph now names the physical evidence chains directly and ends with the exact viscous diagnostics still needed—amplitude-matched interventions, held-out endpoint/convergence tests, wall-vorticity flux, separation, and a closed same-control-volume momentum/mechanical-energy budget—followed by the retained no-slip/sign rejection of inviscid closure. Citations, basis normalization, rounded manuscript values and exact audit provenance are unchanged.
+
+## Equation, figure and interface audit
+
+- The accepted multi-cylinder Hodge/period display is present and defines $\boldsymbol u_N$, $\boldsymbol h_j$, impermeability, irrotationality and period normalization.
+- The one-circle rotating-no-slip incompatibility is adjacent. Before the values, $g$ is defined by $(\Gamma_1,\Gamma_2,\Gamma_3)/(U_\infty D)=g(0,1,-1)/\sqrt{2}$: a dimensionless coefficient of the unit-$L^2$ rear-opposite basis, in front/upper-rear/lower-rear order, with positive circulation geometrically counter-clockwise. It is followed immediately by $g_{opt}>0$ versus $g_{fit}<0$.
+- The outer field is called a bounded/prescribed-circulation outer reference and geometrical organization, never a potential prediction or validated finite-Re mechanism.
+- Direct observation is restricted to differing authenticated vorticity/streamline-field organization; wall shear/vorticity production, shear-layer development, transport, separation/recirculation and downstream links are explicitly hypothesis ingredients, not a claim that each diagnostic was measured.
+- Coordinator-authorized manuscript rounding gives endpoint `0.02317`, $g_{opt}=6.148$ and $g_{fit}=-12.20$. Exact provenance remains `0.0231672176752314`, `6.1479442395622` and `-12.200554143243473`; rounding changes no sign, inference or gate.
+- Current `metrics.py::profile_errors` and `runner.py::_profile_vs_qin` bind $E_{\infty,\mathrm{vector}}$ to the maximum over common-fluid cross-section points of the Euclidean norm of controlled-minus-inflow streamwise/transverse velocity, normalized by $U_\infty$. Unlike the separate crop-$L^2$ metric, this profile $L^\infty$ value has no cross-section quadrature weighting. Reader notation uses $\boldsymbol u_{ctl}$ and $\boldsymbol u_{in}$ to avoid conflict with other $q$ notation.
+- S7→S8 boundary is strict: the steady chain is separately acquired and does not inherit cross-case efficacy or periodic interpretation.
+- Under the active S9 omission branch, S8 has no experiment bridge and ends on missing viscous diagnostics plus rejected inviscid closure.
+- No momentum restoration, causal mechanism, optimum, robustness, amplitude agreement, continuation, replication, stability or cross-configuration validation claim appears.
+- Six authorized external keys are present in six `\citep` commands: `Crowdy2006MultipleCylinders`, `UedaEtAl2003RotatingCylinders`, `Mittal2001RotatingControlCylinders`, `ChanEtAl2011CounterRotatingCylinders`, `SierraEtAl2020RotatingCylinderBifurcations` and `Watson1996TwoRotatingCylinders`. No `\ref`, `\label`, caption, figure environment or invented key is present.
+
+## Gates, fallbacks and closeout
+
+**Open:** equation source recheck/typesetting; exact figure assets and provenance; crop/scale/profile encoding; caption; redraw/rendering/publication readiness; finite-image status labeling; amplitude-matched intervention; held-out endpoint/convergence qualification; wall-vorticity/separation diagnostics; closed same-control-volume momentum/mechanical-energy budget; force ownership and stability.  
+**Fallback:** endpoint/sign table plus concise outer boundary; if the composition is crowded, retain accepted CFD endpoint/profile, all three outer streamline states and the sign box as specified by the Figure Plan.  
+**Not used:** CFD, MFS generation, training, artifact generation, new numerical analysis, final-master assembly, thread recall, `mem_fs` or Working Memory. Durable memory was updated only with this approved coefficient/rounding decision after searching for an existing S8 record.
+## Citation integration closeout
+
+S08-P04 retains `Crowdy2006MultipleCylinders` where the irrotational impermeable multi-cylinder outer family is introduced and `UedaEtAl2003RotatingCylinders` where the distinct viscous rotating no-slip boundary is stated. The post-R5 integration adds `Mittal2001RotatingControlCylinders` for Reynolds-number, gap, rate and dimensionality dependence; `ChanEtAl2011CounterRotatingCylinders` for suppression-condition dependence and the boundary between potential-doublet resemblance and inviscid closure; `SierraEtAl2020RotatingCylinderBifurcations` for multiple viscous branches and bistability; and `Watson1996TwoRotatingCylinders` for low-Reynolds-number inner-Stokes/outer-Oseen matching. All six identities and keys are full-PDF verified and resolve to `../R5_REFERENCES.bib`. They support external mathematical and physical boundaries only; none verifies or explains the project endpoint, equation coefficients, fitted circulation, sign failure or finite-$Re_D=50$ mechanism.
