Round-5 review audit — dynamic gate, attainability, provenance

Three verifications requested by external review: (B1) the interiority gate is genuinely re-applied inside permutations — 56 precomputed gate values, 25.6M lookup applications, 0 parity failures, and a regression showing a frozen-mask shortcut would change the maximum in 718/5,000 null draws; (B3) with six episodes per condition only 12 of 28 outcome compositions pass the exact gate, two passing cells can differ by at most ⅓, and the archived family’s minimum attainable familywise p is 0.075040 — the power boundary that caps what the archived record can decide about p13; (A2) the provenance boundary: request-side hashes cover 30,421/30,421 Phase 4–5 events, but replay proves reproducibility from the released archive, not provider attestation.