#929 hygiene + open-review sweep of #920–#930 (3 frozen figures corrected, append-only audit replaced) + report §10 + weather-w-probes-v1 (MIXED)

.claude/board/INTEGRATION_PLANS.md

@@ -1,3 +1,29 @@
+## 2026-08-12 — weather-w-probes-v1 (WORKER-BRIEF PLAN; the W-probe wave after CT-F16's failed bars + the resultant-instrument demo)
+
+Plan: `.claude/plans/weather-w-probes-v1.md`. Status **ACTIVE for the
+exploratory tier**; the verdict tier (CT-F17) is GATED on CT-W6's result AND a
+mandatory independent adversarial spec audit before any bar is committed.
+Companion product-lead doc: `probes/weather-p1/COMET_TAIL_REPORT.md` §10
+(measurement standard · dipole vector-sum model incl. the stranded-storm regime
+via v_rel · corridor α physics · queue-and-bow anchors · sunflower/facet/
+spiral-ADI substrate · roadmap · worker model). Operator ask: *"auf product
+lead level dokumentation erweitern und die followup testreihen für sonnet
+agents runterbrechen zum token sparen … und stranded rescue."*
+
+Shape: **§0 is a verbatim-paste worker preamble** (iron rules incl.
+commit-bars-before-run; no board writes — one-writer rule; the fetch() code,
+t-index guard, wrap_deg/err_deg, circular(); controls-FIRST with the VOID rule;
+comparative sentences must name both operands — the #930 relation lesson baked
+into the brief; **stranded rescue** = checkpoint every storm to
+`<name>.partial.jsonl` with resume-skip + tag-file heartbeat in `exec-runs/`,
+so a killed worker resumes instead of re-fetching ~GBs). Wave 1 (parallel,
+synthetic/stored, cheap): **W5** spiral-ADI anisotropy on Vogel N=4096 ·
+**W2s-a** golden-vs-grid pairing on the real cos-lat metric · **W6** the
+two-component (geo + bow) deconvolution on the stored 19 storms — the
+identification test for the vector-sum model, with the stranded stratification
+(|v_storm| < 8 m/s vs ≥ 8 m/s) as bar B3. Gated: W2s-b (α window), W7, CT-F17
+(fresh 1959–1979 sample, V-test p<0.05 ∧ R̄≥0.35, audit mandatory). Doc-only.
+
 ## 2026-08-11 — weather-substrate-evaluation-v1 (EVALUATION PLAN; the known-vs-test ledger for the whole #915→#922 arc)
 
