NO BLACK BOX.
A résumé ranking from completed games in the current season. No previous-season team rating, preseason poll, recruiting class, betting line or AP rank enters a score.
ARCHIVED FORMULA: v2.0.001 / WHAT GOES IN.
Every completed FBS game through the stated week, including games against FCS opponents. Scores are calculated for every FBS team. Only current-season results, opponents, location and capped margin are used. AI writes explanations after the rankings are calculated; it cannot choose a rank.
02 / THE FIVE COMPONENTS.
COMPUTER_SCORE = 0.50 × SOR + 0.20 × QUALITY_WINS
+ 0.20 × SOS + 0.05 × DOMINANCE
+ 0.05 × LOSS_QUALITYEach component is scaled to 0–100. Full precision determines order; displayed scores are rounded. Wins, losses and road results remain visible in the résumé, but no longer get separate score weights because those facts already affect the components above.
03 / OPPONENT QUALITY AND SCHEDULE.
A Colley matrix propagates win/loss evidence through the current season's FBS-versus-FBS game network. For team i, the diagonal is 2 plus its number of FBS games; each meeting against j adds −1 to the off-diagonal. The right side is 1 plus half its wins-minus-losses. Solving the system gives a 0–1 rating. All teams start with the same neutral 0.5; no previous-season team rating is imported.
(2 + FBS_games_i) × rating_i − Σ meetings_ij × rating_j = 1 + (FBS_wins_i − FBS_losses_i) / 2
When scoring a team, we solve the matrix again without its games, then read each opponent's rating. This stops the evaluated team's own result from increasing its opponent's quality. FCS opponents receive quality zero. SOS is the arithmetic mean of these opponent qualities across all completed games; FCS games therefore lower it. SOS_SCORE is the FBS cohort's midrank percentile of this raw value.
04 / QUALITY WINS.
FBS_WIN_VALUE = 0.25 + 2 × opponentQuality²
+ locationBonus + min(winMargin,21)/21 × 0.20
Home: +0.00 · Neutral: +0.10 · Road: +0.20
FCS_WIN_VALUE = 0.10 + min(winMargin,21)/21 × 0.05
QUALITY_WINS = percentile(Σ winValue / gamesPlayed)Squaring opponent quality gives elite wins more credit and reduces the base credit for routine wins. An FCS win is worth at most 0.15 résumé units in this component. It still counts in the official record and in strength of record.
05 / STRENGTH OF RECORD.
The benchmark is the 25th-highest current-season Colley rating among FBS teams. We estimate its chance to beat each opponent, then calculate the exact chance of matching or exceeding the evaluated team's win total against that schedule. A harder earned record has a higher raw SOR.
p(FBS game) = sigmoid(0.02443427 + 3.97705589 × (benchmarkRating − opponentQuality) + 0.38435732 × site) site = +1 home, 0 neutral, −1 road p(FCS home) = 0.96260388; neutral/road subtract one/two home coefficients on the log-odds scale. SOR_RAW = 1 − P(benchmark wins at least actualWins) SOR_SCORE = percentile(SOR_RAW)
The FBS coefficients were fit to 2,264 FBS games from 2023–2025 using each team's rating before that game week, then checked with each season held out in turn. The FCS home probability is the smoothed observed rate from 360 FBS-home/FCS games. Neutral and road FCS probabilities extend it using the fitted home effect; those venue cases are sparse and less certain. The final record probability is computed exactly with a Poisson-binomial dynamic program, with no Monte Carlo randomness.
06 / MARGIN AND LOSSES.
DOMINANCE_RAW = mean(clamp(signedMargin,−21,21)) + (SOS_RAW − FBS_AVG_SOS) × 10 DOMINANCE_SCORE = percentile(DOMINANCE_RAW) LOSS_PENALTY = 1 + (1−opponentQuality)×0.75 + locationPenalty + min(lossMargin,21)/21×0.25 LOSS_RAW = 1/(1 + Σ lossPenalty) LOSS_SCORE = percentile(LOSS_RAW)
Home loss penalty is 0.20, neutral 0.10 and road 0. The margin cap is 21 in every component. Undefeated teams receive 100 for loss quality.
07 / TIES AND REPRODUCIBILITY.
A score gap above 1.0 point takes precedence. Within 1.0, the head-to-head winner gets priority. Split series cancel. Remaining order follows strength of record, quality wins, SOS, road diagnostic, fewer losses, dominance, then stable name and team ID. A graph resolves circular head-to-head results deterministically.
Percentiles use midranks among FBS teams with completed games. A single or all-equal cohort scores 50. Teams with no completed games score zero and appear last. Weekly snapshots preserve the version, all input games, all component values and a SHA-256 input hash. Published snapshots cannot be edited or deleted. The public JSON API exposes the inputs and scores; AP ranks are overlaid afterward for comparison.
Version history
v1.0.0 was the first published formula. Its archived methodology and original snapshot remain available. v2.0.0 adds connected-schedule opponent ratings, a historically calibrated top-25 strength-of-record benchmark, convex quality wins and lower duplicate record and road credit. Historical results and the change review are documented in the project repository.
Back to the rankings →