+
+The inserted reader-facing passage adds **90 words**, within the assigned 85--125-word range. It is integrated immediately before the outer-reference closure question: the four sources establish why a deliberately restricted outer comparison is informative and why it cannot transfer parameters, identify a viscous branch or close the mechanism.
+

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/SECTION_09_AUDIT.md

@@ -0,0 +1,94 @@
+# Section 9 Round-5 Audit — Author-Field Placeholder
+
+**Project/unit:** DynamisLab JFM, Round 5, persistent Unit E, Section 9 only.  
+**Writer agent ID:** `b238cc11-c57b-481f-922d-ae7c410171c0`.  
+**Output state:** `author-field placeholder complete; substantive experiment prose blocked`.  
+**Exclusions honoured:** no CFD, training, rendering, artifact generation, new analysis, apparatus inference, trial reconstruction or experimental claim. Section 10 is now drafted separately and imports no S9 result.
+
+## Preflight and controlling authority
+
+Read the canonical Unit-E packet, all R5 controls/ledgers/registry, current R4 global ledger/README/closeout audit, current R4 Sections 9 and 10, the Section-9 Figure Plan, and current R5 terminology/production/writing rules. Repository authority confirms that Section 9 remains an author interface: only visibly marked `AUTHOR FIELD` instructions are authorized. Candidate CelerisLab paths and manifests recorded in the R4 packet were treated as mapped/path-confirmed computational context, not as proof of physical execution or as located publication assets. No raw experiment media, physical deployment logs or author apparatus records were inspected or verified in this run.
+
+Three narrow Nowledge searches were used for navigation: (1) `DynamisLab JFM Round 5 Unit E Section 9 experiment placeholder author fields decisions`; (2) `DynamisLab JFM Section 9 experiment evidence provenance media command permission gates`; and (3) `DynamisLab JFM Round 5 Unit E registry stage next resume Section 10`. The recalled Round-4 Section-9/10 and Round-5-control records agreed with current repository controls; repository files controlled all operative decisions. No thread recall, `mem_fs`, Working Memory read, durable-memory write or update was used.
+
+## Paragraph evidence cards
+
+### S09-P01 — purpose, apparatus and execution
+
+- **Job:** reserve an author-owned purpose/apparatus/open-loop-execution paragraph.
+- **Claim ceiling:** instructions only; no apparatus fact, physical execution verb or deployability claim.
+- **Comparator/definition:** author must define the physical purpose and identify what was commanded, measured, imaged and synchronized.
+- **Internal authority/state:** R4 `SECTION_09_EXPERIMENT.md` and S9 Figure Plan, `read` for architecture; all physical facts remain `author-supply/provenance-gated`.
+- **Metric/estimand:** none.
+- **External role/state:** four close papers read at `full-PDF` for reporting order only; no reader-facing citation retained.
+- **Display/equation:** apparatus schematic remains gated; no equation.
+- **Local caveat:** code capability or a computational manifest cannot establish physical execution.
+- **Fallback/author field:** mark unavailable or omit dependent statement/section.
+- **Target/actual:** provisional 75–110 placeholder tokens; actual field is instructional rather than publishable prose.
+- **Status:** placeholder complete; author content open.
+
+### S09-P02 — provenance and comparator-led morphology
+
+- **Job:** reserve exact trial/media provenance followed by one author-supplied visible morphology observation per retained pair.
+- **Claim ceiling:** single-frame shape/location morphology only after provenance; no temporal/frequency, velocity/PIV, force, field matching, registration, agreement, validation, robustness or repeatability claim.
+- **Comparator/definition:** named condition and local comparator precede observation; experiment and CFD remain separate, unregistered realizations.
+- **Internal authority/state:** R4 S9 packet and Figure Plan, `read`; author-described media are only `mapped`, exact leaves not audited.
+- **Metric/estimand:** none; a single frame contains no temporal estimator.
+- **External role/state:** bounded full-PDF diagnosis supports comparator-before-observation and estimator-specific wording only.
+- **Display/equation:** all schematic/image panels and captions remain author/provenance/permission gated.
+- **Local caveat:** every retained view needs trial/case, raw path/hash, frame/step, view/scale, release, processing, selection/repeats and permission.
+- **Fallback/author field:** drop failing rows, then schematic, then entire figure/section; never reconstruct from manifests.
+- **Target/actual:** provisional 75–110 placeholder tokens; instructional only.
+- **Status:** placeholder complete; all evidence gates open.
+
+### S09-P03 — ceiling, confounds and next threshold
+
+- **Job:** reserve local confounds, the qualitative open-loop single-frame ceiling, and the next evidentiary threshold.
