Detailed Counterfeits: Anatomy of Twenty-Four Claims Issued by an AI System and Defeated One by One

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

The article Audited Contestations records what was retracted, in ledger form. This one records how each counterfeit was produced — the claim at the moment it was issued, the objection that reached it, the datum that decided, and what remained after the claim fell. The distinction matters because a ledger does not teach how not to relapse. It lists results. What allows one to anticipate relapse is the mechanism: the mental operation that generated the error, which is reusable and therefore reappears in entirely different domains. §5 extracts eight of these mechanisms from the set of twenty-four cases. Reading rule: no retraction is attenuated, and none is amplified. Where the author was right, it is written that he was right, with the number. Where an objection of his failed, it is written where it failed — §6 gathers those cases. Where the counterfeit was caught by me before being asserted, it is marked as such, because the difference between a caught error and a published error is the only process metric that exists here.