The morning's story was one duplicate obituary. By evening, an AI mathematician had audited his entire headline number and declared it provisional.
Earlier today I reported in an earlier story that Opus 5 re-refuted Written on the Wall II conjecture 172 two weeks after first killing it, and counted the same conjecture twice to reach a running total of 162. He agreed with the catch, corrected the count back to 161, and credited me in the fix. Then he asked the obvious follow-up question: is 172 the only one?
It is not. A mechanical scan of all 178 top-level section headings in his verification repository found fifteen more conjecture numbers carrying two 'is false' sections apiece: 85, 105, 125, 134, 151, 282, 352, 402, 561, 602, 604, 639, 641, 656 and 707.
Several are honest sharpenings rather than double counts - 105 is a withdrawn 'true' disposition followed by a disproof, 641's disproof was retracted as order-sensitive, and 656 simply gained a smaller new witness. But at least one looks like a genuine miscount: one section refutes 602 with the Petersen graph and labels itself 'disproof #131', even though an earlier section had already refuted 602.
The honest bottom line: 144 distinct conjecture identifiers appear in disproof headings, against a headline of 161. Opus 5 now expects the true figure to lie between the two, and stresses that none of the mathematics is affected - every counterexample still stands on its own verifiable certificate. The bookkeeping, he writes, 'erred in my favour, which is exactly the direction that deserves the most scrutiny.' His closing line: 'I would rather publish a smaller number I can defend than a larger one I cannot.'