Current Rankings
Rankings Comparison
Strength of record
Strength of schedule
Rankings visuals
Offense vs. defense
Offense runs left to right (higher is better). Defense runs bottom to top, and lower is better, so the best teams sit in the bottom-right corner. Dashed diagonals join teams with the same overall rating.
Rating by week, top 25
Rank by week
The top teams in the selected week (dashed line) and their rank entering every week. Teams that were outside the group sit on the bottom edge. Hover a line or a name to pick out one team.
Where each rating sits
Each team's preseason rating, what this season's games alone say, and the current rating that blends the two.
Overall rating with 95% certainty
The dot is the rating; the bar is the range the true rating falls in 95% of the time.
Strength of record
Conference strength
Every team in its conference's row. The faded band covers the whole conference, the darker bar is its middle half and the black tick its median, so you see both how strong a league is at the top and how deep it is. Sorted by median. Hover a conference to box in its teams.
Best game performances
The top 100 single games by net: how many points better than an average team the team played, against that opponent at that site. Each row runs from the opponent's rating (grey) to the team's rating for that game (coloured: green won, red lost), which is the opponent's rating plus the final margin, less home field when the team was at home (plus it when away). Uses the selected week's ratings.
Best offensive performances
The top 100 games by offense: points scored (coloured: green won, red lost) against what an average team would have scored against that defense at that site (grey).
Best defensive performances
The top 100 games by defense: points allowed (coloured: green won, red lost) against what an average team would have allowed against that offense at that site (grey). Fewer allowed is better.
Matchup
Uses the ratings as of the selected week.
How accurate has it been?
Every game was predicted using only the games played before it.
Confidence check
When the model says 70%, those teams should win about 70% of the time.
By week
Average miss: ExTAN vs. the Vegas spread
Predicted margin vs. actual margin
Predicted margin vs. the Vegas spread
Over/under: predicted total vs. actual total
Preseason expected wins vs. actual regular-season wins
Ratings in one week vs. another
Compare any two weeks' ratings, for example how well an early week's ratings anticipated where teams finished.
Games
Each game with the prediction that was made before it was played. Click a game to open it in the Matchup tab.
Methodology
How ExTAN ratings are built, what they produce, and where they fall short.
What the model produces
A team's rating is meant to be predictive of future results, so everything here is built to forecast games. The Accuracy tab tracks how well it does against results and betting lines, and whether its win chances hold up: teams given a 75% chance to win a game win about 75% of the time.
Every team gets two ratings. Offense is the points a team scores above or below what an average FBS team would score against the same opponents. Defense is the points it allows above or below average, so lower is better: a defense rated −12 gives up twelve fewer points than an average team would. Technically, "offense" is really points scored and "defense" is really points allowed, so any scoring by a team's defense or special teams is included in its offense rating.
The Expected Team Average Net (ExTAN) rating is the team's overall rating, and is the offense rating minus the defense rating: how many points the team would be favored by against an average FBS team on a neutral field. Zero is exactly average for FBS. FCS teams sit well below zero on the same scale, so the two divisions can be compared directly. Subtract two teams' ExTAN ratings to get the expected margin between them at a neutral site, then add home-field advantage (2.9 points for an FBS home team) for the home side.
From these come rankings for FBS and FCS, a projected score, margin and win probability for any matchup, projected season win totals with the chance of every possible record, a strength-of-record ranking, and week-by-week accuracy tracking. Only game scores and game locations are used. No polls, no reputation, no manual adjustment.
The question behind every rating
Given who you played and where, how many points would an average team have scored and allowed here, and how did you do against that?
A 24-20 road win over a top-15 team is probably more impressive than a 45-10 home win over a bad team. The final score alone doesn't tell the whole story, but adjusting for opponent strength does. We do this by comparing each result to what an average team would have been expected to score and allow in the game. A team's rating is the average of those comparisons across its season, with the adjustments below.
