No voters. No committee. Four math models, published weights, and every FBS game played this season.
Each component column shows that model's own rank. Δ CFP compares the computer to the committee: ▲ the computer ranks a team higher, ▼ lower. T25 W counts wins over the computer's top 25; Bad L counts losses to teams outside its top 75 or to FCS teams. Luck is wins minus the wins the power rating expected; a big plus means close games broke their way. Click any team for its game log, the math, and a team comparison.
| # | Team | Rec | CFP | Δ CFP | AP |
|---|
The field is picked by the computer's current rankings and seeded straight down the list. Byes go to the top seeds when the field doesn't fill a full bracket. Title odds come from 10,000 simulated tournaments using each team's power rating on a neutral field.
Wins and losses only. Every team starts at .500 and gets credit for beating good teams and less credit for beating bad ones, solved for all 138 teams at once. Score margin is ignored. This is the same method Wes Colley ran as one of the original BCS computers.
How many points better than an average FBS team, on a neutral field. Fit by least squares across every game, adjusted for opponent and home field. Each game's margin is capped at 24 points, so running up the score on a cupcake earns nothing extra. Recent games count more: a week 1 game counts 60% as much as last week's. The fitted home-field edge this season is points.
Rewards teams that move the ball and stop it, regardless of the final score. Four stats, each adjusted for the opponent and weighted toward recent games, offense minus defense:
Points per drive (40%): how often a possession turns into points. Yards per play (30%): how hard the team is to stop. Field position (15%): where drives start, which captures special teams and turnovers. Third-down rate (15%): staying on the field. A team that dominates the box score but loses close games will look better here than its record.
Asks how a top-25-caliber team would have done against your exact schedule, then credits you for the wins beyond that. Going 8–0 against a soft schedule is impressive. Going 8–0 against a brutal one is more impressive, and this is the column that shows the difference.
Each model is converted to a z-score (standard deviations from the FBS average), then blended with the weights at the top of the Rankings tab. The default is 25% each: half of the formula asks "who has the best record?" and half asks "who is actually the best team?" Move the sliders to build your own formula. The Old BCS preset uses résumé only, like the margin-blind BCS computers did after 2002.
When two teams' blended scores are within 0.15 of each other and one beat the other, the winner moves directly above the loser. The move is skipped if the winner also lost to a team it would jump. That keeps loops (A beat B, B beat C, C beat A) from forcing an arbitrary order. Teams moved this way carry an H2H tag. The tiebreaker is on by default and can be turned off under the weights.
Top-25 wins and bad losses are counted against the computer's own current rankings, so they update when you move the sliders. A bad loss is a loss to a team ranked outside the top 75, or to an FCS team.
Luck is actual wins minus expected wins. Expected wins add up each game's win probability from the power rating. A team at +2 has won two more games than its play suggests, usually through close games and turnovers. Luck isn't scored, because a win is a win, but it flags who is likely to regress.