First public release.
Theorem A (complete). For the centred-square spectrum (the odd squares, each twice) the Newton excess M(n,t) = n(p_t^2/(p_{t-1}p_{t+1}) - 1) exceeds 4/5 for every odd n >= 5 and every t < n/2; the constant is sharp. Five machine-checked pieces (a 627-rung ladder of exact certificates and three analytic certificates), each independently validated.
Theorem B (Sibuya's 1988 conjecture, large parts). Proved for every j <= 1000 and every n; for every j >= 1001 with theta <= 0.9; and at the top of the row whenever at most 802 indices are missing. The open region is stated exactly in ARTICLE.md section 7 and OPEN_PROBLEM.md.
Release-day reviews. Two adversarial reviews per theorem (results/DEBATE_*.md). On Theorem B they found a derivation resting on a Poisson approximation off by 2.8x; it was rewritten without one and re-run. On Theorem A they found no fatal step; one check compared against an unjustified threshold and now compares against the actual loss; one boundary description was imprecise and is corrected.
Every computation is exact rational or certified interval arithmetic; every certificate is an exit-0 script with a logged run in results/. AI involvement is disclosed in full in AI_DISCLOSURE.md. We are an independent lab and would be glad to be corrected.