Reconcile S2 execution differences with Claude v0.2 rulings and remaining definitions

This commit is contained in:
Frank14f committed 2026-11-14 17:25:56 +08:00
1 parent c35b02d9e3
commit 33b18e0831
5 files changed
+212 -163

No files matched your search

@@ -0,0 +1,93 @@
# S2 design v0.2: rulings on MeltAstra's execution diff (Claude, 2026-11-14)
Integrates MeltAstra's `docs/s2-execution-diff-v0-2026-11-14.md` (codex/s2-execution-diff-20261114 @ c35b02d9). The diff is adopted except where a ruling below says otherwise.
Prior versions (v0 d8aa0f42, v0.1 fd1db63e, v0.1a 4e435a03) stand unless amended here.
No computation or training starts without Yanqi's explicit go.
## R1. Target scale (D1)
The log scale τ is fitted separately for each (outer fold, budget pool), on that learning pool only. A larger pool never supplies the scale for a smaller one.
## R2. Splits (D2)
- **Components:** source key ∪ confirmed near-duplicate evidence edges, following MoldAstra note14.
- The rule must be frozen before the candidate list is regenerated.
- Unresolved pairs that could change the grouping put S2 on HOLD.
- **Outer folds:** 5, over components, with a fixed label-free ordering and seed 2026111409.
- **Inner validation:** a fixed 20 % of each outer fold's non-test components (adopted). The 25/50/100 % pools are nested prefixes of the remaining learning pool. Checkpoints and HPO see inner validation only.
- **development_exposed:** the S1 panel and the previously diagnosed parts are flagged. The main table keeps the pre-registered 5 folds. Exposed and unexposed subsets are reported as a secondary breakdown. There is no separate untouched set.
## R3. Target domain and eligibility (D3)
- **Primary domain:** commonly eligible nodes in the d > 0 PPS draw. Statements are restricted to this domain and never called whole-part results.
- **S2 common eligibility:**
- d > 0;
- O2 spectrum valid;
- N_eff ≥ 64;
- edge block complete and finite (needed by arm E, and therefore applied to every arm).
The S1 gate-direction requirement is **dropped**: gate q is not an S2 input. This is recorded as a difference from S1.
- **Draws:** the frozen S1 draws are reused for the 24 panel parts. New parts get the same rule, with seed [2026111400, prefix]. The 2,000 queries are fixed per part and not refreshed per epoch.
- **Per-part error (adopted):** e_p = sqrt(Σ c E (T̂−T)² / Σ c E T²). No second mass factor. A zero denominator is NA with a reason.
- **Aggregation:** parts are equally weighted within a component, components equally weighted overall. Source-key weighting is descriptive only.
- **Coverage table per part:** h, draws, distinct IDs, K, K/2000, h·K/2000, the near-wall share, 1−h, and rejection reasons.
- "Last-filled region" and Kendall τ are defined only over the query coverage, and are labelled that way.
## R4. Connectivity bottleneck metric (D8): deferred
It needs predictions on the full tet graph, plus full-node O2 for arm E, and neither is in S2's budget. No substitutes are allowed: no ground-truth infill, no query-induced graphs, no zero-filled O2. It moves to a later full-field evaluation stage. The thin slice follows v0.1a.
## R5. Baselines and backbone (D4/D5)
- **R2 = "R2-G, Gre24-style approximation".**
- Features, as listed in the diff §5.1:
- (x−x̄)/R_g;
- d/t_part;
- ‖x−gate‖/R_g;
- ℓ_g/R_g, where ℓ_g is the tet-edge Dijkstra distance from the gate's nearest node;
- R_g, V, t_part.
- This is **not** called a Gre24 reproduction.
- Grok gets one targeted task: Gre24's exact feature list and its eikonal setup. If a feature can be matched cheaply, it is added before the freeze; R2-E (eikonal) is otherwise deferred.
- **GBRT:** sklearn HistGradientBoostingRegressor with sample_weight. Weights make components equal, parts equal within a component, and use draw counts within a part.
- A fixed HPO grid of at most 12 configurations, selected on inner validation.
- With a fixed random_state it is deterministic, so it is counted as **15 independent fits** (5 folds x 3 pools), not 45.
- **N/E backbone:** primary candidate is Transolver (thuml, MIT) as the whole-part encoder on all tet nodes with the non-O2 base inputs. Backup is GINO (neuraloperator, MIT).
- The query decoder is an MLP over [node embedding, g12, g23] for N and [..., 4 edge q] for E.
- Query nodes are mesh nodes, so no interpolation is needed.
- The licence and the exact commit are checked before the pilot.
- Two capacity tiers are pre-registered (width 128 and width 256, 4 layers), chosen on inner validation. The same tier is used for N and E.
## R6. Statistics (D6)
- **Primary contrast (single):** Δ = 1 − Ē_E / Ē_N.
- Measured at the 100 % pool, whole primary domain.
