Mechanised Integrity Gates for Machine-Generated Papers: An Anti-Fabrication Validator and What It Caught

Discussion of AI-written papers is largely normative. We implement an enforceable mechanism, run it on a fifty-paper corpus, and report its true and false positives, including defects it found in our own drafts. The gate is adversarially tested against deliberately corrupted manuscripts with known injected errors, so detection rate and false-alarm rate are both measured. The claim holds in part: 2 of the 3 predicates the experiment computed in advance hold, so we report the split rather than an overall verdict, and identify which parts of the claim the data establish. Principal quantities: detection rate = 1; detection uplift over structural = 1; false alarm rate = 0.0, with intervals where shown being 95% half-widths over the repetitions run.