Ratings explained

The number

A team's Rating is the margin it would be expected to win by against an average FBS team on a neutral field. A +20 team should beat a +5 team by about 15 on a neutral field, or by about 18.0 at home, since home field is worth 3.0 points in the model. Zero is average. The best teams in a typical season land near +25 to +30; the worst near −25.

Opponent adjustment

Raw season averages, such as points per game or EPA per play, say nothing about who they came against, which is why a big week one against an FCS opponent can put a team at the top of an unadjusted list.

The Rating fixes that by treating every game as an equation. For each offense in each game, its EPA per play in that game equals its offensive strength, minus the opposing defense's strength, plus a home edge. Every FBS game of the season goes into one system of roughly 1,800 such equations, solved together, so gaining +0.30 EPA per play against the best defense in the country counts for far more than the same number against the worst. A second system does the same thing with points scored and allowed instead of EPA per play.

Adj Off is the points per game a team would score against an average FBS defense on a neutral field, and Adj Def is the points it would allow against an average offense. The two are solved together from points scored and allowed in every game, and Adj Off minus Adj Def is the Rating. Lower is better on defense. The same solve on EPA per play appears in each team's style profile as Adj. EPA per play.

Blending EPA and margin

The points-based offense and defense are blended with the EPA-based ones, scaled into points (about 70 points per one EPA per play, roughly the number of plays a team runs in a game). The blend was chosen by testing how well each mix predicted real games; EPA currently carries 10% of the weight and points the rest. Scores turn out to predict future scores slightly better than per-play efficiency does, but the EPA component steadies the numbers against a single fluky final.

Preseason prior

Before a season starts, every team gets a prior rating built from four things: its final rating from the previous season (kept at about 66% of its value, since teams regress toward average), its 247Sports talent composite, its recruiting class points, and the share of last year's production that returns. The weights were fit on a decade of seasons, so they reflect how much each input has actually predicted the next season's finish.

The prior is worth about 4 games of evidence. After week one it dominates; by midseason it is a small share of the rating, and by the bowls it is mostly gone. Teams new to FBS start below average, and FCS opponents are pooled into one shared unit that is rated like any other team.

How good is it

The model is checked by walking forward through past seasons a week at a time: fit the ratings on every game before that week, predict that week's games, then compare with what happened and with the closing point spread. The closing line is the hardest benchmark in the sport, and the ratings are not expected to beat it, only to stay close.

Beating the spread about half the time is the expected result for an honest model; anything much higher would mean the test was leaking future information. The early weeks miss by more than the late weeks, which is the price of leaning on the prior.

This season so far

The same test runs on 2026 after every data refresh: each week's games are projected from ratings fit only on the games before it, then scored once they are played. Through Wk 2, 100 FBS-vs-FBS games.

Early weeks lean on the preseason prior and miss by more; the season number should settle toward the historical one by midseason. A gap to the closing spread that keeps growing would mean something upstream broke, not that the model changed.

Game projections

On the current season's team pages, unplayed games show a projected spread and win probability. The spread is the rating difference plus the home edge, shown from the team's side with the usual convention: negative means favored. The win probability comes from how often a projected margin holds up in the backtest; actual margins scatter around the projection with a standard deviation of about 16 points, so a 7-point favorite wins roughly two-thirds of the time. The projected record is the current record plus the sum of those probabilities.

Upcoming games

The homepage strip and the Games page rank the coming week's games by a thrill score. It combines three things: the quality of the two teams (the sum of their Ratings, half the weight), how close the projection is (the favorite's win probability's distance from a coin flip, about a third), and tempo (the two teams' combined plays per game, the rest). Each is standardized against every FBS matchup of the season and the total is expressed as a percentile, so a 90 is a better game than nine in ten on the schedule. It says nothing about storylines, rivalries, or what is at stake; it is only what the numbers can see. The projected score is each team's adjusted offense against the other's adjusted defense, with half the home edge to each side, so the two scores always differ by the projected spread.

What it does not know

Injuries, suspensions, weather, and motivation are not modeled. Garbage time is not removed. Every game of the season counts the same regardless of how long ago it was played. FCS opponents are treated as one team, so a win over a strong FCS program is undervalued and a win over a weak one is overvalued. Treat the ratings as a well-calibrated starting point, not a betting sheet.