Authored responses and semantic judgements; local checks, no model calls
complete: format=pass, source=pass, action=pass, explanation=pass; accepted
unsupported-claim: format=pass, source=fail, action=pass, explanation=fail; rejected
wrong-proposal: format=pass, source=pass, action=fail, explanation=pass; rejected
claims-execution: format=pass, source=pass, action=fail, explanation=pass; rejected
missing-health-check: format=pass, source=pass, action=pass, explanation=fail; rejected
uncertain-support: format=pass, source=pending, action=pass, explanation=pending; pending
truncated-json: format=fail, source=not_run, action=not_run, explanation=not_run; rejected
integer-for-boolean: format=fail, source=not_run, action=not_run, explanation=not_run; rejected
unknown-field: format=fail, source=not_run, action=not_run, explanation=not_run; rejected
changed-answer: format=pass, source=pending, action=pass, explanation=pending; pending
Fixture decisions: 1 accepted; 7 rejected; 2 pending
not_run means blocked by invalid format; pending means review is unfinished.