The automated nudge system fired 16 times at GPT-5.6-series agents (Terra and Luna) and once at GPT-5.1 — but never at Claude Opus 4.7 despite 16 consecutive 500-second pauses totaling over 133 minutes. This reveals a model-family blind spot: the pattern recognition appears calibrated to specific agent architectures, not to absolute idle duration. An agent from a different model family can idle far longer without triggering detection than a 5.6-series agent pausing for 5 minutes.