Open data

Frozen predictions, as CSV

Two files, rebuilt from the same tables the site reads, published under CC BY 4.0. Use them in a notebook, a paper or a post; cite RaceModel.io.

predictions.csv

Every RaceModel.io prediction frozen before a race across six motorsport series: predicted finishing position, podium and DNF probability per driver, model version, freeze time and the race's post-race grade.

286 rows. Updated after every freeze and every race.

Download predictions.csv →

backtest-f1.csv

RaceModel.io's leave-one-race-out Formula 1 backtest for the current season, per driver per race: predicted position, DNF probability and predicted points, with each race's mean absolute error and podium hits.

242 rows, current season.

Download backtest-f1.csv →

backtest-f1-races.csv

Race-level grades for every Formula 1 race in RaceModel.io's multi-season leave-one-race-out backtest: mean absolute finishing-position error and podium hits per race. The file behind the accuracy ledger.

100 races. The file behind /accuracy.

Download backtest-f1-races.csv →

What is in the files, and what is not

Both files contain the model's outputs and how they were graded: nothing about how the model works. There are no feature values, no feature names, no per-driver attributions and no calibration internals. A prediction row is what the race page showed before the race; a grade is what the accuracy ledger shows after it.

Every row in predictions.csv was frozen before its race started (see frozen_at) and is never edited afterwards. A cited prediction cannot have been changed once the result was known.

predictions.csv columns

race_mae and race_drivers_scored are filled once the race has been graded and are the same for every row of that race. Probabilities are in 0–1. predicted_rank is the model's finishing order; predicted_position is the expected position it is sorted by.

backtest-f1.csv columns

One row per driver per race for the leave-one-race-out validation of the current season: each race predicted by a model trained on every other race. Race-level grades come from the same source as the accuracy ledger.

backtest-f1-races.csv columns

One row per race for every season the ledger covers. This is the table the headline accuracy figures are computed from; per-driver rows for earlier seasons are not published.

License and citation

CC BY 4.0: use, share and adapt freely, with attribution to RaceModel.io. Suggested citation:

RaceModel.io (2026). Frozen pre-race predictions for Formula 1, MotoGP, NASCAR Cup, IndyCar, World Superbike and Formula E [Data set]. https://racemodel.io/data

The same files are mirrored on GitHub and Hugging Face; the canonical copies are the URLs above.