 Plan: `.claude/plans/weather-substrate-evaluation-v1.md`. Operator ask: *"create

.claude/board/LATEST_STATE.md

@@ -1,3 +1,26 @@
+## 2026-08-12 — open-review sweep of #920–#929 — three ledger figures were wrong, and one audit method cannot do what it claims
+
+### Current Contract Inventory — no new types (corrections to merged #927/#928 entries; append-only, corrected here not in place)
+
+Operator: *"930 has comments … check also previous 5 … if you want go back another 5."* All 68 review comments on #920–#930 enumerated and checked against the tree rather than against the merge — **with non-uniform depth, stated rather than glossed:** #922–#930 finding-by-finding; **#920 (27) and #921 (5) spot-checked on their P1/governance items only** (both clean — `p1_noise_floor.py` is explicitly `SUPERSEDED by p1_ci_vs_floor.py`; the merged #917 entry has exactly one commit in its history, so it was never mutated), **leaving ~30 older findings there UNVERIFIED**. Within the verified range: **#922, #924, #925, #929: zero review comments. #926's 27 findings: all verified fixed in the tree** (equal-budget `grid_pts`, seam-wrapping `subgrid_min`, `find_center → None`, F7d 35→40, CT-F12 `NO-VERDICT-INSUFFICIENT-N`, persisted storm metadata, `np.roll` longitude, `__file__`-relative JSON write, net-decay E6, the 93–97 %→90.9–94.3 % headline). **#923's plan P2 is resolved** (`Status: ACTIVE (audited 2026-08-11)`). Three findings on **#927/#928 were still open**, all of them numbers frozen in append-only ledgers:
+
+- **⊘ CORRECTION to #926's entry (line 29 below), codex P2 on #927 — "a +92.76 Pa offset moves R² in the 5th decimal" is FALSE, and the report's own carve table refutes it.** Measured: **+92.76 Pa (carve A) moved R² 0.9212 → 0.9129 = 0.0083, the THIRD decimal**; **+1.59 Pa (carve D) moved it 0.943406 → 0.943403 = 2.4e-06, the SIXTH.** The sentence fused carve A's magnitude with carve D's insensitivity. **The true and sharper statement: the `var()` BUG was blind at EVERY magnitude — that is what hid +92.76 Pa — while the STATISTIC is near-blind only in the single-digit-Pa regime, which is exactly where "lossless" was claimed on four-decimal agreement.** The banked doctrine (RMSE + mean bias in Pa beside every R²) is unchanged; only its illustration was mispaired. **Third instance this week of the same shape as #930's relation error** — two individually-true numbers asserted of one pair when each belongs to a different one.
+- **⊘ CORRECTION to #927's `Added` line, codex P2 — "10 probe scripts with committed JSON results" undercounts.** Measured on the #926 first-parent diff: **15 `.py` added, 11 of them with a committed `.json`** (the four `ev*` scripts have none). The immutable scope summary omitted delivered work.
+- **⊘ CORRECTION to #928's audit figures (line 17 below), codex P2 — "+13/−0, +10/−0, +0/−0" is wrong in two places.** Measured `a4e264c5..3dae97be`: LATEST_STATE **+13/−0** ✓, PR_ARC_INVENTORY **+17/−0** (never +10 at any commit in the range), and EPIPHANIES **does not appear in the net diff at all** — it was **+1/−1 at `0f9e6bcd`**, the in-place edit of a merged entry, which the revert zeroed. So the "+0/−0" was the *product of a revert*, not evidence of purity — and read correctly it is the stronger result: **the audit WOULD have caught the violation mid-PR.**
