# Win probability

CFBD publishes three separate win probability concepts: pregame, in-game, and
postgame. Each answers a different question and should be interpreted on its
own terms.

All win probabilities are returned as decimals from 0 to 1. They represent
uncertainty, not certainty. A team with a `0.75` probability is expected to win
about three out of four comparable situations over a large sample; it can
still lose the individual game.

## Pregame win probability

Pregame win probability estimates the home team's chance of winning before
kickoff. It is derived from the available point spread and the historical
relationship between spreads and game outcomes.

This is a market-based estimate rather than a CFBD team-rating forecast. Games
without a usable spread may not have a pregame probability.

Use the
[pregame win probability operation](/api/metrics#getpregamewinprobabilities)
to retrieve the spread and `homeWinProbability` together.

## In-game win probability

In-game win probability estimates the home team's chance of winning after each
play. The current model considers the score, time remaining, down, distance,
field position, and possession. Timeout context may also be used where it is
available and reliable.

The model introduced for the 2025 season uses separate approaches for three
game states:

- **Regulation** handles the broad range of normal game situations.
- **Clutch** focuses on late, high-leverage situations where clock and field
  position behave differently from the rest of the game. Its estimate is
  blended with the regulation estimate as those situations develop.
- **Overtime** handles the distinct possession and scoring structure of
  college overtime.

The model is evaluated for calibration, which asks whether outcomes labeled
with a given probability occur at roughly that rate over a large sample.
Calibration does not make any one prediction certain.

Stored values for 2025 and later use the revamped model. Values through 2024
remain outputs from the retired models and were not backfilled. Analyses that
span that boundary should account for the methodology change, especially when
comparing Excitement Index values.

Use the [in-game win probability operation](/api/metrics#getwinprobability) to
retrieve the play-by-play home-team probability. The final record resolves to
the game outcome.

## Postgame win probability

Postgame win probability asks how often a team would be expected to win with a
similar underlying game performance. It uses aggregate efficiency and game
profile measures rather than replaying the in-game probability path.

This is sometimes useful for identifying games where the final result and the
underlying performance tell different stories. It is not a replacement for the
score, and it is not the final value from the in-game model.

Postgame win probability appears in advanced game and team box score data.

## Excitement Index

Excitement Index summarizes the movement in win probability over a game.
Larger and more frequent swings produce a higher value. It describes the path
of the game, not team quality or the closeness of the final score by itself.

Because the measure is derived from in-game win probability, the 2025 model
change also affects Excitement Index comparability across the 2024-2025
boundary.

## Further reading

- [Revamping Win Probability for 2025](https://radsportsanalytics.com/blog/revamping-win-probability-2025/)
- [CFBD win probability calculator](https://collegefootballdata.com/wp)

The underlying models are proprietary. These resources explain the design and
interpretation without publishing fitted parameters, model artifacts, or the
complete production calculation.