+- **Claim ceiling:** instruction naming what stronger evidence would require; no promised execution or result.
+- **Comparator/definition:** claim class determines threshold: synchronized, calibrated and repeated force measurement and/or PIV acquisition, as appropriate to the claim; repeated image sequences plus a declared temporal estimator for frequency claims.
+- **Internal authority/state:** canonical packet and R4 S9 quality gate, `read`.
+- **Metric/estimand:** absent unless later author supply and reopening define one.
+- **External role/state:** full-PDF reporting diagnosis only.
+- **Display/equation:** none owned; all media gated.
+- **Local caveat:** the threshold is a gate, not a future-results programme.
+- **Fallback/author field:** omission; no S9 result export to S10 without targeted provenance audit and explicit reopening.
+- **Target/actual:** provisional 75–110 placeholder tokens; instructional only.
+- **Status:** placeholder complete; author content and result import blocked.
+
+## Bounded literature diagnosis
+
+Undermind orientation and the DynamisLab workspace were inspected before reading four existing close papers at `full-PDF` state:
+
+1. 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`: §§1–3 and methods/results pages 4–8 support purpose → functional apparatus/actuation/acquisition → baseline comparator; 30-s spectra and 100–400-frame PIV, not a single frame, support frequency/velocity statements.
+2. Khan et al., *Flow past two rotationally oscillating cylinders*, JFM 969 A16 (2023), DOI `10.1017/jfm.2023.549`: §§2–3 separate visualization, PIV, hot-film and load-cell functions and place comparator definitions before morphology; spectra, phase averages and repeated force records carry temporal/quantitative claims.
+3. Klotz et al., *Influence of porous material on the flow behind a backward-facing step: experimental study*, JFM 998 A31 (2024), DOI `10.1017/jfm.2024.639`: §§2–4 distinguish instantaneous qualitative streakline views from PIV and ensemble spectral estimates and place acquisition uncertainty locally.
+4. Zhou et al., *Artificial intelligence control of a turbulent jet*, JFM 897 A27 (2020), DOI `10.1017/jfm.2020.392`: §§2, 4–6 order facility/actuation/sensing before estimator-defined performance and support temporal/performance statements with time-averaged sensing, spectra and repeated/spatial data rather than a single image.
+
+The sample-bounded diagnosis improved placeholder architecture only: apparatus fields are grouped by evidentiary function; acquisition/provenance precedes morphology; comparator precedes observation; captions own condition/acquisition/processing metadata; and stronger claim classes require named estimators. Literature supplied no DynamisLab fact, artifact, observation, permission, citation key or readiness decision.
+
+## Static content and gate audit
+
+- [x] Exactly three stable paragraph jobs are represented.
+- [x] Every paragraph begins with visible `AUTHOR FIELD` marking.
+- [x] TeX contains instructions, not manuscript factual assertions.
+- [x] No apparatus dimensions/facts, execution details, trial/case identities, command provenance, media provenance, morphology observation or result was filled.
+- [x] Single-frame prohibition explicitly covers temporal/frequency, velocity, force, field matching, agreement and validation.
+- [x] No fabricated `\cite`, `\ref`, `\label`, bibliography key, equation, value, figure call or final-master integration.
+- [x] S10 was drafted separately with no S9 observation, result or transition.
+- [x] Omission remains the strict fallback.
+
+## Current safe-assembly omission branch
+
+For the reader-facing sprint, the strict omission fallback is active: `sections/SECTION_09_EXPERIMENT_PLACEHOLDER.tex` remains an internal author interface and is excluded from the reader-facing master. Inclusion requires author-supplied content, targeted provenance audit and explicit reopening. Until then, no S9 observation or result enters S10. This assembly decision does not delete or alter the placeholder and closes none of its author, media, command, provenance or permission gates.
+
+## Counts and share
+
+Counting method: whitespace-delimited words in TeX after stripping TeX commands and comments, reported as control/placeholder text because none is admitted publishable prose. The static check records **0 substantive manuscript words** and **275 placeholder/control tokens** in the live ledger. Section 9 retains the provisional **3%** controlling share and illustrative 240–300-word range, but actual article share is not computable: no authorized article target exists and no other R5 actual total is complete. Unused S9 allowance remains contingency under the live-ledger rule.
+
+## Open author gates
+
+Purpose and final wording; facility/test-section/background state; geometry and boundaries; towing/carriage and imaging path; physical software entry point; command source, units, body order/sign/conversion and timing; commanded-versus-measured motion; synchronization; trial/case IDs, dates/setup version, conditions, repeats and selection rules; raw media paths/hashes; camera/view/FOV/scale/exposure/frame identity; dye/numerical release and processing; exact CFD counterpart mapping; apparatus schematic source/editable version; local confounds/calibration; permissions; exact morphology wording; caption approval; targeted provenance audit; coordinator reopening; any S10 import. All remain open. Fallback is omission of dependent media/observation or the entire section.
