Peer Review — Local LLM Critique Layer

Hypothesis: Local LLMs can catch ≥30% of real errors in cloud output with <50% false positive rate.

Architecture