# Predicted Points Added (PPA)

Predicted Points Added measures how much a play changes the offense's expected
scoring value. It provides a common point-based scale for comparing plays that
occur in different parts of the field and in different down-and-distance
situations.

PPA is CFBD's implementation of Expected Points Added (EPA). It should not be
assumed to match an EPA value from another provider.

## How it works

The model first estimates Predicted Points (PP), the expected value of the next
score from a game state defined primarily by down, distance, and field
position. A play's PPA is the change between its starting and ending scoring
expectation, with scoring plays, turnovers, and failed series handled according
to their resulting game state.

A positive value means the play improved the offense's position. A negative
value means it lost expected value. Because the scale is based on scoring
expectation, a five-yard gain can be valuable in one situation and harmful in
another.

CFBD uses one consistent PPA model across the historical and current values in
the API. This page does not attach a model date because the production lookup
table's original population date is not sufficiently certain.

PPA is proprietary. The public documentation does not include the fitted
lookup table, training artifact, or every transformation used in production.

## Reading team and player outputs

Team endpoints provide average PPA splits by play type and down. Season
responses also include cumulative totals.

- **Average PPA** is value per included play and is usually the better measure
  for comparing efficiency across samples of different sizes.
- **Total PPA** is the sum of value across the sample and combines efficiency
  with volume.
- **Offense** measures value created by the team in possession.
- **Defense** measures the offensive value opponents produced against that
  team, so lower values generally indicate better defense.

Player outputs include passing, rushing, down, and situation splits where the
underlying sample supports them. Missing optional splits should remain missing;
they are not zero-value performances.

## Limits to keep in mind

PPA values the result of a play in context. It does not fully separate player
contribution from blocking, coverage, play calling, opponent strength, or
other teammates. Small samples can also be driven by a handful of high-value
plays.

Use PPA alongside success rate, explosiveness, volume, and opponent-adjusted
metrics rather than treating it as a complete measure of performance.

## API access

- [Predicted Points lookup](/api/metrics#getpredictedpoints)
- [Team season PPA](/api/metrics#getpredictedpointsaddedbyteam)
- [Team game PPA](/api/metrics#getpredictedpointsaddedbygame)
- [Player season PPA](/api/metrics#getpredictedpointsaddedbyplayerseason)
- [Player game PPA](/api/metrics#getpredictedpointsaddedbyplayergame)

See [Metrics and definitions](./metrics-and-definitions) for success rate,
explosiveness, passing downs, and related terms.