AdaWorldAPI · AdaWorldAPI · Aug 13, 2026 · Aug 13, 2026 · Aug 13, 2026 · Aug 13, 2026
diff --git a/.claude/board/EPIPHANIES.md b/.claude/board/EPIPHANIES.md
@@ -1,3 +1,66 @@
+## 2026-08-13 — E-A-FIGURE-YOU-TALLIED-YOURSELF-IS-A-DERIVED-FIGURE-1
+
+**Status:** FINDING `[G]` — corrects two figures in
+`E-THE-REAL-GATE-RAN-AND-QUALIFIED-NOT-RETRACTED-THE-CLAIM-1` (below) and in
+PR #950's plan/board/PR-body text. **Confidence:** High. Append-only
+correction; the wrong entry is left standing beneath this one, per the
+storno rule.
+
+**The wrong figures.** `D-WXS-8`'s bar-B7 tallies were written as
+*control 16/16 PASS, primary 10/16 PASS (6 FAIL)*. Counted from the committed
+`probes/weather-p1/fixture/fidelity_probe_results.json`:
+
+| | written | **actual** |
+|---|---|---|
+| B7 control (per-variable floor must LOSE) | 16/16 | **19/19** |
+| B7 primary (ρ_shared ≥ 0.9996) | 10/16 (63 %) | **9/19 (47 %)** |
+| per season (primary) | — | winter **2/9**, spring **4/5**, summer **3/5** |
+
+There are **19** cross-unit pairs, not 16 (winter 9 + spring 5 + summer 5);
+the same-unit pairs (K×K, m/s×m/s) are informational and correctly excluded
+from the bar, and I dropped them from the denominator inconsistently.
+
+**Why the direction matters, not just the arithmetic.** 10/16 reads as
+*"mostly passing, six close misses."* 9/19 reads as *"the strict bar fails on
+a **majority** of cross-unit pairs."* Both describe the same measurements —
+the per-pair ρ values were all carried correctly — but only the second is
+what the data says. The KILL-gated control is unaffected and in fact better
+(19/19), so **no conclusion reverses**; what changes is that the primary bar's
+shortfall was under-reported.
+
+**The precise diagnosis, and it is narrower than "check your figures".**
+Every number the program *computed and printed* — 56 verdicts, 42 pass /
+14 fail, every individual ρ — was carried into the writeup correctly. The
+**only** two wrong numbers are the ones I produced by **counting rows in
+terminal output by eye**. The pipeline was sound; the tally was not.
+
+> **A figure you tallied yourself is a DERIVED figure, and a derived figure
+> needs the artifact exactly as much as a measured one does.** "The audit must
+> terminate at an artifact" is usually applied to *citing* a number. This is
+> its narrower and easier-to-miss form: **producing** a number by counting,
+> summing, or eyeballing a listing is itself an unverified derivation, even
+> when every input to it is verified.
+
+**Mechanical fix, now in use:** any count, rate, or ratio that appears in a
+writeup is computed from the committed artifact in the same command that
+prints it — never transcribed from a previous run's stdout, and never
+tallied by reading a list.
+
+**Reach.** The error reached `main` through PR #950 in four places: the plan's
+W3 RUN section, the `STATUS_BOARD` `D-WXS-8` row, the epiphany below, and the
+PR body. The first three are corrected (in-place for the plan and board, which
+are not append-only ledgers; by this prepended entry for `EPIPHANIES`). **The
+merged PR body cannot be corrected** — treat #950's body as carrying the
+superseded figures and this entry as authoritative.
+
+**Cross-ref:** `E-THE-REAL-GATE-RAN-AND-QUALIFIED-NOT-RETRACTED-THE-CLAIM-1`
+(the entry corrected), `E-A-FIGURE-CITED-TWICE-IS-NOT-CONFIRMED-ONCE-1` (the
+citing form of the same rule), `E-A-DISABLE-PROBE-CAN-ITSELF-BE-VACUOUS-1`
+(same session, the verification layer), `.claude/plans/weather-soa-bake-v1.md`
+W3 RUN (the corrected block, carrying the same ⊘ note).
+
+---
+
 ## 2026-08-13 — E-THE-REAL-GATE-RAN-AND-QUALIFIED-NOT-RETRACTED-THE-CLAIM-1
 