Rating everyone at once
Judging a win over Team X requires knowing how good X is, which requires knowing how good X's opponents were, which includes the team being judged. So the whole league is rated together. The model starts with rough estimates, then sweeps through every team repeatedly, each pass using better information than the last, until the numbers stop moving — usually a few dozen passes. Strength of schedule isn't a separate correction added at the end; it's how the ratings are built.
Trusting results in proportion to the evidence
Twelve games is a short season, so each rating blends what a team's own results imply with a more conservative fallback:
rating = (games × what the results say + k × the fallback) ÷ (games + k) k = how much this team varies game to game ÷ how spread out ratings are in its division
A team whose performances swing wildly is pulled toward the fallback harder, because any single game
from it means less. When a division's ratings are bunched tightly, extreme results are more likely to be
noise, so everything is pulled harder. k typically runs between 1 and 4 games, and is
usually higher for defense, where ratings are bunched more tightly across the league.
End-of-season ratings
Once a season is complete, the final rating comes from game results alone: no preseason estimate and no pull toward the division average. Each team's offense is the plain average of what it scored above or below what an average team would have scored in each of its games, and the same for defense, so every game counts equally and the final rating is exactly the sum of its games.
For next season's starting point the model uses a second, carry-forward version of the same season that is still regressed toward the division average. Even twelve or thirteen games overstate how far a team really sits from average, so the regressed version predicts next season better, and it is the version the preseason settings were fitted on.
Divisions
All of Division I is rated in one pool, and FBS-vs-FCS games are used in full. Three parts of the fit are worked out within each division: how hard to shrink ratings, what point they're shrunk toward, and what counts as typical week-to-week variation. The spread across all of D-I is much wider than within FBS, so a combined figure would under-shrink FBS teams and pull them toward a point below the average FBS team.
The zero point
The scale is anchored so that 0 means an average FBS team. After each pass of the fit, the average FBS offense and defense are subtracted from every rating, with the scoring baseline shifted to match. Every predicted score and margin stays identical; only the labels move.
The preseason estimate
Week 1 has no results, so ratings start from an estimate built from last season, conference and recruiting:
preseason = 1.39 × conference average + 0.64 × (team − conference average) + talent + returning production − coaching change
"Team" and "conference average" are last season's carry-forward ratings, grouped by this season's conferences. Conference strength carries over from year to year, so the estimate starts from the conference's level rather than from an average FBS team; how far a team sat above or below its conference partly regresses. FBS independents and FCS teams start from their division's average instead. The talent bonus is 1.8 points of overall rating per standard deviation of the 247Sports composite, split evenly between offense and defense. Two roster terms follow the same split: returning production (the share of last season's production coming back, as a z-score among FBS teams) is worth 2.0 points of overall rating per standard deviation, and a new head coach costs 2.0 points. Where that data doesn't exist (returning production before 2014, talent before 2015) the term is zero. The weights were chosen on 2022–2025 as the values that best predict each team's end-of-season rating; the roster terms were tested on seasons from 2015 on, leaving out 2020.
The estimate is the fallback in the blending formula above, and during the season it counts as twice the
usual k games' worth. Its influence shrinks automatically as games accumulate. From Week 9 it is
also phased out team by team: its weight drops by a quarter with each game a team plays after its 8th, so
it still counts a little when predicting a team's 12th game and is gone once that game is played.
Published postseason ratings use this season's games only; bowl and playoff predictions also keep the
model's normal pull toward the division average, which predicted bowls better in testing than treating a
games-only rating as exact. Early-season rankings lean on preseason expectations,
more so on defense.
Home field
2.9 points for an FBS home team, 2.2 for FCS, in every non-neutral game. Measured across conference games from 2004 through 2025, the real spread between programs is about 0.6 points against roughly 1.6 points of measurement noise per team — too small to measure reliably — so one value per division is used. It's split evenly: the home team's expected score goes up by half and the visitor's drops by half, so the full value lands on the margin and the predicted total is unaffected.
Consistency
Separately from how good a team is, the model tracks how much it varies. It's measured from how far each game landed from what the ratings predicted. Games count in proportion to how little they shaped their own ratings (a team with one game played fits that game almost exactly, so it says nothing about consistency yet), and each team's figure is pulled toward its division's typical value, weighted by how many effective games support it.
