Sliding Scale

Methodology

Two models, one slider: how each team's score is built, how accurate it is, and what it leaves out.

Last updated: Tue, 17 Nov 2026 05:43:20 GMT

2026 (out of sample)
12.0
pts avg miss · 70% winners
2023–2025
12.4
pts avg miss · 72% winners
2016–2022
12.9
pts avg miss · 72% winners

Two questions, one slider

Every FBS team gets two scores from our own models. Most deserving asks what a team's results have earned. Best team asks who would win the next game. The slider mixes them: score = s × best + (1 − s) × deserving, after putting both on the same scale (z-scores across FBS). Slide to 0 for pure deserving, 100 for pure best team. The value is in the URL so you can share a ranking exactly as you see it.

The two ends are the cleanest answers. Points in between are a judgment call about how much quality should count alongside results; there is no single right mix.

Most deserving (resume model)

  • We rate every team's strength from this season's games only (no preseason expectations), using scoring margins capped at 21 points and a fitted home-field advantage.
  • For each game, we estimate the chance that a benchmark team, the 12th-best team in the country (the playoff cut line), would have won it at the same site. Opponents are rated without the team's own games against them, so beating a team badly doesn't make that win look less impressive.
  • Each game earns credit = result (1 for a win, 0 for a loss) minus that chance. Beating a top team on the road earns a lot; losing to a weak team costs a lot. A team's own margin never matters: a win is a win.
  • The resume score is the sum of a team's credits.
  • Top-12 odds and the likely rank range come from 1,000 simulations of the uncertainty in opponent ratings.

Best team (predictive model)

  • A power rating: how many points better a team is than an average FBS team on a neutral field, adjusted for opponents and home field.
  • Game margins get full credit up to 40 points and half credit beyond. Each game's margin is blended 85% with the final score and 15% with the success-rate gap, which filters out some luck.
  • A preseason starting point from last season's FPI, returning production and recruiting talent. It counts as six games of evidence before the season and halves every six games a team plays (about 2.4 games after eight games), so by November this season's results carry most of the weight. We tested this against a fixed weight: it predicts as well or slightly better, and leaning less on reputation late in the season is fairer. Dropping the starting point entirely makes predictions clearly worse, even late in the season, so it stays. The team details show each team's rank from this season's results alone.
  • Offense and defense ratings split the rating using points scored and allowed.
  • Tested by predicting each week's FBS-vs-FBS games using only earlier games (week 2 on): a 12.0-point average miss and 70% of winners picked on 2026 games so far; 12.4 points and 72% across 2023–2025; 12.9 points and 72% across 2016–2022. The settings were tuned on 2016–2025, so those seasons flatter the model a little; 2026 was not used for tuning.

Schedule strength

Schedule strength is how often the benchmark team (the 12th-best team, the same one the resume model uses) would win a team's games so far, at the same sites. Lower means harder. Opponents are rated by the best-team model without any of the team's own games. We judge schedules for a contender rather than an average team because what separates schedules for a playoff case is how many strong opponents are on them.

Other ratings

Team details also show where each team ranks in SP+, FPI, Elo and SRS for reference. They don't feed the slider. Only Elo is available week by week for past seasons; SP+, FPI and SRS are end-of-season values for historical weeks (look-ahead), and we snapshot them weekly going forward.

Blind ranking

On the blind ranking page you rank three anonymous teams at a time. Each card shows record, average scoring margin, offense and defense efficiency (EPA per play, ranked among FBS teams), postgame win probability (CollegeFootballData's estimate of how often a team would be expected to win with a similar underlying performance, based on efficiency measures rather than the final score, averaged over its games), and its record against each opponent quadrant. We fit a Plackett-Luce model to your rankings: it learns how much you weight each stat, plus a small per-team adjustment, then scores every FBS team. Until your own rounds say otherwise it assumes you like every stat a little, and it picks the next cards to be the ones that tell it the most about what you value. That is why about 15 rounds give a usable ranking, and more rounds sharpen it. Everyone's votes also feed a crowd ranking that resets every Monday. To keep one visitor from outvoting everyone, each browser session counts as at most 40 votes: beyond that its votes are scaled down, and the 200-vote threshold counts votes the same way. Our own model ranks are deliberately kept off the cards, so the ranking reflects your judgment rather than ours.

A G6-only switch limits the game to Group of 6 teams (American, Conference USA, MAC, Mountain West, Sun Belt and the Pac-12, plus UConn; the same group as the rankings table's G6 toggle). In that mode the record, average margin and postgame win probability on each card count only games against other G6 teams; every other stat still covers all games. G6 rounds are saved separately, build their own G6 ranking after 15 rounds, and stay in your browser: they don't feed the crowd ranking, because they were judged on a different record.

Your results also show your blend: where your picks sit on the rankings slider. For every slider setting from 0 (most deserving) to 100 (best team) we ask how well that blend of our two models explains the order you actually put each three teams in, and report the setting that explains it best, with the range of settings your rounds can't tell apart from it (about 95% confidence). It uses your raw rounds, not your fitted ranking, because the fitted ranking leans on default stat weights early on. The two models agree on most teams, so the range is wide after 15 rounds and narrows as you play. The crowd's blend is the same calculation over this week's votes, with each session capped as for the crowd ranking.

Quadrants group opponents by our best-team rating, and the bands widen toward the bottom, like college basketball's NET quadrants. For a home game, Q1 is an opponent ranked 1–15, Q2 16–40, Q3 41–75 and Q4 76 or lower. Neutral-site and road games reach further down, by the model's home-field advantage: typically Q1 goes to about #21 on a neutral field and #33 on the road (Q2 about #52 and #63, Q3 about #85 and #100), varying a little each season with the fitted home field. In a typical season a top-12 team wins about 62% of Q1 home games. FCS opponents are placed by their own rating (nearly always Q4). Opponents are rated without any of the team's own games (as for schedule strength on the stats page), so blowing a team out can't drop it a quadrant and make the win look worse. Tied games count in no quadrant. The voter model uses two quadrant stats: Q1 wins and Q3/Q4 losses.

Fine print

A team's best win and worst loss are the best team it beat and the worst team it lost to, judged by our best-team ratings (FCS opponents are rated from how they actually performed against FBS teams). Ratings are only good to a few points, so opponents within 4 rating points of the best (or worst) are treated as tied and their records settle it: the better record is the best win, the worse record the worst loss. Where or how close the game was doesn't change which one is picked. FCS teams are not shown, though games against them count. Each FCS opponent is rated individually from its games against FBS and FCS teams, pulled toward the average FCS team when it has few games, so beating a strong FCS team counts for more than beating a weak one. Historical weeks are recomputed with only the games played through that week. The best-team model's starting point uses last season's final FPI, which is known before the season starts.

Attribution

Data from CollegeFootballData.com, including game results, advanced stats, returning production and 247Sports team talent. SP+ is created by Bill Connelly, FPI by ESPN; Elo and SRS are computed by CFBD. Team logos come from CollegeFootballData.com and are trademarks of their schools.