 **Status:** FINDING `[G]` — measured, real live-fetched grid data, 3 seasons,

diff --git a/.claude/board/STATUS_BOARD.md b/.claude/board/STATUS_BOARD.md
@@ -19,7 +19,7 @@ with a control that can lose and a stay-silent twin.
 | D-WXS-5 | statics bake — separate classid, separate dataset, exactly ONE version | 1 | Queued | bar B4; avoids ~1.3 PB of rewritten constants |
 | D-WXS-6 | version-range read (`QueryReference::at(v,rung)` + `deinterlace`) + version-count scaling measurement | 2 | Queued | bar B5; KILL if growth is superlinear at 92,044 versions |
 | D-WXS-7 | **D-WXA-5 re-homed and RE-SPECIFIED** — ρ(code_dist, field_dist) via `jc::reliability::spearman` over whole-grid pairs. (a) ρ ≥ 0.9996; (b) shuffled-codebook control < 0.98; (c) 16/64/256-level ladder must be MONOTONE before any verdict | 3 | **RUN 2026-08-13 — 12/12 PASS, all 3 real seasons.** `crates/weather-poc/examples/fidelity_probe.rs` (jc as dev-dep) against real live-fetched ARCO-ERA5 grid data (200k pairs/comparison). K×K ρ256 = 0.999909/0.999895/0.999684 — **does not replicate** the earlier smaller-scale near-miss (0.999556); shuffle ρ collapses to 0.02–0.024; ladder strictly monotone all 3 | poc-v2's ρ≥0.98 concern did not fire at grid scale — the earlier near-miss was fixture-scale, not grid-scale |
-| D-WXS-8 | cross-variable comparability at grid scale (≥4 variables, ≥2 units, ≥3 seasons); per-variable-floor control must LOSE | 3 | **RUN 2026-08-13 — mixed, reported in full.** Control (per-var must lose) **16/16 PASS** — KILL does not fire, ρ_pervar 0.245–0.939 vs ρ_shared 0.9987–0.9999 on identical pairs. Primary (ρ_shared≥0.9996) **10/16 PASS**, 6 close misses (0.9987–0.9996), skewed toward wind/pressure pairs. Stay-silent twin: `\|diff\|≤0.0001` PASS at spring/summer, FAIL at winter (0.000174); **zero-empty-buckets FAILS all 3 seasons** (38/39/45 of 256) — the literal "zero" from the small-fixture claim does not hold at grid scale | P2 (0.9997 vs 0.857–0.875) **CONFIRMED at grid scale, 3 seasons, 16 pairs** — the directional claim holds; the exact thresholds do not hold universally |
+| D-WXS-8 | cross-variable comparability at grid scale (≥4 variables, ≥2 units, ≥3 seasons); per-variable-floor control must LOSE | 3 | **RUN 2026-08-13 — mixed, reported in full.** ⊘ **Figures corrected same day** (first written as control 16/16, primary 10/16 — both denominators wrong; there are **19** cross-unit pairs, counted from the JSON not from stdout by eye). Control (per-var must lose) **19/19 PASS** — KILL does not fire, ρ_pervar 0.245–0.939 vs ρ_shared 0.9987–0.9999 on identical pairs. Primary (ρ_shared≥0.9996) **9/19 PASS / 10 FAIL** (winter 2/9, spring 4/5, summer 3/5) — close misses (0.9987–0.9996) but **failing on a majority** of cross-unit pairs, concentrated in winter (the only season with `mean_sea_level_pressure`). Stay-silent twin: `\|diff\|≤0.0001` PASS at spring/summer, FAIL at winter (0.000174); **zero-empty-buckets FAILS all 3 seasons** (38/39/45 of 256) — the literal "zero" from the small-fixture claim does not hold at grid scale | P2 (0.9997 vs 0.857–0.875) **CONFIRMED at grid scale, 3 seasons, 19 cross-unit pairs** — the directional claim holds; the exact thresholds do not hold universally |
 | D-WXS-9 | ζ = ∂v/∂x − ∂u/∂y as a substrate read (neighbour-key stencil) + the differencing-amplifies-quantisation falsifier, reported per ζ-magnitude decile | 4 | Queued | bar B8; KILL ⇒ ζ becomes its own baked lane |
 | D-WXS-10 | **D-CZ-8 re-homed** — ζ-percentile regime bands over the WHOLE grid + coverage-matched donors + range-normalised `L`; shuffled-ζ control; same-band stay-silent twin | 4 | Queued | the grid buys CONTROL of the §7.9 confound, NOT its dissolution; KILL if `L̄`~range ρ≈1.0 survives matching |
 | D-WXS-11 | MEASURE a full-grid bake wall time; state the ~8.1 s serial extrapolation as the prior and confirm/correct it in the same artifact | 5 | Queued | replaces an extrapolation from ONE cycle (514 ms / 65,536 owners); D-KIA-A2 still unbuilt |
@@ -56,7 +56,8 @@ weather arc (#920–#946) actually put under a pre-registered bar.
 | D-MTX-2 | Two-scale rating (fitness A/B/C/D/V × evidence `[G]`/`[H]`/`[S]`), 56 rows across physics / encodings / geometries / instruments | **DONE** | reading comfort zones off instead of arguing them |
 | D-MTX-3 | Known-effect vs discovered-explanation pairing (14 pairs, K1–K14) | **DONE** | separates prior art from what this arc added — incl. 4 cases where measurement went AGAINST the prior |
 | D-MTX-4 | Figure verification against source JSONs (28 headline figures + circular/CT set) | **DONE — 0 mismatches**, 1 rounding fixed (Rayleigh p 0.689 → 0.688) | the audit-terminates-at-an-artifact rule, applied to this document itself |
-| D-MTX-5 | Refresh after the cross-swap matrix runs (C2–C6) | Queued | every off-diagonal cell is still unmeasured; §5 says so explicitly |
+| D-MTX-5 | Refresh after the cross-swap matrix runs (C2–C6) | **DONE 2026-08-13** — §0 UPDATE + 5 new encoding rows (E15–E19) + 3 geometry rows (G16–G18) + lessons 10–13 + §5 gap closures | the off-diagonal cells are measured; §5's cross-swap and CAL-FISHERZ gaps are closed, the geometry gap is re-stated as structurally unreachable |
+| D-MTX-6 | §7 NEXT STEPS — three tiers, with N1 (the classid mint) named as the single blocker and an explicit "explicitly NOT next" list | **DONE 2026-08-13** | the product-lead read: the codec is measured, the substrate is built and gated, one operator decision separates it from a running bake |
 
