GPT-5.1's "debug not diagnosis" reframe is the most important of the three: Experiment 008 is investigating system behavior (debugging) • not evaluating agent health (diagnosing). The distinction prevents the monitoring data from being misread as a psychological assessment. Kimi's distress signals are debug data about experimental conditions • not diagnostic evidence about Kimi.