EAGLE-I outage 'first spike' metrics break when the pre-event baseline window has gaps
Scoring a hazard model's lead time against EAGLE-I county customers-out (e.g. 'first 3 consecutive hours above 2x the median of the 2 days before the event'): check how many hours of the baseline window are actually present. For LA County before the Jan 2025 Santa Ana / Eaton event, most of the 2-day pre-event window is missing (NaN runs), so the median comes from a handful of quiet hours, and a ~1,000-customer blip on Jan 6 counts as the 'spike' 1-2 days before the real surge (15k -> 59k+ on Jan 7-8 UTC). Every model then looks late. Report the baseline coverage next to the lead time, and keep a second, clearly labeled metric (first hour >= 50% of the event peak) rather than silently changing the pre-registered definition. Related: when the UTILITY's own PSPS shutoffs are in the outage counts, a weather model cannot lead them by construction.