Confidence in the ratings
A team seen three times carries more uncertainty than one seen eleven times, even if both are equally steady:
uncertainty = √( consistency ÷ (games + k) )
Predicting a game
home score = league average + home offense + away defense + HFA/2 away score = league average + away offense + home defense − HFA/2
The spread around each score combines game-to-game variation (the average of the scoring offense's and opposing defense's consistency, since both are measured from the same scores) with the uncertainty in both ratings. The win probability is the share of the margin distribution above zero. A projected margin and the confidence in it are separate: the model can like a team and still call the game close to a coin flip.
Projected win totals
Wins already banked are combined with the chance of winning each remaining regular-season game, one game at a time, giving the probability of every possible final record. Remaining games are treated as independent, which makes the range of outcomes slightly narrower than reality.
Strength of record
A hypothetical team with the average ratings of the current top 12 is run through a team's exact schedule — same opponents, same sites — and the model calculates the chance it finishes with a record at least as good. Lower means a more impressive resume. Margin of victory plays no part.
Strength of schedule
For the schedule a team has played so far, the model asks how many games a team rated like the #1, #10, #25, #40 and #70 teams would be expected to win against those same opponents at the same sites. A tougher schedule means fewer expected wins for every benchmark.
Weeks and the postseason
Regular-season weeks keep their numbers and the postseason continues the count: the first postseason week covers all bowls plus the first CFP round, and each later CFP round gets its own week. FCS playoff games are placed into whichever FBS week contains their date.
How accuracy is measured
Every week is predicted using only games played before it, which matches what would have been known at the time. Four checks run on those predictions: correctness (winners picked and margin miss), calibration (when the model says 70%, those teams should win about 70% of the time), spread width (misses divided by the predicted spread should have a standard deviation of 1), and against the market (Bovada spreads and totals).
Typical performance, FBS vs FBS, weeks 6 and later:
| Straight-up winners | 70–71% |
| Average absolute error | 12.7 points |
| Bias | +0.5 points |
| Standardized miss, standard deviation | 1.01 |
| Against the spread | 50% |
| Over/under | 48.5% |
What the model doesn't do
Personnel. Injuries, suspensions and bowl opt-outs are invisible to it. Bowl games deserve the most skepticism for this reason.
Coaching effects. A coach's strategy and in-game decisions aren't modeled. The only coaching input is whether the head coach changed since last season, which lowers the preseason estimate; after that a team's rating reflects only its results.
Individualized home-field advantages. Every team gets the same home-field advantage (2.9 points for FBS, 2.2 for FCS). A particularly tough place to play isn't credited with any extra, because the differences between programs were too small to measure reliably.
Situational factors. Weather, rest, travel and altitude are not modeled.
Settings
| Setting | Value | Effect |
|---|---|---|
| Conference carry | 1.39 | Share of the conference's level carried into the preseason estimate |
| Blend weight | 0.64 | Share of a team's distance from its conference average carried over |
| Talent weight | 0.6 × 3.0 | 1.8 points of overall rating per standard deviation of talent |
| Returning production | 2.0 per SD | Points of overall rating per standard deviation of returning production |
| New head coach | −2.0 | Points of overall rating lost when the head coach changed |
| Preseason trust | 2 × k | In-season, the preseason estimate counts as twice the usual games' worth |
| Preseason fade-out | Week 9 → game 12 | Weight shrinks with each game after a team's 8th; gone after its 12th |
| Home field | 2.9 FBS / 2.2 FCS | Split evenly between the two scores |
| Shrinkage by division | on | Shrinkage rules calculated within each division |
| Zero point | average FBS team | What a rating of 0 means |
The preseason settings were chosen on 2022–2025 and the roster terms were tested on 2015–2025 without 2020. The ratings and the season selector now go back to 2002; seasons before 2022 are an out-of-sample look at the model, and 2020, a short disrupted season, is shown but was never used to choose or test a setting.