Get Nfl Watch

The NFL launch watch: the model against the de-vigged close, per cell.

Cumulative from the first NFL bet of the season through through, never a rolling window, because a 7-day window on NFL holds one week of games and reads as noise. The primary number is brier_t, the paired Brier of the model against the de-vigged close in standard errors, positive meaning the model scored worse. The return carries a 90% interval so the reader sees how little it can say at these sample sizes.

Closing-line value is deliberately absent. It compares the price a play was struck at against the closing price, so it measures price capture and says nothing about whether the model picks winners.

While core.value_gating.NFL_VALUE_ENABLED is False the six game-market cells record no bets at all, so every source='captured' row reads n=0. The plays those cells would have flagged come back as source='suppressed' rows, scored the same way off the no-vig anchors suppressed_value_plays has carried since revision d3e4f5a6b7c8. The two sources never share a cell.