Methodology

How our numbers are made

Win probability

The live win probability on an NFL or college football game page is our own in-game model, one per sport. It reads the state of the game — the score, the clock, down and distance, field position and who has the ball — and is anchored on the pre-game market spread, taken once at kickoff. A game with no pre-game line gets no number.

The college football number is labelled Calibratinguntil our own 2026 results validate it. Every live reading is kept, so each model can be checked against how the games actually ended. It is context for following a game, not advice: we explain, we don’t pick.

NFL power ranking

Our NFL power ranking uses this season’s games only — nothing carries over from last year — and is recomputed after each week’s last game, once Monday night is final. Early weeks are small samples, and the ranking says so.

It combines four measures of each team’s season so far:

• Scoring, adjusted (25%). Points per game against what those opponents usually allow. Scoring 30 on a defence that usually allows 20 counts as +10.
• Defence, adjusted (12%). Points allowed per game against what those opponents usually score, turned so that higher is better.
• Win percentage (55%). Wins over games played; a tie counts as half a win.
• Quality of record (8%). Each result against what a typical team would expect against that opponent. Beating a team that wins 80% of its games is worth +0.8; losing to it costs only 0.2; losing to a team that wins 20% of its games costs 0.8.

Every opponent figure leaves out the opponent’s game against the team being rated, so no team helps rate itself. A game against an opponent with no other games yet is left out of the three adjusted measures, but still counts toward win percentage. Each measure is put on the same scale — how many standard deviations a team sits from the league average — and the four are combined with the weights above. A power of 0 is a league-average team.