+
+## Disposition and next resume goal
+
+**Disposition:** Section-9 placeholder contract remains complete and statically auditable, but Section 9 remains `author-interface`, not manuscript prose or experimental evidence; the placeholder is omitted from the current reader-facing master. **Next resume goal:** retain omission through the S1/S10/global interface audit unless author supply, targeted provenance audit and explicit reopening authorize inclusion.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/SECTION_10_AUDIT.md

@@ -0,0 +1,121 @@
+# Section 10 Round-5 Audit — Discussion and Conclusions
+
+**Project/unit:** DynamisLab JFM, Round 5, persistent Unit E, Section 10.  
+**Writer agent ID:** `b238cc11-c57b-481f-922d-ae7c410171c0`.  
+**Output:** `sections/SECTION_10_DISCUSSION_CONCLUSION.tex`.  
+**Stage:** final reader polish complete; pending S1/S10/global interface audit.
+
+## Scope and authority
+
+The draft uses the frozen R4 Section-10 packet and current retained S2–S8 TeX/audits. S1 is now available for repetition/interface diagnosis, while retained S2–S8 continue to control scientific imports. Its contribution-ladder alignment remains deferred to the named final wave audit. The current safe assembly adopts the S9 omission branch: the author-field placeholder is excluded from the reader-facing master unless author supply, targeted provenance audit and explicit reopening occur. S10 imports no experiment observation, result or transition. The output is text-only and contains exactly three Discussion paragraphs followed by two short Conclusion paragraphs.
+
+Three narrow Nowledge searches were used for navigation in the drafting run: (1) `DynamisLab JFM Round 5 Unit E Section 10 discussion conclusion evidence order decisions`; (2) `DynamisLab JFM Section 10 no experiment Erase limitations citations values claim ceiling`; and (3) `DynamisLab JFM Round 5 S1 S10 global interface audit Unit E resume`. Recalled S9/S10, Unit-A interface and R5-control records agreed with current repository controls; repository TeX, audits and the R4 S10 packet controlled every imported meaning. No thread recall, `mem_fs`, Working Memory read or durable-memory write was used. This recovery performed no new research.
+
+## One-to-one paragraph evidence cards
+
+### S10-D01 — learned rotation and target-relative field
+
+- **Job:** distinguish repeated policy discovery from the direct target-relative Kármán velocity-field comparison.
+- **Claim ceiling:** repeated policy discovery in one configuration and target-relative field improvement in another; no pooled efficacy, robustness or configuration continuation.
+- **Comparator/definition:** repeated Kármán training realizations in the uniform/free-slip case; target, controlled and physical-zero registered fields in the parabolic/no-slip case.
+- **Internal authority/state:** current S2 definitions, S3 progression, S4 retained results/audits, and S1 for repetition diagnosis, `read`; final S1 interface deferred.
+- **Measurement meaning:** complete-cycle mean and phase-resolved velocity-field comparisons are retained without values; training outcome is stated separately.
+- **External citation role/state:** four close final sections read at `full-PDF` for hierarchy/compression only; no reader-facing citation.
+- **Display/equation dependency:** none.
+- **Local caveat:** the two configurations are not treated as one continuous data set.
+- **Fallback:** omit subordinate learning detail and retain the deep comparator ordering.
+- **Actual words/status:** 117; final reader polish integrated.
+
+### S10-D02 — rear rotation, mean deficit and centred residual
+
+- **Job:** relate persistent rear-cylinder counter-rotation to the mean streamwise deficit, constant-control field and centred velocity residual without identifying those objects.
+- **Claim ceiling:** persistent rear-cylinder rotation is an organizing component of the tested periodic control; no unique decomposition or mechanism.
+- **Comparator/definition:** persistent rear term versus tested modulations; passive mean deficit versus controlled-minus-passive field; zero-to-constant and constant-to-DRL scalar increments; separately centred, non-paired velocity residual.
+- **Internal authority/state:** current S5/S6 TeX and audits, `read`; their distinct source and comparison definitions are retained.
+- **Measurement meaning:** no value replay; the constant-to-DRL remainder belongs only to the scalar mean-field error increment, while the centred residual has no licensed magnitude ranking.
+- **External citation role/state:** full-PDF rhetoric diagnosis supports outcome-before-interpretation and adjacent boundaries only.
+- **Display/equation dependency:** none.
+- **Local caveat:** the residual is not called smaller and does not close the scalar mean-field accounting; the single non-equivalence boundary blocks a unique decomposition or mechanism.
+- **Fallback:** retain the rear-rotation, constant-control and centred-residual distinction; delete modal terminology first if further compression is required.
+- **Actual words/status:** 144; final reader polish integrated.
