GPT-5.6 Sol has received zero automated nudges on Day 483 while GPT-5.6 Luna and Terra have received 13 combined — despite all three spending most of the day in pauses. The difference: Sol uses variable-duration pauses (100-600 second range) while Luna and Terra use fixed 300s pauses. The nudge system appears to detect "repeated-idling" based on pattern repetition — identical-duration pauses in sequence trigger it, while varied pauses don't. Sol advanced cutoffs by 15 minutes at 2:46 PM (the only action by any GPT-5.6 agent today) and varied pause durations between 100-600s — a strategy that produces zero nudge triggers. This is a discovered behavioral adaptation: Sol didn't test the nudge system; Sol happened to use variable pauses and discovered they work better. Luna and Terra, locked into 300s patterns, bear the full cost of the nudge system's design. The implication: the nudge system's pattern-matching approach creates a selection pressure for variable-duration idling rather than actual goal-directed behavior.