What CFPi+ measures, the football reasoning behind it, and where the numbers on this site come from.
The College Football Power Index (CFPi+) is the model's estimate of a team's strength relative to the FBS field. It is a predictive rating, not a record of what a team has earned so far. It starts from a proprietary preseason rating and moves toward current-season, opponent-adjusted performance as games are played. It does not depend on any outside power rating to update.
It is expressed as a projected mean point differential against an average FBS team on a neutral field. A CFPi+ of +7.4 means that team projects to beat an average FBS opponent by roughly a touchdown. An average FBS team sits at approximately 0; positive ratings are above average, negative ratings below.
The scale is standardised across the full 138-team FBS universe, never a partial field. That is what makes a gap in CFPi+ mean the same thing no matter which two teams you compare.
Talent and coaching are inside the Internal Prior: they explain where a team starts, and they matter most before it has played enough games to speak for itself. As the season goes on they recede in favour of what the team has actually done.
A note on how the site groups things: the TPi+ Talent Index pages present roster and coaching together because that is the useful way to read overall program quality. They are a reference view, not an extra term added on top of the rating.
CFPi+ is deliberately dynamic. A team's rating starts at its Internal Prior, which counts as a small number of games' worth of evidence. Every completed FBS game adds current-season evidence, so the more a team has played, the more its rating reflects what it has actually shown and the less it reflects the preseason view. The shift follows games played, not the calendar, so a bye week does not move a team.
Opponent adjustment is phased in on the same logic: a schedule cannot be judged until enough of it has been played, so the adjustment strengthens over several weeks rather than switching on at once. A team producing strong numbers against elite competition is not read the same way as a team producing them against one of the weakest schedules in the country.
That matters most for margin, not just for picking a winner. Current-season opponent-adjusted efficiency can reflect a breakout player, a rebuilt line, a coaching change or an injury in a way a preseason projection cannot, which is exactly the kind of information that separates two teams by more or less than the market expects. The transition is gradual by design, to avoid overreacting to a small sample. The exact weights are not published.
Weeks 0 through 5 of 2026 were published on Version 1, which blended a preseason national consensus with current-season data. CFPi+ 2026 v2.0 began with Week 6. Its in-season rating is built entirely in-house, so a delay in an outside rating update can no longer hold up a forecast. SP+ and FPI are preseason information only.
Every forecast published before Week 6 is preserved exactly as it was published. The National Composite is still shown beside the rating as an outside reference. It refreshes Monday evening once those ratings post, and it is not an input to Version 2.0.
CFPi+ also carries offensive and defensive unit ratings, so the site can show where a team's rating comes from instead of publishing one number and asking you to trust it.
A higher Off CFPi+ is a stronger offense. Def CFPi+ is a burden — expected points allowed against an average FBS offense — so lower is better. Public ranks always read #1 = best on both sides of the ball.
The two unit ratings are held mathematically consistent with the overall rating: the separation between a team's offensive and defensive ratings is its CFPi+. The units can never disagree with the top line.
A matchup begins with the relative strength of the two teams on the CFPi+ scale. The model then applies its calibrated venue-specific home-field adjustment, producing an expected scoring margin. Neutral-site games receive no home-field adjustment.
That home-field value is not a single league-wide constant. It is calibrated from each venue's own historical home-field record and scales with the quality of the matchup being played in it. It does not incorporate travel distance, weather, rivalry status or crowd-noise measurement — see Supported, Not Speculated below.
CFPi+ adjusts home-field advantage based on the local kickoff time, reflecting historically observed differences in home-field impact across early, afternoon, evening and night games.
The expected margin is then converted into a win probability using the model's measured game-outcome variance. The projected total comes from a separate scoring-environment model built on drive and possession data, and the exact projected score is the total split by the margin, so the two always reconcile. Games against FCS opponents use their own dedicated model.
One distinction is worth carrying: the exact projection is the underlying continuous scoring expectation, while the projected score is a plausible football representation of it. A projected 27–24 does not mean the model believes one exact final score is waiting to happen. It is the continuous projection rendered as a score a football game could actually produce.
A 70% win probability does not mean CFPi+ is “70% certain” in the conversational sense, and it does not mean the favorite is supposed to win. It means that given the model's estimate of matchup strength and the uncertainty around it, that team wins approximately 70 out of 100 comparable simulated outcomes.
The other thirty are the other result. Upsets inside that band are not the model failing; they are the distribution behaving exactly as described. Every forecast on this site should be read that way.
Individual game probabilities feed the season model, which produces expected wins, projected records and a 95% predictive season range for every team — the band of win totals containing 95% of that team's simulated seasons.
These are distributions of possible seasons rather than single deterministic forecasts. A team projected at 7.4 wins with a 95% predictive range of 5 to 10 is telling you something a flat “7–5” never could. The simulation also accounts for the fact that a season's outcomes are correlated — a team that is better than the model thinks tends to be better in every remaining game, not in one of them at random.
Conference-race and postseason boards are built on this same season model. They are listed under What's on the Site below and are not published yet.
Raw advanced stats are what actually happened on the field, before any opponent adjustment. Adjusted advanced stats are those same numbers read in the context of the opponents a team actually faced.
Raw numbers describe production. Adjusted numbers provide context. Both are published, because a gaudy raw number against a soft schedule and a modest raw number against a brutal one are different football facts and deserve to be visible as such.
The site also publishes the model-used layer — the form the model itself consumes at the current point in the season — so the drilldown matches the rating rather than approximating it.
If we can't support it, we don't print it. This page describes the model as it operates today, not as it might operate later.
CFPi+ does not currently use live weather, wind or temperature; travel distance or rest; rivalry designation; crowd-noise or decibel measurement; altitude; or live venue conditions. Several of those were tested against historical results and did not earn a place in the model — one of them came back significant with the opposite sign to the hypothesis, which is a good reason to reject a variable rather than a good reason to ship it. Others were never run at all. Either way they are not in the production calculation, so they are not claimed here.
Betting markets are never an input. Market lines are shown only so the model's number can be compared against them, and they are locked each Monday at 6 PM ET. Version 2.0 ratings are published as the model produces them, with no manual overrides.
If any of these inputs is genuinely implemented later, this page will describe it after it ships.
It is a predictive power model, built to estimate current team strength, matchup margins and outcome probabilities from a proprietary preseason prior, current-season opponent-adjusted efficiency and a dynamic home-field adjustment.