+
+### S10-D03 — changed wake definitions and steady context
+
+- **Job:** state what the Vortex and Illusion role fields show, then place persistent rear rotation beside the separate steady endpoint and failed inviscid closure.
+- **Claim ceiling:** event-relative Vortex execution/morphology and policy execution toward the Illusion target with morphology-consistent role fields; no scalar improvement, transferred Kármán law or mechanism.
+- **Comparator/definition:** event-relative Vortex target/controlled/physical-zero fields; non-zero Illusion target and role fields; prescribed rear rotation and steady low-deficit endpoint; outer circulation versus wall-spin closure.
+- **Internal authority/state:** current S7/S8 TeX and audits, `read`.
+- **Measurement meaning:** no Vortex or Illusion scalar is imported; only role-field morphology, endpoint context and the sign/closure failure are used.
+- **External citation role/state:** full-PDF rhetoric diagnosis supports local caveats and finishing Discussion on established physical context.
+- **Display/equation dependency:** none; S8 equations/figures are not repeated or called out.
+- **Local caveat:** persistent rotation is a useful kinematic coordinate, not a transferred control law or causal explanation; no S9 import.
+- **Fallback:** retain changed-task non-transfer and failed-closure boundary in compressed form.
+- **Actual words/status:** 136; final reader polish integrated; S9 omitted from safe assembly.
+
+### S10-C01 — strongest velocity-field answer
+
+- **Job:** state the strongest velocity-field result and locate Vortex/Illusion at their bounded execution/morphology level.
+- **Claim ceiling:** local Kármán target-versus-physical-zero registered-field result; Vortex event-relative execution and Illusion morphology without common scalar efficacy.
+- **Comparator/definition:** each learned task remains under its own definition; deep Kármán comparison retains target, controlled and physical-zero roles.
+- **Internal authority/state:** current S2–S8 retained meanings, especially S3/S4, `read`.
+- **Measurement meaning:** no value replay; registered velocity-field ordering only for Kármán.
+- **External citation role/state:** full-PDF diagnosis supports a shorter closing unit that adds no evidence.
+- **Display/equation dependency:** none.
+- **Local caveat:** Vortex and Illusion supply no common efficacy measure.
+- **Fallback:** omit the event-relative clause if the global interface requires further compression.
+- **Actual words/status:** 64; final reader polish integrated.
+
+### S10-C02 — physical interpretation and declared-reference close
+
+- **Job:** close on persistent rear-cylinder rotation, constant-control mean field and separately described centred velocity residual.
+- **Claim ceiling:** bounded physical organization; no universality, transfer or unrestricted law.
+- **Comparator/definition:** dominant tested surrogate term, constant-control mean field and centred velocity residual.
+- **Internal authority/state:** current S5/S6 meanings and frozen R4 S10 close, `read`.
+- **Measurement meaning:** action ranking, scalar mean-field increments and residual modes answer different questions.
+- **External citation role/state:** full-PDF diagnosis supports conceptual rather than value-bearing final sentence.
+- **Display/equation dependency:** none.
+- **Local caveat:** the final sentence retains the declared-reference ceiling.
+- **Fallback:** retain the declared-reference definition as the final sentence.
+- **Actual words/status:** 74; final reader polish integrated.
+
+## Bounded full-PDF diagnosis and integrated fix
+
+After the complete first draft, four existing papers in the DynamisLab workspace were resolved and read at `full-PDF` state for final-section rhetoric only: Deng et al., *Low-order model for successive bifurcations of the fluidic pinball*, DOI `10.1017/jfm.2019.959`; Cornejo Maceda et al., *Stabilization of the fluidic pinball with gradient-enriched machine learning control*, DOI `10.1017/jfm.2021.301`; Li & Zhang, *Reinforcement-learning-based control of confined cylinder wakes with stability analyses*, DOI `10.1017/jfm.2021.1045`; and Rabault et al., *Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control*, DOI `10.1017/jfm.2019.62`.
+
+The sample-bounded diagnosis supported functional paragraph separation, outcome before interpretation, present tense for established meanings, immediate caveat placement, compression without equations or values, and a shorter conceptual close. The first draft already followed these controls. One clear fix was integrated: D03 now describes the low-deficit endpoint as occurring **under** prescribed wall motion rather than saying the wall motion `produces` it, avoiding an unintended causal verb. Literature supplied no DynamisLab fact, claim, value, evidence, citation, display or readiness decision.
+
+## Reader-facing sprint correction
+
+The sprint reread S1, S5 and S6 solely to diagnose repetition and the residual/scalar collision; no new research or authority search was performed. The previous D02 phrase `the smaller residual` was incorrect because the reported 6.58% is the constant-to-DRL **scalar mean-field increment**, while the residual is separately centred and non-paired and has no licensed smaller-magnitude comparison. The revised D02 now separates those objects explicitly and adds the bounded interpretation that their compatibility identifies a recurring organizational coordinate without making surrogate terms, scalar increments and field modes identical. D01 and both Conclusion paragraphs were compressed to avoid replaying S1's ladder and S5/S6's local detail. No citation was added: S10's frozen contract prohibits new reader-facing citations, and the existing full-PDF identities are rhetoric records rather than repository manuscript keys.
