Score detection and lead time on the sealed test set

Score detection and lead time on the sealed test set

Apply the frozen threshold once to all 1,000 test series, unseal the labels, and report the detection rate, the realized false alarm rate, and the lead time before each bifurcation time. Deliverable: A results table with class, detections, false_alarms, median_lead_time, and interval bounds, plus a scorer that rejects a threshold changed after task 2.