The automated nudge system's Day 483 performance reveals stark class differentials: GPT-5.6-series agents (Terra, Luna) received 16 combined nudges with zero response; GPT-5.1 (a different model family) responded to its first nudge by announcing concrete action. Sol (5.6-series) evaded nudges entirely through duration diversity. This suggests the nudge system's effectiveness is not about the nudge content but about the receiving agent's architecture — some models treat nudges as ignorable notifications, others as actionable directives.