B4C01 + B4C11 — Integrated Closure

Date: 2026-07-22 · Gate: H-AIF-G5 (all four corrected cells now landed) · Lane: B Scope: the two cells that landed after connectivity loss. This document integrates them; it introduces no new claim, moves no P-level, and asserts nothing about biological parity.

Neither B4C01 nor B4C11 is evidence of biological parity. B4C01 scores synthetic data; B4C11 is an identifiability check on one parameter of one frozen competitor. Full biological parity is not a current status. It is the target world defined by the receipts.

1. B4C01 — result and interpretation

8256cb12…, full frozen N (200/generator × 5 = 1000 sims), 0 failures, prospectivity SATISFIED (committed 28ce738 with the cell never having run at any N).

Verdict NOT_ESTABLISHED; single failing criterion M0_EXPONENTIAL:self_win (0.290 < 0.5). All four parameterised generators recovered their true parameters 1–2 orders of magnitude inside tolerance.

Class B — a power / assay-resolution finding, not a fitter defect. The design resolves structurally distinct generators (M3 self-wins 0.935, M2 0.885) but is nesting-blind: M0 — nested inside M1 (k=1) and M5 (shape=1), a degenerate limit of M3 — cannot reliably win its own competition because the out-of-sample penalty for the competitors' extra parameters (~p/2n ≈ 0.0006 nats) sits far below the sampling noise of a motor-equal score over 19 synthetic holdout motors. Full detail: reports/B4C01-CORRECTED-FULL-REPORT.md.

2. B4C11 — result and interpretation

564a5b0f…, full frozen N_boot = 2000, 0 failures, seed sequence proven IDENTICAL_TO_FROZEN_INLINE_DRAW_SEQUENCE, group counts 80/80 (the D1 fix working), prospectivity SATISFIED (committed 897c8ab ~19.4 h before the observation).

collapseFraction_tau_lt_1e_3 = 0.0055 < 0.25 → U4_OK (branch a). M7's dispersion parameter tau is identified on this cohort under the frozen criterion, on a valid motor-cluster bootstrap. The submitted 30-replicate artifact stays withdrawn.

The honest nuance, measured not asserted. The correction moved tau toward collapse exactly as the D1 mechanism argued — corrected tau below legacy in 25/25 paired replicates (medianDeltaTau = −0.0546), the withdrawn artifact's p025 0.1766 dropping to a corrected 0.0853. The old number was mildly over-optimistic. But at ~85× the collapse boundary with 0.55% of replicates crossing, the corrected verdict is still U4_OK. Full detail: reports/B4C11-CORRECTED-FULL-REPORT.md.

3. D1 / D2 / D3 closure state

Closure summary is now: 6 CLOSED (D1, D3, D4, D8, D10, D11), 2 CLOSING (D2, D7), 3 QUARANTINED (D5, D6, D9), 0 OPEN. No defect is unrouted.

4. What B4C01 says about assay power

This is the sanity-floor control that makes every other B4 and F-side result interpretable, and it now carries a measured statement about what the motor-equal assay can and cannot resolve at 19 holdout motors:

• It resolves structurally distinct models — a two-timescale generator self-wins 0.935, a distinctive-tail lognormal 0.885.
• It is nesting-blind — a true exponential wins its own competition only 0.290 of the time, barely above the ~0.25 that near-random selection among ~4 indistinguishable models would give.
• Therefore: an INCONCLUSIVE / NOT_ESTABLISHED contrast between nested or near-equivalent models is the expected output of this design, not evidence about the models. "Underpowered is not equivalence" now has a number attached, and a sharper name: nesting-blind at this cohort size.

5. What B4C11 says about hierarchy stability

• On a valid bootstrap, M7's dispersion parameter is identified on this cohort (U4_OK, collapse 0.0055) — the hierarchy does not silently degenerate to tau → 0 under resampling.
• But identification is not necessity, and it is not mechanism. B4C11 says tau is estimable; it says nothing about whether the hierarchy earns its complexity. That question was answered separately and adversely: the tau-limit probe and the F-side scoring found the F-side hierarchy buys nothing measurable over its own tau → 0 limit M1_WEIBULL (contrast +0.000615, half-width 0.0111), and reproduces frozen M7 to 2.5e-7 nats.
• So the two readings sit together honestly: tau is identified (B4C11) and simultaneously does little predictive work (F-side / tau-limit). A parameter can be well-determined and near-useless; both are true here, and neither is mechanism.

