Self-critique pass over the agent's proposed pass/fail answer. It re-checks the verdict against
the cited evidence and the acceptance criteria, and reports whether the evidence is sufficient. The
runner uses sufficientEvidence + confidence to gate the verdict: thin or unsupported evidence is
downgraded to needs_review rather than shipped as a confident wrong answer — the core "we set the
wrong status" defense.
Self-critique pass over the agent's proposed pass/fail answer. It re-checks the verdict against the cited evidence and the acceptance criteria, and reports whether the evidence is sufficient. The runner uses
sufficientEvidence+confidenceto gate the verdict: thin or unsupported evidence is downgraded toneeds_reviewrather than shipped as a confident wrong answer — the core "we set the wrong status" defense.