B4C11 — Corrected Full-N Result Report

Gate: H-AIF-G5 · Cell: B4C11_M7_STRUCTURAL_IDENTIFIABILITY · Lane: B · Date: 2026-07-22 Result: hierarchical-aif/results/motor_stack_aif/B4C11_CORRECTED_FULL_RESULT.json sha256 564a5b0f4461c83fe2c8792a2304dc136799b0923d887a27a60572361a851b53 Committed prediction: hierarchical-aif/protocols/B4C11-CORRECTED-FULL-PREDICTION.md @ 897c8ab Defect routed: D1_C11_CLUSTER_COLLAPSE Maps to: P6 for C11 U4 only — a scoped, per-cell identifiability statement. No P-level is raised.

1. Executive result

B4C11 ran to completion at the full frozen N_boot = 2000 with the D1 cluster-collapse correction: 2000 completed, 0 failed, resourceBoundPartial: false, runStatus = ELIGIBLE_FOR_FROZEN_VERDICT.

collapseFraction_tau_lt_1e_3 = 0.0055 (11 of 2000) < 0.25 → verdict U4_OK. This is pre-committed branch (a).

• M7_status = IDENTIFIED_ON_THIS_COHORT — U4 newly computed on the corrected bootstrap; U1/U2/U3 carried forward verbatim from the frozen artifact.
• The withdrawn artifact stays withdrawn. The submitted U4_OK — 30 of 2000 replicates under the defective cluster bootstrap — is not restored by this run. D1 moves OPEN → CLOSED-BY-RERUN for the C11 U4 lane only.
• The frozen B4 prediction PROFILE_FLAT_OR_WEAK is REFUTED at full N (both arms — see §4).

2. Result detail

U1 / U2 / U3 carried forward (CARRIED_FORWARD_FROM_FROZEN_ARTIFACT, source sha256 f361e4dc…): U1 TAU_INTERIOR, U3 SUPPORTED_OVER_M1, U2 U2_OK (flatLogspan_normalized = 0.06666666666666665, far below the 0.50 fire threshold). U2 is a deterministic profile scan on the full unresampled cohort and never touches the bootstrap, so D1 cannot reach it — copied, not recomputed.

3. Provenance and integrity

NO FLOOR was applied: a non-finite log density or a None fit would have incremented failed; failed = 0, so every one of the 2000 replicates produced a finite optimum.

4. Prediction grading

4a. The frozen B4 prediction — REFUTED at full N

PROFILE_FLAT_OR_WEAK is a disjunction (flat tau profile and/or non-trivial collapse). Both arms fail: U2 flatLogspan_normalized = 0.0667 ≪ 0.50 (carried forward), and U4 collapseFraction = 0.0055 ≪ 0.25. The predictionOutcome field records REFUTED. The earlier REFUTED_U4_PARTIAL label is superseded, not upgraded.

4b. My committed directional prediction (record §7.3) — HIT on all five sub-commitments

The committed record predicted branch (a) — explicitly "against the mechanism's own direction of concern," on measured magnitude. Graded against the numbers I committed:

A correction I owe on the record. Earlier this session, watching the live counter at 330/2000 with collapsed=0, I said my committed prediction "appears headed for refutation." That was a misreading of my own committed prediction. I conflated the mechanism's direction of concern (the defective bootstrap should have suppressed collapse, so removing it makes collapse more likely) with what I actually committed in §7.3, which was branch (a): collapse stays low. Record §7.4 states this explicitly — "I am predicting against the mechanism's own direction of concern." The committed prediction was branch (a), and it hit. The live-counter aside should never have been offered as a read on the committed prediction, and the counter should not have been read at all.

4c. The mechanism's direction was ALSO confirmed — and it is the honest nuance

The correction did move tau toward collapse, exactly as §7.1 argued: the paired legacy-vs-corrected diagnostic (25 replicates, identical draws) shows corrected tau below legacy tau in 25 of 25 (medians 0.21298 → 0.16822, medianDeltaTau = −0.0546), and legacy group counts inflated to 46–59 against the corrected 80/80. So:

• The defective 30-replicate artifact was mildly over-optimistic about tau stability. Its tau median was 0.2196 and its p025 0.1766; the corrected full-N run gives median 0.168 and p025 0.0853 — a distribution shifted toward zero, with a p025 roughly half the submitted value. The direction of concern that motivated D1 was real and is now measured.
• But the effect is nowhere near the collapse boundary. The smallest of the collapse-relevant quantities, p025 = 0.0853, is ~85× the 1e-3 threshold, and only 0.55% of replicates cross it. The corrected verdict is still U4_OK.

