CFB Power Board

Ranking methodology

Scores, opponents, and venue. Fixed model settings. Every usable completed 2026 result contributes to one independent rating.

Reading the numbers

  • Rating is in points: the expected scoring margin against a perfectly average team on a neutral field. A team at +21.0 would be favored by about 21 over an average opponent — and by about 14 over a +7.0 team.
  • Strength of schedule is the weighted mean rating of every opponent a team has faced, on the same points scale.
  • Rank is within a team's own division. All four divisions are solved together on one scale, so cross-division ratings are directly comparable.

How ratings are calculated

Each game is treated as a single noisy measurement of the gap between two teams, and all 2,436 measurements are solved simultaneously for the set of team ratings that best explains them.

The model

If team A beat team B by 10 points, that is evidence that A is about 10 points better than B. Written down for every game at once, that is an over-determined system of equations, and the rating is the least-squares solution:

minimize  Σ w · (rating[home] − rating[away] − adjusted margin)²  +  λ · Σ rating²

The first term asks the ratings to reproduce the actual results. The second is a regularization term that pulls every team toward average. This is the classic Massey / Simple Rating System formulation, with a few adjustments described below.

Strength of schedule is not a separate ingredient. A team's rating is defined relative to its opponents' ratings, which are defined relative to their opponents' ratings, and so on through the whole season. Beating a good team therefore raises your rating more than beating a bad one, automatically — nothing has to be bolted on to make that happen, and adding an explicit opponent-quality multiplier on top would double-count it.

Adjustments

Home field

2.5 points are subtracted from the home team's margin before the game is used. This is a fixed model setting; neutral sites get no adjustment. A narrow home win provides less evidence of a power advantage than the same win on the road.

Margin of victory, with diminishing returns

Margins are passed through a saturating curve capped at 28 points. A 7-point win counts as roughly 7 points; a 63-point win counts as far less than nine times a 7-point win. Larger wins still count more, but each extra point contributes less evidence than the last.

Winning, on its own

A flat 2.5-point bonus is added for winning regardless of score, so a one-point win is meaningfully better than a one-point loss. Ties remain in the model, with no win bonus.

Recency

Older games decay by 3% per week, down to a floor of 60% of full weight. September still matters in December — teams change over a season, but not so much that early results become noise.

Games between divisions

An FBS team beating a Division II team is weak evidence about the FBS team, so those games carry reduced weight, scaled by how far apart the two divisions are. They are not discarded: they are the only thing connecting the divisions to each other, and without them the four divisions could not be placed on a common scale at all.

Teams with few results

The regularization term shrinks teams with few games toward the average. That limits the effect of a single result. Teams with no completed games are left unranked. Regularization also keeps the calculation stable when schedules barely connect to the rest of the country.

Rank history and movement

After every completed week, the season is solved again using only the games played through that week. That produces each team’s rank history and the ▲▼ movement shown in the tables. Earlier weeks are never revised with later information, so a team’s chart shows what the rating actually said at the time.

Projections and win probability

A projected margin is the rating gap between two teams, scaled to real scoring margins, plus a home-field term for games that are not at a neutral site. Both numbers are fitted to this season’s own results, and the spread of real results around the projection turns a margin into a win probability.

  • Scale: each point of rating gap is worth about 1.43 points of real margin.
  • Home field: about 3.6 points.
  • Uncertainty: real margins land within about 19 points of the projection two times in three.

Every projection is out of sample. A game is projected only with ratings solved from earlier weeks, and both teams need at least two prior games. This season, the favorite has won 1,219 of 1,699 projected games (72%), and projected margins have missed by 15.0 points on average. The model does not account for injuries, weather, or depth charts.

Résumé rank

The power rating asks how good is this team? Résumé rank asks how much has this team accomplished? For every game a team has played, we take the probability that the 25th-best team in its division would have won that same game at the same venue. Add those up and you get the wins a top-25 team would expect against that schedule. A team’s actual wins minus that benchmark is its wins above benchmark, and résumé rank orders teams by it.

A hard schedule lowers the benchmark, so a 7-2 team against strong opponents can out-rank a 9-0 team against weak ones. Losses are never rewarded, though: a loss counts zero no matter who it came against. Quality wins are wins over teams rated at or above that benchmark, and bad losses are losses a team was at least a 75% favorite to win.

Data and updates

Results come from College Football Data. Rankings are calculated from completed games. Opponent ratings in game logs reflect the current rankings. The full computed dataset is available as rankings.json.

Limitations

This is a résumé-and-margin rating, not a prediction market. It knows nothing about injuries, weather, personnel changes, or who was playing their backups in the fourth quarter. Early in the season, when few games connect the teams to one another, it is uncertain and should be read that way. It is published because it is transparent and reproducible, not because it is authoritative.