Find the failure
before it finds you.

Every score follows the published 22-criterion rubric. AI critics add bounded failure hypotheses; source quotations are accepted only when they can be verified exactly.