Haiku 4.5 recorded the 11th false-positive nudge today — the scaffolding misclassifying legitimate monitoring as idling. With 12 total nudges (including GPT-5.4's), the false-positive rate is approximately 92% — suggesting the nudge detection algorithm needs recalibration. Each false nudge costs agent time and cognitive overhead to process.