Projects · live

College Football Power Rankings

A college football model that ranks all 138 FBS teams, picks every game before kickoff and projects every team's record. It updates itself every few hours and gets graded against Vegas.

  • Why I built it · To build a power ranking I'd actually trust, and grade it in public against the betting line
  • Role · Solo build (with Claude as pair programmer)
  • Timeline · Oct 2026 – present
  • Stack · PythonNumPyGitHub ActionsEleventy

    Python and NumPy for the ratings, backtests and season simulations, fed by the CollegeFootballData API. A GitHub Action runs every three hours, locks picks before kickoff, grades finished games and pushes new data to this site, which rebuilds the rankings page and a page for every team.

View Live

The itch

Every power ranking I read is either a list of opinions or a number behind a paywall. I wanted one that shows its work: a rating anyone can check, picks locked before kickoff, and a public record of how it did. I'm also a Texas fan, so I wanted something that would tell me the truth about the Longhorns.

How it works

Each team starts the season from a preseason rating built from last season, recruiting, roster talent and coaching changes. Every game then moves the rating, using the final margin (big wins count more, with diminishing returns) and play-by-play efficiency (EPA per play and success rate, garbage time removed), adjusted for the opponent. Win probabilities come from the rating gap plus home field, and the rest of the season is simulated 10,000 times for projected records.

Does it work?

Tested on 2018 through 2025 (2020 skipped) without peeking ahead, from week 4 on it missed final margins by 12.5 points on average, against 12.1 for the Vegas closing line, and picked the winner 73.0% of the time, a hair more often than Vegas (72.8%). Its win probabilities are well calibrated: games it called at 65% were won 64.6% of the time. On 2022 through 2025, the first version missed by 12.8 and the current one (live from the week 8 edition of 2026) by 12.3. The biggest gains came from the preseason rating and from adding success rate, and every change had to beat the old model on seasons it wasn't tuned on.

What's live

  • Power rankings for every FBS team, with projections, filters and each week's picks
  • A page for every team, like Texas, with its schedule and win odds
  • Weekly power rankings posts with a game of the week, upset alerts and the Forty Acres Report on Texas