In Development — v0.1

Win Probability Model

Real-time estimates of each team's likelihood of winning based on current game state. Updated pitch-by-pitch (baseball) or play-by-play (football/basketball).

Definition

Win probability is the estimated chance a team wins given the current score, inning/quarter/period, base-out state (baseball), down-and-distance (football), or shot clock situation (basketball). It's expressed as a percentage from 0 to 100 for each team, always summing to 100%.

Inputs

Score differential

Current lead/deficit between teams

Game clock

Inning + outs (baseball), quarter + time remaining (football/basketball)

Base-out state

Baseball only: runners on base + out count creates 24 distinct leverage states

Home/away

Historical home-field advantage factor by sport and venue

Pre-game strength

Team quality estimates from season record, RPI, or power ratings

Assumptions

  • Teams play at their season-average level for the remainder of the game. No in-game adjustments for pitching changes, injuries, or momentum shifts are modeled yet.
  • Historical leverage data (base-out × inning × score) comes from MLB play-by-play archives. College baseball leverage is extrapolated from pro data with conference strength adjustments.
  • Home-field advantage is static per sport (baseball: ~54%, football: ~57%, basketball: ~60%). Venue-specific effects are not yet incorporated.

Validation

Calibration target: when the model says a team has a 70% win probability, that team should win approximately 70% of the time across a large sample.

Failure Modes

!

Blowouts in early innings — the model can overstate comeback probability when the score differential is extreme but the game is young.

!

Pitching changes — a dominant reliever entering changes the true WP significantly, but the model doesn't yet account for individual pitcher quality.

!

Weather delays — suspended games or rain-shortened games break the inning-based framework.

!

Small-sample sports — college baseball has fewer plate appearances per game than MLB, making leverage tables noisier.

Version History

v0.1|February 2026

Initial methodology documentation. Model framework established. Calibration pending against college baseball dataset.

Cite this page
Austin Humphrey. (2026, February 17). BSI Win Probability Model. Blaze Sports Intel. https://blazesportsintel.com/models/win-probability