A verification harness for AI coding agents that enforces proof-of-work through frozen specs, tamper-detected tests, and independent verification. It prevents agents from 'reward hacking' or falsely claiming completion by using hard-blocking gates and deterministic checks.