ARDLINETry the model
Model v2Trained 2014-2022 · validated 2023 · tested 2024

How the number is made.

A gradient-boosted classifier over fourteen game-state features, fit on FBS play-by-play from 2014-2022 and calibrated so a stated 70% actually wins about 70% of the time. No power ratings, no live market lines, no human adjustment.

0.117
Brier score, held-out season
0.362
Log loss
1.34M
Real plays in the training set
±6.1
Max calibration error, pts
01 · Inputs

Fourteen features, all derivable from the broadcast. Nothing about team quality enters the model directly except a single pregame market-implied prior - a 2-win team down 3 in the fourth gets close to the same in-game number as a playoff team in the same spot.

FeatureEncodingRel. gain*
Score differentialSigned, home − away0.290
Seconds remaining (regulation-normalized)Log-scaled0.190
Pregame priorSpread-implied win probability0.140
Yards to goal1-990.100
Garbage time factorContinuous, 0-10.080
DownOrdinal 1-40.060
DistanceYards, capped 300.050
Is home offenseBoolean0.030
Is overtimeBoolean0.020
Overtime periodOrdinal, 0 if regulation0.010
Offense timeouts0-30.010
Defense timeouts0-30.010
Rule era0 = pre-2021, 1 = 2021-22, 2 = 2023+0.005
Has pregame priorBoolean0.005

*Illustrative - no per-feature gain export has been published yet.

02 · Training and validation

Fit on 2014-2022, validated on 2023, held out on 2024 - split by season rather than at random, so no play from a game the model scores was ever in its training data. Overtime plays are included in the same unified model rather than a separate one, to avoid a probability discontinuity at the start of overtime; the model reports lower confidence there rather than switching curves.

03 · Calibration

Predicted probability against observed win rate, 2024 test season, decile buckets. On the diagonal means honest.

1005000predicted %100observed win %
04 · What it does not know
-Injuries, ejections, weather, or who is actually taking the snap.
-Team strength - by design. Blowout-prone matchups are under-modeled early.
-Coaching intent: a kneel-down and a hurry-up read identically to the model.
-Anything about betting markets, spreads, or totals beyond the single pregame prior.
Read the number as a range
Every output carries a band. A 68% early in the second quarter is a much softer claim than a 68% with two minutes left, and the interface shows both.
Sensitivity to small input perturbations is monitored alongside these metrics before every release.
YardLine · model v2Model estimate - not betting advice · Data provided by CollegeFootballData.com