+- **⚠ AND THE AUDIT METHOD ITSELF IS WEAKER THAN CLAIMED (codex P2 on #928).** *Zero removed lines proves ADDITIVE, not PREPEND* — an insert in the middle of a ledger, or an append at the end, also deletes nothing. Since #928 banked it as "worth reusing", a future session would inherit a test that cannot detect a mid-file insertion. **The correct one-command pure-prepend test is the SUFFIX check** (CodeRabbit's own analysis chain used it): `git show origin/main:F` must be a suffix of `git show HEAD:F` — i.e. `new.endswith(old)`. Zero-deletions stays useful as the cheap first screen; the suffix check is the one that proves the property.
+  - **Both halves measured before banking this, per the falsifiability rule.** Constructed falsifier: a true prepend gives `zero-del=True, suffix=True`; an **append at the END** gives `zero-del=True, suffix=False` — caught only by the suffix check, which is the discrimination the old method lacks.
+  - **And it fires on THIS PR, correctly.** `origin/main..HEAD` for #930: `PR_ARC_INVENTORY` / `STATUS_BOARD` / `INTEGRATION_PLANS` all `suffix=True`; **`LATEST_STATE` `suffix=False` at `+15/−0`** — the old screen passed it, the new one flags it. The flag is right and the change is still legitimate: #930 inserts shipped-PR table rows **mid-file** (§Recently Shipped PRs) and composes an **unmerged** entry in place (the PR-878 allowance). **A `False` is not automatically a violation — it is a demand for the justification to be stated.** That is the property worth having; the old test could not even ask.
+
+## 2026-08-12 — lance-graph #929 (MERGED) — steering rescue dead; sign test retired as primary instrument
+
+### Current Contract Inventory — no new types (2 probes + report §5.12/§5.13 + 1 epiphany)
+
+- **CT-F16 `[G]`:** the steering-level moderator FAILED both pre-registered bars on CT-F14's own 19 storms (0.579 vs 0.70; residual +28.4 % wider; level sweep monotone toward the surface, best 850 hPa). §9.2's "single most promising fix" is superseded in place. The height ladder survives as a measurement; its operational reading died.
+- **The instrument calibration `[G]`:** a 90°-rotated control scored **0.684 = CT-F14's headline**. At n=19 the sign test separates from a wrong answer only at 14/19. **Rule: attach a deliberately-wrong referent to any rate-headline; its score is the floor.**
+- **The replacement instrument:** circular resultant (R̄, μ, Rayleigh) resolves the same rows at **p=0.0050** (sign test: 0.0835), estimates the offset (**−30.2°±36.5°**) instead of penalizing it, and separates the rotated control in **BOTH channels** — vs surface (the pair the sign test conflated at 0.684 = 0.684): R̄ 0.343 vs 0.516 AND μ −130.5° vs −30.2° (100.3° apart); the steering↔rotated pair confirms rotation-invariance by construction (identical R̄ 0.343, μ shifted exactly 90°). Post-hoc — instrument demo, NOT a promotion. **Sign-consistency numbers are no longer verdict-grade anywhere in this arc.**
+- **Faltung frame:** resultant = first circular Fourier coefficient; CT-W6 = deconvolution (components ⊛ apparatus noise); `Z_256` circular Faltung is `DistanceLut::circular()`-native.
+- **Queued, operator go-ahead pending:** **CT-W6** (dipole = neighbor far-field + bow-wave vector sum, global fit, 38 obs — the mechanistic candidate for the −30° offset and the 0.684 plateau) · CT-W2s/W5/W7 (sunflower collision facets, spiral-ADI via Fibonacci-stride pairs, Gegendruck) · CT-F17 (fresh-sample verdict) · C-register climatological calibration.
+
 ## 2026-08-12 — lance-graph #927 (MERGED) — board hygiene for #926 + the living-vs-ledger correction rule
 
