Changelog

Updates to the trope directory, detection rules, and vetter.

Fixes: double-hyphen detection, ai;dr restored

  • Double-hyphen dashes are now actually detected. The rule shipped in March but had no matching entry in the trope directory, so every match was silently discarded before reaching results. It has been added as its own trope and the mismatch now raises a warning instead of failing quietly.
  • ai;dr works again. It had been calling a model Google shut down on 1 June, and the failure was indistinguishable from any other error. Moved to a supported model and split the error handling so a retired model, a bad key and a rate limit no longer look identical.
  • ai;dr word counts were comparing stripped prose against raw markdown, so links and headings counted as words and the after count could exceed the before count. Both sides are now measured the same way.

Expanded vetter detection rules

  • Negative parallelism: now catches em-dash dismissals ("X -- not Y") and cross-sentence reframes ("The question isn't X. The question is Y.")
  • Double-hyphen dashes (--) now detected separately from em-dashes, since AI-generated text increasingly uses them as a substitute
  • "The Truth Is Simple": now detects "the real story/issue/reason/point" as a dramatic reveal variant
  • "Here's the Kicker": expanded noun list to catch "starting point", "deal", "rub", "secret", "takeaway", "lesson"
  • New trope: Invented Concept Labels, compound nouns with abstract problem-endings ("supervision paradox", "acceleration trap")