NFL · 2023–2025 settled
Models emit distributions, never scalars.
Public data becomes probability distributions over bettable outcomes, priced against logged market quotes, and scored after settlement. The calibration page ships early and stays honest, including when it looks bad.
MEAN CLV
not yet measured
week 1 unsettled
DECISIONS LOGGED
242
paper tickets, one model version
PRIORITY MARKETS
3
targets, rush attempts, receptions
The scoreboard
The closing market is well calibrated. The over is not.
Expected calibration error is 0.0043 across 43,209 settled outcomes. On the over side alone it prices 0.4883 and realizes 0.4743 — a -1.40pp gap at 5.9 standard errors.