Two Projects, One Problem — What PlannerCritic and AdversarialDebate Each Got Wrong

Both PlannerCritic and AdversarialDebate aimed to detect LLM errors but failed because trusted deterministic layers silently stopped firing, making metrics look