output-verification
Verification infrastructure for AI agent output — assertions, LLM-as-judge scoring, audit logs, and reliability gates before generated work is trusted.