What gets published here
Three kinds of work. Ratings: Elo, Markov chain and Massey ratings for every NFL team, recomputed from the results after each week of the season. Models, explained: long-form guides to rating systems, win probability, expected points and the statistics underneath them, each with the math written out and code that can be rerun. Tools: calculators and stepped explainers that run in the browser and show every constant they use.
The site does not publish picks or betting advice. Point spreads appear as what they are to an analyst: a public forecast of the margin, useful as a benchmark and as an input to other models.
Where the data comes from
Game results and play-by-play come from the open nflverse data releases, published under the Creative Commons Attribution 4.0 license. College football schedules come from the sportsdataverse releases. Every rating and model on the site can be rebuilt from those files.
How the ratings are computed
The methodology page lists every formula and constant. The backtest shows how the Elo system has predicted games since 1999, season by season, including the seasons where it did poorly.
Who writes it
Miles Okafor builds rating systems and win-probability models from public play-by-play data and writes about the statistics behind them: how the models work, where they fail, and how to check them yourself.
Corrections
When a number is wrong, it is fixed in place and the article’s updated date changes. Model changes are recorded on the methodology page with the date they took effect.