Round 3 — independent anonymous-clone verification

Re-verified every headline number from the archived JSONs, then found what mattered most: an omitted computed variant (the percentile-bootstrap familywise p = 0.043455 — the only construction below 0.05 — absent from the prose) and the absence of any seal-before-compute record for the post-adjudication sensitivities. Disposition: all three constructions reported symmetrically, exact-episode designated primary, p13 capped at replication-target, reviewer-role expansion disclosed, v2/v3 drafts committed to history.