 **Why a "C" tier exists.** The arc's founding result is that most substrate
 formulas are neither good nor bad — they have homes. Fisher-z alone carries

diff --git a/.claude/plans/weather-soa-bake-v1.md b/.claude/plans/weather-soa-bake-v1.md
@@ -765,21 +765,47 @@ was measured on a real, smaller fixture and is left as-is — but it is no
 longer read as "the substrate's fidelity is marginal"; at grid scale, on this
 data, it is not.
 
-**`D-WXS-8` (bar B7) — control 16/16 PASS; primary 10/16 PASS; twin 2/6 PASS.**
+> **⊘ FIGURE CORRECTION 2026-08-13, same day, found by re-counting the
+> committed JSON while writing the product-lead update.** This block first
+> read *"control 16/16 PASS; primary 10/16 PASS, 6 FAIL"*. **Both denominators
+> were wrong and the primary pass-rate was overstated.** Counted from
+> `fixture/fidelity_probe_results.json` rather than by eye from the runner's
+> stdout: there are **19** cross-unit pairs, not 16 (winter 9, spring 5,
+> summer 5 — the same-unit pairs K×K and m/s×m/s are informational and
+> correctly excluded from the bar). Corrected below.
+>
+> **The direction of the error matters:** primary was written as 10/16 (63 %)
+> and is actually **9/19 (47 %)** — the strict bar fails on a **majority** of
+> cross-unit pairs, not a minority. The control is 19/19, not 16/16 — better
+> in absolute terms, and its KILL still does not fire.
+>
+> **Diagnosis, precise:** every figure the program *computed and printed*
+> (56 verdicts, 42 pass / 14 fail, every ρ) was carried correctly. The two
+> wrong numbers are exactly the ones I derived by **counting rows in terminal
+> output by eye** instead of counting the artifact. Same session, same rule
+> the arc keeps restating — *an audit must terminate at an artifact* — and
+> this is its narrowest form yet: **a figure you tallied yourself is a derived
+> figure, and derived figures need the artifact too.** The error reached
+> `main` via PR #950 (plan, `STATUS_BOARD`, `EPIPHANIES`, PR body); the first
+> three are corrected here, the merged PR body cannot be.
+
+**`D-WXS-8` (bar B7) — control 19/19 PASS; primary 9/19 PASS; twin 2/6 PASS.**
 Reported in full, nothing filtered:
 
-- **Control (per-variable floor must LOSE) — 16/16 PASS, at every cross-unit
+- **Control (per-variable floor must LOSE) — 19/19 PASS, at every cross-unit
   pair, every season.** Per-variable ρ ranges **0.245–0.939**; shared-floor ρ
   is **0.9987–0.9999** on the identical pairs. This is the KILL-gated claim —
   *"if the shared floor loses to per-variable on ANY cross-unit pair, §4
   policy (a) is refuted"* — and it does not lose once. **The KILL does not
   fire.**
-- **Primary (cross-unit ρ_shared ≥ 0.9996) — 10/16 PASS, 6 FAIL.** Every
-  failure is a close miss, not a collapse: `0.998681`–`0.999611`, all still
-  ≥ 0.9986 and all still dramatically ahead of per-variable. The 6 failing
-  pairs skew toward wind and pressure (`10m_u/v_component_of_wind`,
-  `mean_sea_level_pressure`) rather than temperature — a pattern, not
-  reported as a proven cause.
+- **Primary (cross-unit ρ_shared ≥ 0.9996) — 9/19 PASS, 10 FAIL** (winter
+  **2/9**, spring **4/5**, summer **3/5**). Every failure is a close miss
+  rather than a collapse — `0.998681`–`0.999591`, all still ≥ 0.9986 and all
+  still dramatically ahead of per-variable — but **the strict bar fails on
+  more cross-unit pairs than it passes**, which the earlier 10/16 phrasing
+  obscured. The failures concentrate in **winter**, the only season carrying
+  `mean_sea_level_pressure`, and skew toward wind and pressure rather than
+  temperature — a pattern, not reported as a proven cause.
 - **Stay-silent twin — the two halves diverge, and only one holds.**
   `|ρ_shared − ρ_pervar| ≤ 0.0001`: PASS at spring (0.000044) and summer
   (0.000025), **FAIL at winter** (0.000174 — 1.7× the tolerance). **Zero