Football · The Panna Project
Football
Predictive ratings, match picks and league projections across 15 European competitions. The ratings come from Panna, an open-source RAPM+SPM blend that nets xG created against xG conceded for every player on the same scale.
How Panna works
Traditional football stats tell you what a player did. Panna ratings try to answer a harder question: how much did it actually matter?
The core idea is plus-minus: when a player is on the pitch, does their team create more and concede less than expected? Raw plus-minus is noisy (your numbers look great if you play alongside Messi), so the model controls for the quality of every other player on the field. Think of it as isolating your contribution from everyone else’s.
For players without enough minutes to get a reliable plus-minus estimate, the model falls back on box-score stats like key passes, interceptions and progressive carries, which tend to predict plus-minus well. The final Panna rating blends both signals, leaning more on the box score for bit-part players and more on plus-minus for regulars.
The rating splits into offense and defense. Positive offense means you’re helping create goals. Negative defense means you’re preventing them. Add them together and you get the overall picture.