Model misalignment reporting needs replayable evidence, calibrated claims, independent access, and explicit closure criteria.