How it works

Each game is played out snap by snap, 10,000 times. Every play's result comes from what really happened in similar spots in past NFL games, adjusted for who is on the field: the offense, the defense, the quarterback and home field. Clock, fourth downs, field goals and special teams are all part of it.

From those runs: win chance is how often each team wins, the line is the middle result, and the total and score ranges come from the spread of outcomes. Player projections count the same simulated plays, so a receiver's number moves with his quarterback and with how the game tends to go.

Team strength on offense and defense, passing and running, adjusted for the opponents they faced and weighted toward recent games. Quarterbacks are rated on their own, so a quarterback change moves the numbers. Players listed Out are removed; questionable and doubtful players sit out in part of the simulations. Every input uses only games played before kickoff.

Data comes from nflverse, an open NFL data project. Market prices come from the closing sportsbook lines, Kalshi and Polymarket.

College numbers come from a simpler model than the NFL's. With 134 FBS teams on very different schedules, each team gets a strength rating fitted from the games played so far this season, starting from last season's rating and recruiting talent. Home field and a wide spread of outcomes (college has more blowouts) turn that into 10,000 simulated scores per game. There are no player projections for college.

Tested on the 2024 and 2025 seasons it never saw, it picked the winner 73.7% of the time; the betting favorite, 75.3%. Early in the season, before teams have played many games, it leans on last season and recruiting more, and disagrees with the market more. College data from CollegeFootballData.com.