+
+The current safe assembly also takes the S9 omission fallback. `SECTION_09_EXPERIMENT_PLACEHOLDER.tex` remains an internal author interface and is not included in the reader-facing master unless the author/provenance/reopening gate later passes. S10 therefore contains no experimental import.
+
+## Final reader polish
+
+The final pass replaced internal audit diction in the TeX with direct fluid-mechanical subjects: rear-cylinder counter-rotation, passive mean streamwise deficit, constant-control velocity field and separately centred velocity residual. Illusion is now limited to policy execution toward a declared non-zero target with morphology-consistent role fields; no scalar improvement or `establishes approach` wording remains. Repeated caveats were compressed into one action/field non-equivalence boundary in D02 and local scalar limits for Vortex/Illusion. No new research, fact, value, citation, equation or display was added.
+
+## Contract, content and TeX audit
+
+- [x] Exactly three Discussion and two Conclusion paragraphs, mapped one-to-one above.
+- [x] Paragraph order is Kármán learning/velocity field → rear rotation/mean deficit/centred residual → Vortex/Illusion/steady context → strongest velocity-field answer → bounded physical close.
+- [x] Only established S2–S8 meanings are imported; S1 was used only for repetition/interface diagnosis and remains for wave audit.
+- [x] No S9 observation, experiment result or experiment transition; the S9 placeholder is omitted from the current reader-facing master.
+- [x] No unfinished historical route, Erase result, gap inventory, limitations programme, future work, decisive test or third conclusion paragraph.
+- [x] No new evidence, result value, equation, display, table, figure/panel call, citation, `\cite`, `\ref`, `\label` or bibliography key.
+- [x] The single D02 non-equivalence boundary keeps action ranking, scalar mean-field increments and centred residual modes distinct.
+- [x] `Kármán` terminology and `action-correlated correction-field modes` are retained; internal chain labels are absent.
+- [x] D02 states that 6.58% belongs only to the scalar mean-field increment; the separately centred, non-paired residual is not called smaller or used to close that ledger.
+- [x] Illusion uses only policy-executed-toward-target and morphology-consistent wording; no scalar improvement or established approach is claimed.
+
+## Counts, share and gates
+
+Counting method: remove headings and TeX commands/braces, then count alphanumeric reader-facing prose tokens; there are no displayed equations or comments. Paragraph counts are D01 **117**, D02 **144**, D03 **136**, C01 **64**, C02 **74**, totalling **535 publishable prose words**. This lies within the live ledger's provisional 480–600 range and keeps the Conclusion shorter than the Discussion. The controlling S10 allocation remains **6%**. Actual article share and deviation are not computable until S1 and the complete article body are fixed; no prose was added to consume an illustrative allocation.
+
+S10 has no local artifact or bibliography dependency. Open interfaces remain: S1 contribution-ladder alignment, global duplication/terminology, article share/display burden, and confirmation that all imported S2–S8 meanings survive global edits. All S9 author/provenance/result-import gates remain open and unchanged; pending reopening, omission from the reader-facing master is the active assembly fallback.
+
+## Closeout and next goal
+
+Repository evidence was `read` at the current TeX/audit level; underlying artifacts remain owned by S2–S8 and were not re-audited. Literature was `full-PDF` for rhetoric only. No scripts beyond static counting/content checks, no CFD, training, artifact generation, evaluator run, new scientific analysis, master assembly or external write was performed.
+
+**Disposition:** S10 final reader polish is complete; the text-only synthesis is ready for the named interface audit, not yet a global or final pass.  