The honest one-line reading: the old number was right for the wrong reason and slightly too confident; corrected, on a valid bootstrap, M7's dispersion is still identified on this cohort.

5. D1 closure

OPEN_UNTIL_CORRECTED_C11_FULL_RUN → CLOSED_BY_CORRECTED_RERUN.

• The corrected full-N run completed (2000/2000, 0 failed) on the D1-fixed builder, with the seed sequence proven identical to the frozen inline construction.
• The withdrawn artifact stays withdrawn. A defect that happened to produce a right-looking answer (0/30 collapse → U4_OK) is still a defect: it rested on 30 of 2000 replicates and a bootstrap that collapsed 80 draws into 46–59 groups. Neither is repairable retroactively, and the frozen artifact is not edited.
• The D1 effect size is measured, not asserted: −0.0546 median tau shift, 25/25 directional consistency, group-count inflation 46–59 → 80. This is the second demonstration after D2 that a defective/partial run produced a reading that a corrected full run revises.
• The N=5 paired diagnostic (C11-PAIRED-DIAGNOSTIC.json) and the 25-replicate LEGACY_DEFECTIVE_FOR_COMPARISON_ONLY arm remain diagnostic only and licensed no verdict; the verdict comes solely from the 2000 corrected replicates.

6. Prospectivity — SATISFIED

Verdict moves PENDING_NO_OBSERVATION_YET → SATISFIED, pinned by test_b4c11_prediction_was_committed_before_its_observation. Per D9, the flip PENDING → PROSPECTIVE belongs in the result commit, and the prediction and result must never share a commit — the result is currently uncommitted, so the ordering is intact.

7. Lane impact — scoped

B4C10's favourable M4 result was never transferable here, and is not invoked — different model (M4 pooled-i.i.d. vs M7 motor-grouped), different likelihood structure. This U4_OK stands on its own corrected bootstrap.

8. P-ladder mapping — map, never redefine

P6 is carried per scope. The scoped statement now in force:

P6 for C11 U4 is re-established on the corrected full-N run only (U4_OK, collapseFraction 0.0055, N=2000) — M7's dispersion parameter is identified on the derived_eligible_1_to_8 cohort under the frozen collapse criterion. The submitted 30-replicate artifact remains withdrawn. P6 for duration-only B3/B4 is unchanged; P6 for the Wadhwa mark-process is retrospective-only; P6 for the full motor-stack AIF is pending.

Identifiability is not correctness and is not mechanism. U4 is a stability check on one parameter (tau) of one frozen competitor (M7) on one cohort. It does not make M7 correct and is not evidence for any mechanism. P8 remains FULL_PARITY = false; first unsatisfied level P4 transfer. No P-level is raised.

9. Wording

Allowed: "corrected full run", "U4 re-established on the corrected run only", "the withdrawn U4_OK remains withdrawn", "CI-bound verdict" (where one applies), "not established". Forbidden as claims: the canonical list in claim_guard.FORBIDDEN — referenced, not re-transcribed. In particular this report does not assert that any diagnostic proves the U4 result, nor that any mechanism, parity, or active-inference claim follows from an identifiability check.

10. Limitations

One cohort; 80 training motors. U4 is a stability check on a single parameter of a single competitor under a single criterion. The corrected tau distribution is shifted toward zero relative to the withdrawn artifact — this cell shows it does not reach the collapse boundary, not that tau is sharply determined. Carried-forward U1/U2/U3 are copied from the frozen artifact and were not recomputed here. The retained adverse B3 finding (M2 over M3, ~0.0369 nats event-pooled) is unchanged and reported alongside.

NEXT_ACT = write hierarchical-aif/reports/B4C01-B4C11-INTEGRATED-CLOSURE.md