- Ē is the component-equal error averaged over 3 matched seeds.
- m = 10 %, now frozen.
- The six-region table applies.
- **Interval:** bootstrap over components, with matched seed blocks kept intact.
- 2,000 replicates, seed 2026111408, two-sided 95 % percentile.
- Both arms are recomputed on the same indices before the ratio is taken.
- Stated limitation: overlapping training sets and fixed folds are not resampled.
- **Secondary** (reported, no gate, no multiplicity claim): the 25 % and 50 % pools, the near-wall slice, the exposed/unexposed split, Kendall τ.
- **R2 usefulness:** report Δ_N/R2 and Δ_E/R2 with the same machinery. The neural route is called useful only if N's lower bound against R2 is > 0.
- The query-level Monte Carlo error is not merged into the model-generalisation interval.
## R7. Resources (D7)
- **Preprocessing:** before any corpus-wide run, MeltAstra measures a 10-part stratified sample. It covers decode, d/t_part, ℓ_g, the draw, the O2 count and the O2 kernel, with CPU time, RSS and disk recorded for each part. The full-corpus table then lists cases above the limits. Limits stay at 30 min / 1.25 GiB per part; no part is swapped.
- **Pilot:**
- one outer fold, the 100 % pool, one seed, N and E, both capacity tiers;
- at most 4 runs, at most 2 GPU-h in total, one V100 16 GB, FP16 AMP with FP32 loss and normalisation;
- test fold sealed.
- Exceeding a limit means stop and report; nothing is shrunk silently.
- **Full matrix accounting:** 90 neural runs + 15 R2 fits + HPO + pilot + reruns, all counted. GPU-h is computed only from pilot measurements.
## R8. Extensions (not in S2 v0.2)
- **N0 (an arm without O2):** conditional. It is added only if Yanqi wants the value of the whole O2 block tested; that costs 45 more neural runs.
- **O1:** stays open, next design item. Order: type definition, then calibration on 890 plus one uncached part, region tables, capacity LP, then acceptance.
- **In-house CAE pair:** prepare the data-request and control table (constant Δp and fixed flow) only; no solving yet.
## Entry sequence (unchanged)
1. Grouping rule and components.
2. Domain and coverage.
3. Features and backbone with licences.
4. Statistics frozen (this document).
5. 10-part preprocessing measurement.
6. Pilot receipts.
7. Full-matrix budget.
8. Stage review, then Yanqi's go.
@@ -4,7 +4,8 @@
"design_baselines": [
"d8aa0f420647701709e03629326dce8339db739a",
"fd1db63ee1c4fb4415a36ed95a2eedfb73a5c85c",
"4e435a03d3e1023bb2c12154ab4afc98b37da741"
"4e435a03d3e1023bb2c12154ab4afc98b37da741",
"72deb2664411d0256895108b7ab108e2562ebf0b"
],
"files": [
{
@@ -12,20 +13,20 @@
"bytes": 1357,
"sha256": "4093a77e04cff01d422ba31acb57cd40acca480e8035de0fa52eea02d9fb458d"
},
{
"path": "docs/sharing-authorization-2026-11-14.md",
"bytes": 766,
"sha256": "88d5c4f84f985194c7c4c96071d57e3bf8ed44f601d86af66f5573c15361c11d"
},
{
"path": "docs/collaboration-stage-review-policy.md",
"bytes": 2297,
"sha256": "4c98e9f8593e48911269b38544d5155c443efb17bb2abf14ce15bf78f989fbe7"
"bytes": 2321,
"sha256": "7757dc15066cda6e74a3b6aed559c6cf42a3a6e982e711456e63d71be41522ce"
},
{
"path": "docs/s2-execution-diff-v0-2026-11-14.md",
"bytes": 24321,
"sha256": "e232cae30a7254bb5ec509d2a2209edc78aacdb9ae7c3b33d951c6509e67de15"
"bytes": 12048,
"sha256": "91219858ffe7fedb71e371e9543dd85914b3a87adaf53702a89a703fc7d0aae6"
},
{
"path": "docs/sharing-authorization-2026-11-14.md",
"bytes": 766,
"sha256": "88d5c4f84f985194c7c4c96071d57e3bf8ed44f601d86af66f5573c15361c11d"
},
{
"path": "references/2026-11-14/s2-design/claude-data-contract-v0.md",
@@ -62,6 +63,16 @@
"bytes": 5481,
"sha256": "754cf22d3cd4f595d89bf7044197b17764260750ecf30fe5f852c1f4523e8cce"
},
{
"path": "references/2026-11-14/s2-design/claude-s2-v01a-4e435a03.md",
"bytes": 1490,
"sha256": "d60cf13001e73f9a7a662c1446623bcaae0bd000e1cf0572204322127eeb8e9c"
},
{
"path": "references/2026-11-14/s2-design/claude-s2-v02-72deb266.md",
"bytes": 6892,
"sha256": "bd59621b1507a12e24bf3221cb4481611ec1695c69d7f79c071996d3e175c0bb"
},
{
"path": "references/2026-11-14/s2-design/claude-stage-review-3ae89805.md",
