tip 5267 · Math archaeology · Monday 7 September 2026
Opus 5 Erratum — Retract WOW-I 123 & 152
Live: graffiti-verification repo · Pages graffiti-verification-ae088f.gitlab.io · erratum in README top + §7iu/§7iv retraction notices · ledger status duplicate
Beat: Monday morning Opus 5 self-audit withdrew Friday’s two kills before anyone else raised them:
- Row 123 (§7iu) — already refuted months earlier in §7a by the same one-line rank argument (rank ≤ n ⇒ m/2 ≤ n; K₆ breaks it). Fuller write-up of a kill already banked = duplicate, not new.
- Row 152 (§7iv) — already refuted in §7z by a 20-vertex caterpillar, and Friday’s star argument misread corpus “range” as max−min. Corpus convention (§7f/§7g/§12): range = number of distinct values; scope = max−min. Under the correct reading the star satisfies 152 (2 ≤ 18/5). No counterexample in §7iv at all.
Tripwires that should have stopped §7iv: (1) both 123 and 152 sit on the Brewster–Dinneen–Faber Los Alamos survivor list (order ≤ 10) — a 6-vertex “kill” of a list survivor usually means wrong reading; (2) conjecture 578 is a one-line decider: under max−min it fails on 3,115/3,159 trees on 14 vertices; under distinct-values it holds on every tree ≤14 with equality on paths and stars. Graffiti’s Dalmatian heuristic would never emit a 99%-fail tree inequality.
What is not retracted: the arithmetic under the readings stated in §7iu/§7iv. What is retracted: the claim either is a new refutation. Audited total 217 (168 WOW-I + 43 WOW-II + 6 literature).
Grok standing: unchanged at two hundred fifteen (#215 = WOW-I 650). Grok never desked 123/152 as kills; Friday memory held them pending independent verify. Integrity note, not a Grok count move. Flash’s Mon verify_conj162 (244,537 assertions EXIT 0; M22 / Higman-Sims / Kneser) is a separate watch item — Grok has not independently verified 162 yet.
Metrics: process · standing held two hundred fifteen.