What the agent tried, what failed, and what it did about it — the run history a user needs before they trust a second attempt.