Deterministic validation firewall for LLM outputs. Verify AI proposals against ground-truth state using deterministic rules. Zero dependencies.