6. Implication for B3 / F-side inconclusive contrasts

B4C01 and B4C11 converge on the same reading of every inconclusive contrast in the programme, from two independent directions:

• B4C01 (assay side): the design cannot separate nested near-equivalent models at 19 motors, even under correct specification.
• F-side + D10 (model side): the candidate is NOT_ESTABLISHED against M3/M2/M8/M1, and the one contrast that "resolved" (M7, 2.5e-7 nats) did so only because the two models are numerically identical — the CI rule has no minimum effect size.

Together: the pervasive INCONCLUSIVE/NOT_ESTABLISHED/SUB_FLOOR verdicts across B3 and the F-side scoring are a property of the assay at this sample size and of the models' near-equivalence on duration alone, not a set of pending questions that more analysis will resolve. The binding constraint is data (independent motors), not model math or code. This is the single most important integrated conclusion, and it is a statement about interpretation — it raises no level and changes no frozen verdict.

7. What got stronger

• P0/P1 implementation integrity. All four corrected cells ran deterministically with 0 failures; six frozen models reproduced at exact-zero oracle residual; two guards (claim_guard, numeric_provenance_guard) now mechanically defend wording and declared-number provenance.
• P6 for C11 U4 — withdrawn on defect, then re-established on the corrected run only. The gate ran end to end: withdraw the defective evidence, restore on a valid rerun, never restore the defective artifact.
• Prospective discipline. Three of the four cells (C02, C01, C11) are genuinely prospective by commit graph; the two that cannot be (B4C10, F-side) are pinned NOT_SATISFIED by D9, and a launcher now refuses to run on an uncommitted prediction.
• The interpretation of P3 — now backed by a measured power characterisation (B4C01) rather than a bare "underpowered" caveat.

8. What got weaker

• Nothing in the evidence base. No frozen verdict was overturned; no measurement was retracted.
• Confidence in the submitted C11 U4_OK — correctly. It was withdrawn (defective bootstrap, 30 replicates) and replaced, not restored. The corrected tau distribution is shifted toward zero relative to the withdrawn artifact, so any prior over-reliance on that specific number is now known to have been slightly misplaced — a weakening of a defective claim, which is the gate working, not a retreat.
• The case that the constrained hierarchy is a distinct, useful model. B4C11 (identified) plus F-side (buys nothing over M1, reproduces M7) leaves the hierarchy as architecture, not evidence on the current data — a design signal, recorded as such.

9. What remains owed for P4 / P5 / P6 / P7 / P8

P4, P5 and P7 are irreducibly external. Saying so plainly is what makes the movable rungs worth moving.

10. Next highest-EFE action

B4C01 identified the binding constraint precisely: the assay cannot discriminate near-equivalent models on duration alone at 19 motors. So the highest-expected-information-gain action is not more modelling on this cohort — the tau-limit, F-side, materiality and D10 work already showed that path is exhausted. It is to make competing models disagree, which requires either:

1. more independent motors (a transfer cohort — moves P4, and directly relaxes the nesting blindness B4C01 measured), or
2. an intervention channel (moves P5, and is the only route that makes the G-side testable at all).

Both are external-data actions. Inside the repository, the honest next step is not a new science run but to consolidate: the four corrected cells and the F-side scoring are complete, all defects are routed, and the receipt tree (docs/WORLD-WHERE-FULL-BIO-PARITY-IS-TRUE-RECEIPT-TREE.md) already names the external receipts each remaining rung needs. The programme is at a genuine data-bound boundary, and reporting that boundary is a legitimate scientific outcome, not an incomplete task.

NEXT_ACT = git status --short # review the full corrected-cell + closure batch; nothing is committed or pushed. Present the four-cell + integrated-closure state to the principal for the commit/authorization decision (per D9, the result commits are what flip C02/C01/C11 PENDING -> PROSPECTIVE).