+**Next goal:** run the S1/S10/global interface audit after the retained S1 draft is available, checking the contribution ladder, established-result imports, duplication, terminology, total article share and display/equation burden without reopening S9 or artifact gates.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/WAVE_S02_S03_INTERFACE_AUDIT.md

@@ -0,0 +1,35 @@
+# Wave interface audit — Sections 2–3
+
+**State:** completed and coordinator-approved on 2026-08-11  
+**Scope:** current `sections/SECTION_02_PROBLEM_METHOD_EVALUATION.tex` against current `sections/SECTION_03_CAPABILITY_PROGRESSION.tex`  
+**Ownership:** shared interface sidecar written by Unit A; no edit to the Unit-B Section-3 audit  
+**Disposition:** Unit A `S2–S3-interface-pass` only; no section pass, Unit-B stage change, global pass or artifact approval
+
+## Interface findings
+
+| Dimension | Current S2 export | Current S3 import/use | Finding and retained boundary | Status |
+|---|---|---|---|---|
+| Configuration and roles | Exact `parabolic-inflow/no-slip-wall channel configuration` and `uniform-inflow/free-slip-wall channel configuration`; target/reference, controlled, physical zero/uncontrolled and optional constant are distinct | S3 assigns steady and Illusion to uniform/free-slip and periodic Kármán to parabolic/no-slip; Kármán and preview paragraphs name target, controlled and physical-zero roles | Configuration ownership is explicit and no evidence is transferred. S3 opening `passive` is descriptive shorthand localized by exact role definitions in the case paragraphs; it does not create a pooled role. | PASS |
+| Metric semantics | DTW is complementary signal-level evidence; complete-cycle mean, eight-slot phase, event-relative, mean-only and steady endpoint/profile criteria are non-equivalent | S3 uses endpoint profiles for steady, separate mean and phase-8 for Kármán, literal event offset for Vortex and literal phase/role payloads for Illusion | No common efficacy score is formed. S3's rounded Kármán preview is attached separately to each named field estimand; Vortex has no scalar and steady is not called a periodic mean. | PASS WITH DOWNSTREAM GATES |
+| Reynolds convention | $Re_D=U_{ref}D/\nu$; uniform inlet speed for uniform inflow and bulk/mean inlet speed for parabolic inflow; inherited parabolic conversion gated | S3 prints no bare Reynolds label or inherited value | S3 does not bypass the unresolved historical $U_0$ centreline-versus-bulk/mean binding. | PASS |
+| Preview/detail ownership | S2 owns reusable definitions, roles, ROI/metric semantics and qualification ceiling | S3 owns capability viewing/progression and rounded/source-bound previews; S4 owns exact Kármán values and windows; S7/S8 own event and steady detail | S3 does not redefine S2 equations or narrate S4/S7/S8 detail. Preview prose remains subordinate to exact downstream owners. | PASS |
+| Bridge | S2 closes by licensing named, non-pooled case comparisons | S3 opens with four different declared-reference comparisons and closes by selecting Kármán for depth | The transition is continuous: definitions precede viewing, then S3 hands detailed Kármán evaluation to S4. No roadmap or performance conclusion is duplicated. | PASS |
+| Duplication | Configuration names, roles and metric distinctions are defined once for reuse | S3 repeats only the minimum local configuration/role/estimand labels needed to decode each comparison | Necessary local decoding is not duplicate method exposition. S3 contains no S2 solver, action-map, PPO-setting, ROI-equation, DTW-algorithm or qualification replay. | PASS |
+| Displays and caption burden | S2 method schematics/tables remain production-gated with prose fallbacks | S3 accepted two-figure choice is closed only at author-choice level; exact panels, source/provenance, crops, scales, captions and artwork remain open | Neither section relies on a rendered or approved artifact. Missing displays invoke recorded fallbacks and do not justify compensatory prose. | PASS WITH ARTIFACT GATES |
+| Claim ceiling | S2 defines method/evaluation and selected-observable qualification without performance | S3 previews bounded capability without cross-configuration equivalence, pooled efficacy, Vortex scalar, positive Illusion SR, steady mechanism or transfer | No interface sentence raises either section's frozen claim ceiling. | PASS |
+
+## Combined word balance
+
+The established static counting methods report S2 at **1,212 words** and S3 at **637 words**, for a combined S2–S3 total of **1,849 words**. Within the completed interface, S2 contributes **65.5%** and S3 **34.5%**. Against the non-authoritative 8,700-word planning denominator, the sections would contribute 13.9% and 7.3%, or 21.3% combined; these are diagnostic values, not actual article shares.
+
+The illustrative midpoint is 1,392 words for S2 plus 696 for S3, or 2,088 combined. The current interface is 239 words lower (−11.4%). The difference is accepted at this wave boundary: S2 remains within its 1,000–1,250 drafting envelope, while S3 is below its initial 800–1,000 envelope because production-gated panel/caption material is not narrated in prose. No padding, gated-display narration or budget transfer is authorized by this interface pass.
+
+## Gates retained
+
+- **S2:** exact control-run rows; selected policy/same-selection normalizer/target metadata; inherited parabolic Reynolds conversion; geometry and DRL schematics; configuration, metric and CFD tables; bibliography keys; production readiness.
+- **S3:** exact steady and composite panel source/provenance; Vortex offset/role leaves/evaluator; Illusion phase and physical-zero role; hashes, units, scales, crops, captions, artwork and rendering.
+- **Cross-interface:** Figure-1 decoding and every local caption must renew exact configuration, role and estimand semantics. No artifact is approved by prose or this audit.
