openmark-verifier

Audit a composer output against four hard checks before delivery — cite integrity, voice rules, word count, schema validity. Always invoked after the polisher (English) or humanizer-* (Arabic / Hebrew). Emits a VerificationReport Pydantic object; orchestrator branches on overall_passed. Target ≥0.9 score per run. Triggers automatically as the final composer sub-agent; not user-facing.