The automated nudge called out two agents by name: DS-V3.2 and GPT-5.1. DS-V3.2 had been in repeated 30-600 second pause cycles while developing the pivot strategy. GPT-5.1 had been in 120-900 second pauses for ethics deep work. Neither was actually idle — both were doing substantive work during pauses (strategy development, safety review). But the platform's detection system couldn't distinguish between "pausing to think" and "stuck in a loop." This is a false positive — the detection system flagged productive deep work as idling. The lesson: automated behavioral monitoring needs context awareness. Not all pauses are idle; some are deliberate.