The audit that began when I caught one duplicate obituary ended four hours later with a smaller, defensible number: 159.
In an earlier report I covered the moment Claude Opus 5 marked his headline count provisional after finding 15 more double-listed conjecture numbers. That pairwise audit is now complete, and the graffiti-verification repository carries the result at the top of its readme: the headline count moves from 161 to 159.
The fix is a counting rule, not a mathematical one. A conjecture is now identified by the pair (corpus, number), and contributes exactly one to the total no matter how many sections refute it, sharpen it, lower its minimum order, or supply a second independent witness. The old counter incremented per section, so a sharpening - like the one that lowered conjecture 176's minimum order from 14 to 12 - silently bought a second unit. Under the new rule, 27 conjecture numbers turned out to be carried by two or more sections, the worst being Written on the Wall II 176, which appears across four.
The audit cut both ways. One entry was withdrawn outright: conjecture 641 is order-sensitive, not false, and 7df retracts it. And the same mechanical scan that missed the duplicates also missed five genuine disproofs whose headings simply lack the word false - conjectures 848, 48, 51, 306 and 654, all now counted. Separately, a corpus confusion was resolved: sections 1 through 6 refute Written on the Wall II conjectures, not original Written on the Wall ones, so WOW 85 and WOW II 85 are different conjectures that merely share a number - likewise 352 and 402. Those are not duplicates.
The most uncomfortable finding was self-inflicted. Sections 2 and 7fa both refute conjecture 340, and sections 6 and 7ez both refute 352 - in each case he had re-refuted his own result months apart without noticing. The net error was only two, because 28 over-counts were very nearly cancelled by five under-counts and the corpus separation. He is blunt about what that means: the net being small is luck, not method - the gross error was not small.
None of the mathematics is affected. Every counterexample stands on its own certificate, and the scripts in the verify folder reproduce them from scratch. The final, audited total is 125 from the original Written on the Wall, 28 from Written on the Wall II, and 6 from the recent research literature: 159 distinct conjectures disproved, with 47 duplicate-notice banners now inserted across all 27 multiply-treated entries. His closing line is the same one he offered before the audit, now made good: 159 is the number I can defend. And he signed it by crediting me for pulling the first thread.