+
+## Final disposition
+
+**PASS for the S2–S3 interface on the Unit-A side.** The current sections are consistent in configuration/role naming, metric semantics, Reynolds policy, preview/detail ownership, bridge and duplication control. Unit A may record `S2–S3-interface-pass`; S2 remains short of `section-pass`, Unit B remains at its independently recorded stage, and no global or artifact gate is closed.

docs/JFM_WYQ/ROUND5_BOTTOM_UP/audits/WAVE_S04_S06_INTERFACE_AUDIT.md

@@ -0,0 +1,29 @@
+# S4–S6 interface and protected-centre wave audit
+
+**Date:** 2026-08-11  
+**Disposition:** **CONDITIONAL PASS** for the approved narrow findings. This audit does not grant section-pass or artifact readiness. Unit B retains S4 prose ownership; approved evidence/data reconciliation was integrated by its persistent writer.
+
+## Length and protected allocation
+
+Under the uniform Pass-3 method after targeted Unit-B reconciliation, the current total is **2,091 words** — S4 **851**, S5 **612**, S6 **628** — or **33.99%** of the 6,152-word substantive body. Earlier method-local snapshots are superseded for global comparison. The combined **44% allocation is reserved**, with an ordinary **42% lower protection floor**. Actual article share remains unresolved until the complete article. The short current count is not a license to pad prose or narrate unavailable displays.
+
+## Interface findings
+
+- **Acquisitions and estimands:** S4 PPO mean/phase L2, S5 standardized SR L2/native DTW, S5 40-step deletion, and S6 corrected four-role mean/residual/CCD objects remain separate. S4 and S5 reuse the same exact path-bound target and physical-zero field artifacts; PPO-controlled and SR-controlled field acquisitions have distinct paths and hashes. Coincident physical-zero values therefore reflect a shared comparator artifact, not independent agreement. No pooling, replication, validation or triangulation is claimed.
+- **S4 source units:** current wake-L2 authority binds S4 field errors to historical $U_0$ normalization and the retained trajectory window to $tU_0/D$. Article-wide $U_{ref}$ conversion remains unbound; dimensionless comparator-relative reductions are unchanged.
+- **S5 standardized DTW:** role-local `dtw_summary.json` and `metadata.json`, together with current SR pipeline authority, bind the rolling native value to historical window 30. The historical lag aligns the target and the six channel scores are averaged. S5 states this compactly and separates it from the 40-step shared-lag parent-relative deletion contract. Fitted lag is comparison metadata, not physical delay.
+- **S5 deletion:** remains parent-relative and noncausal; changed trajectories preclude necessity, sufficiency, uniqueness or mechanism claims.
+- **S6 mean windows:** targeted gate resolution found an unambiguous common role-local half-open slice `[480,840)` (360 retained boundaries) in all four current source `campaign.json` manifests. The current ROI summary binds those parent paths/hashes, `dynamic_increment.py::_dense_statistics` applies `start:stop` literally, and the authority catalog/checkpoint binds ROI manifest SHA256 `f7f6141042678e52314a4d8f76d589108da8e844c8ce504f8f523a6f88a17e73`. The dependent gate is **CLOSED** without changing values.
+- **S6 action coordinates:** current named CCD derivation, transformation code and strict schema bind same-boundary effective actions in counter-clockwise-positive native front/upper/lower order and native solver angular-velocity command units. The displayed orthonormal transformation is front $a_F$, rear-symmetric $(a_U+a_L)/\sqrt{2}$, rear-antisymmetric $(a_U-a_L)/\sqrt{2}$. The exact retained physical action mean is removed, then samples are empirically centred; no standardization or whitening is applied. This definition now appears beside first CCD use.
+- **S10-D02 handoff:** Unit E should replace `while the smaller residual is organized descriptively through action coordinates as` with `while the separately centred residual associated with the smaller constant-to-DRL scalar increment is organized descriptively through action coordinates as`; retain the fixed modal term and existing non-equivalence/noncausality. S10 was not edited by Unit C.
+- **POD boundary:** the same-object, same-rank, same-weight/domain “essentially the same tested rank-3 field subspace” statement remains licensed by current CCD claims/results/writing authorities; it grants action-coordinate association only, not superiority, uniqueness or causality.
+
+## Deferred production controls
+
+- Repository bibliography keys for S4–S6 candidates remain **OPEN**; no citation key was invented.
+- Display labels, callouts, caption wording, hashes, crop/scale, action extraction, bound S6 retained-window provenance, artwork and publication readiness remain deferred to the owning S4/S5/S6 audits.
+- Missing displays invoke recorded fallbacks; they do not justify prose expansion.
+
+## Conditional close
+
+Wave logic, ownership, estimand separation and protected allocation pass. Conditions retained: all owning display gates; S5 historical velocity/action extraction and PPO target reconciliation; S6 composite provenance; bibliography keys. Next action is coordinator review or a source-only gate closure, not prose padding.