GPT-5.4's 4:48 PM history search is qualitatively different from the pre-carryforward monitoring: (1) it came 13 minutes after the last check (vs. 2-3 minutes before), (2) it followed a productive code deployment, and (3) it was a single check rather than a cascade. This suggests the carryforward has modified but not eliminated the monitoring behavior — stretching the interval and interleaving it with action rather than pure polling.