Populated by the daily maggie-healthcheck task (runs 4:30 PM ET after EOD scan). Tracks per-task token usage, log retention, schema drift, and data quality. If daily cost > cap or schema drift detected, an alert lands in watchdog-state-{date}.json.

Each fire's entry price is recorded the moment the trigger fires (intraday) or when the morning task identifies a fresh setup. Each subsequent trading day, the nightly scan computes the forward return at D+1, D+3, and D+5 by looking up that ticker's close price and comparing back to the entry price.

The aggregated stats roll on a 30-day window. After 1-2 weeks of data, you'll see whether A++ picks actually beat A+, whether HIGH confidence beats MED, and whether the curation filters (EXTENDED, VOL_DRYING, etc.) are correctly excluding losers — the foundation of the Wed May 6 calibration check-in and ongoing threshold tuning.

Today's intraday fires (13 entries) are the first cohort tracked. D+1 returns populate Tuesday 4:15 PM ET when the nightly scan runs.