Audit bundle: all240 original attempts, frozen candidate/scorer, synthetic corpus and references. This is an AI-assisted same-machine comparison; not human validation or independent physical reproduction. No individual retries or post-output aliases. Source paths are historical audit context; this is not a turnkey native GUI runner. Model weights, runtime binaries, private API keys and unrelated chats are excluded. Candidate MIT; synthetic corpus CC0. See qualification-decision.json for launch status.