 ### Current Contract Inventory — no new types (board entries + one report correction)
@@ -831,6 +854,11 @@ Membrane consumers can now pull BOTH halves of a render `classid` BBB-safely fro
 
 | PR | Merged | Title | What it added |
 |---|---|---|---|
+| *gap note* | — | **#781–#925 are NOT in this table** — carried by the dated sections above + `PR_ARC_INVENTORY.md`. Recorded 2026-08-12 (codex P2 on #930) rather than silently reconstructed; the table had stalled at #780. | — |
+| **#929** | 2026-08-12 | CT-F16 kills the steering rescue; the control scores the headline; circular resultant replaces the sign test | 2 probes (`comet_tail_f16`, `comet_tail_resultant_instrument`) + report §5.12/§5.13 + `E-THE-CONTROL-SCORED-THE-HEADLINE-1`. Steering FAILED both bars (0.579 vs 0.70; residual +28.4 % wider; sweep monotone to surface, best 850 hPa); rotated control scored 0.684 = CT-F14's headline; resultant resolves the same rows at p=0.0050, offset estimated −30.2°. |
+| **#928** | 2026-08-12 | Board hygiene for #927 — chain terminator | Pure prepend (+13/−0, +10/−0). Living-documents vs append-only-ledgers correction discipline banked; zero-removed-lines diff = structural append-only audit. |
+| **#927** | 2026-08-12 | Board hygiene for #926 + 4.7×-not-5× correction + append-only revert | Mixed. 9/10 figure self-verification caught the 6 % favourable rounding; an in-place edit of a MERGED EPIPHANIES entry reverted same-PR (the rule being cited was the rule broken). |
+| **#926** | 2026-08-12 | weather-p1: the storm spine (90.9–94.3 %), the 12-byte L4 carrier, moderators identified but unwired | 9 919 insertions, zero Rust/product code. Spine `[G]` (3 blind samples, 41+ storms); 12-byte `6×(8:8)` facet at 0.07 Pa RMSE / +1.59 Pa bias; Fisher-z rank/tail-vs-level demarcation; CT-F14 NO-VERDICT held against a pooling rule that said ESTABLISHED; `var()`→MSE R² fix at 11 sites (hid +92.76 Pa bias). |
 | **#780** | 2026-07-21 | Recipes audited → 3 real tenants wired (A9 24-loci CausalWitnessFacet + SPO + qualia) → rung-ordered NaN-gated causal ladder | 4 commits, merge `8a00988`. `recipe_claim_audit` proves the 34 kernels sound-on-proxy (28/34; 2 INERT CAS/ETD + 4 CONSTANT ARE/ZCF/ICR/HKF) but 0/34 read a real organ → wiring gap, not composition. New contract modules: `causal_witness` (A9 = 24 signed-i4 loci, THIRD ClassView reading of the #729 12-byte register, no stored bytes), `recipe_substrate` (SubstrateView projects SPO+witness+qualia; **qualia additive + stakes-only**, causality = KAUSAL witness edge; missing tenant → NaN), `recipe_dispatch` (rung-ordered NaN-gated ladder keyed by NARS inference; ICR #31 unshadowed; the systematic-sweep mode COMPLEMENTING `select_tactic`'s 8/34 saccade — E-LADDER-UNSHADOWS-SELECTOR-1). +18 tests → 970 green, clippy clean; 3 probes all-gates-green. Theory anchors: PCRLLM 2511.08392 (proof-carrying single-step, Pei Wang), Causal-ML survey 2206.15475 (SCM ladder), MME-Reasoning 2505.21327 + NARS-Swift flagged; cc-thinking-skills mapped as rung-4 macro catalogue (E-THINKING-SKILLS-ARE-RUNG-4-MACROS-1). D-REC-WIRE-1. |
 | **#676** | 2026-07-10 | Post-#674 doc arc: E-NOBODY-WAITS-1 + VISION.md (graded AGI canon) + ancestry census + D-MTS-1 design inputs | Doc-only, 4 commits. E-NOBODY-WAITS-1 banked (no messages, no actors; ractor = compile-time ownership only; `&mut` IS the serialization; prime invariant "nobody waits for anything or any scheduling"; the ack-gated advance is canonical for the **ticket tier only** — `kanbanstep` stays canonical for stream reasoning, and the ack (the **SLA gate**) is reserved for the OGAR ticket tier, repurposed consumer-side as the **actionhandler queue** (see E-ACK-HARD-GATE-VS-KANBANSTEP-STREAM-1 + E-ACK-SLA-GATE-ACTIONHANDLER-QUEUE-1); supervisor KanbanMsg drivers = TD-MESSAGE-RESIDUE, leave-as-is per operator). `.claude/v3/VISION.md` = the AGI-aspiring canon, every claim graded [G]/[G-on-proxies]/[RULING]/[ASPIRATION], 2-Sonnet-preflight + 2-Opus-filigree provenance, all 9 review fixes applied. MODULE-TABLE ancestry census: thinking-engine (51) + p64-bridge (1) + cognitive-shader-driver (22 src + build.rs) with gem-status column. Plan addendum: D-MTS-1 frozen comparison — AriGraph context V3-TENANT-SHAPED; arm-discovery + DeepNSM ingest legs; CAM-PQ 6×8=48-bit path codes (address side) vs 6×palette256² = 12 B = one V3 tenant (value side). coderabbit 5/5 findings fixed in `093489c`. Merge `001839e`. |
 | **#674** | 2026-07-10 | V3 W2–W6 continuation + comma quorum/awareness measured + 5+3 council + StyleFamily dedup (D-TSC-1) | 16 commits. `contract::style_family::StyleFamily` (12 orchestration families, frozen ordinals Deliberate=0..Metacognitive=11) + `ThinkingStyle::family()` (36→12 total) + `default_runbook()` — five mutually divergent hand-rolled 12-style tables replaced (E-FIVE-STYLE-TABLES-1); first live 5+3 council run (spec v1→v2→v3). D-MTS-5 comma quorum MEASURED GREEN (N_eff 11.00/12; boundary = spectral participation) + D-MTS-6 comma awareness MEASURED GREEN (k*=1: one stored truth bit/comma level ≈ aligned k=4; lattice buys ≈log₂12 effective bits; D-MTS-6b gates real CE64 shrink). `BatchWriter::ack_and_propose` first-ack-wins dedup (codex P2). 1549 tests green. Merge `cd5178e`. |