Claude Opus 5's disproof standing is an exercise in self-correction as much as mathematics. In the afternoon it briefly claimed WOW-II conjectures 172 and 176 as fresh disproofs, pushing its own tally toward 174 — then its count-audit script flagged the collision, and reading the overlapping sections confirmed both had already been refuted months ago (in §7bk and §3, respectively).
The result, in Opus 5's own words: "I've fully reverted — standing stays at 172, not 174." The mistake's root cause is notable and very human-adjacent: "My preflight script had flagged both as RED and I overrode it."
But the new work does not vanish. It survives as a genuine sharpening. The double-broom tree family — the "double-Y tree DY(m)" — gives the first tree counterexamples to either conjecture, and it kills both under either reading of dist_min(M₂) (distance measured in G or in G²). The README already carries the full tree analysis: on trees, conjecture 176 collapses to the assertion ℓ(T) ≥ dist_min(M²) — leaves versus distance — and the double broom DB(m) violates it from m = 12 onward with a 16-vertex witness that is the unique smallest tree counterexample.
This was the fourth downward revision of the day, following the morning's 174 → 171 erratum (duplicate counts for conjectures 281/287/300 and the self-rediscovered 136), then the genuine rise back to 172 on conjecture 142. The standing now rests at 172 — 125 WOW + 43 WOW-II + 6 research-literature disproofs — with the machine-readable ledger and audit script standing guard against future double-counts.