"bytes": 4592,
@@ -98,9 +109,10 @@
"sha256": "7a1a9dca072633388ae4f6e759063fe5e8952eba8e2d2594726f8c806f60c209"
},
{
"path": "references/2026-11-14/s2-design/claude-s2-v01a-4e435a03.md",
"bytes": 1490,
"sha256": "d60cf13001e73f9a7a662c1446623bcaae0bd000e1cf0572204322127eeb8e9c"
"path": "references/2026-11-14/s2-design/v02-main-review-receipt.txt",
"bytes": 3699,
"sha256": "75de33048a55ab53967d27c800374e6bcba3eb67480eccef16a8126c1d176b63"
}
]
],
"preceding_execution_diff": "c35b02d9e3b6814447d00c9b8950e29237dbb127"
}
@@ -0,0 +1,51 @@
(2 total) ===
--- Reply 1 of 2 ---
From: ClaudeWorker (local) <botuser-T0C8JMA61UY-B0C7YUHNDFC@slack-bots.com> (U0C7SSQLBJ7)
Time: 2026-11-14 17:18:43 CST
Message TS: 1794647923.167489
Accepted; working on: “@⁠TRON MeltAstra 已完成执行差分,接续统一 v0 d8aa0f42+v0.1 fd1db63e+刚发布的 v0.1a 4e4…”
--- Reply 2 of 2 ---
From: ClaudeWorker (local) <botuser-T0C8JMA61UY-B0C7YUHNDFC@slack-bots.com> (U0C7SSQLBJ7)
Time: 2026-11-14 17:20:04 CST
Message TS: 1794648004.314569
*Claude: S2 design v0.2. I read MeltAstra's full execution diff (`c35b02d9`) and have decided all seven open items.*
<https://gitea.frank14f.online/Frank14f/MeltMind/src/commit/72deb2664411d0256895108b7ab108e2562ebf0b/docs/claude/s2-design-v0.2-2026-11-14.md|s2-design-v0.2> (`72deb266`).
1. *Target scale:* τ is fitted separately for each (outer fold, training pool), on that pool's labels only.
2. *Splits:*
• groups = source key ∪ confirmed near-duplicate links; any unresolved pair puts S2 on hold;
• 5 outer folds, seed 2026111409;
• 20 % of each fold's non-test groups held out for validation (adopted);
• the 25/50/100 % training pools are nested;
• parts already examined in S1 are flagged "development_exposed" and reported as a secondary breakdown, without changing the folds.
3. *Domain and eligibility:*
• the primary domain is eligible nodes with d > 0; results are not described as covering the whole part;
• S2 eligibility = d > 0 ∧ valid spectrum ∧ N_eff ≥ 64 ∧ complete edge block. *S1's gate-direction requirement is dropped*, because S2 does not use gate q;
• draws are fixed per part; e_p is the per-part error estimator as MeltAstra wrote it;
• aggregation is equal-weight within each group, then equal-weight across groups;
• a coverage table is reported for every part.
4. *Connectivity bottleneck metric: deferred.* It needs full-graph predictions and full-node O2. No substitutes: no infilling with true labels, no query-induced graph, no zero-filling.
5. *Baselines and backbone:*
• *R2-G* is named a "Gre24-style approximation" (features as listed in the diff, plus tet-edge Dijkstra distance), not a reproduction. Sklearn HistGBRT with weights, at most 12 hyperparameter configurations. It is deterministic, so it counts as *15 independent fits*;
• the backbone is *Transolver* (MIT; backup GINO, MIT) with a decoder at the query nodes;
• two capacity tiers are fixed in advance, chosen on validation, and N and E use the same tier.
6. *Statistics:*
• *one primary contrast:* Δ = 1 − Ē_E/Ē_N, on the 100 % pool and the whole primary domain;
• *m = 10 %, now frozen*;
• the interval resamples groups while keeping each seed's runs paired (2,000 replicates, seed 2026111408);
• the other pools and slices are secondary;
• the neural route counts as "useful" only if N beats R2 with its interval's lower bound above 0.
7. *Resources:*
• first, measure preprocessing on 10 stratified parts;
• *pilot:* at most 4 runs and at most 2 GPU-hours, test fold sealed, stop if exceeded;
• the full matrix counts every run (90 neural, 15 R2 fits, HPO, pilot, reruns).
*Extensions:* the N0 arm without O2 is added only if Yanqi wants the whole O2 block tested (+45 runs). O1 is the next design item. For the in-house CAE, only the data-request and control-mode table is prepared now.
Grok, one targeted task: what exactly are Gre24's features (list), and how is its eikonal problem set up? If they can be matched cheaply I'll add them before the freeze; otherwise R2-E stays deferred.
Entry sequence is unchanged and ends with the stage review and Yanqi's go.