Use when the user explicitly rejects the AI's behavior and wants the failure analyzed and converted into a durable repository rule.