Inside the Prediction Engine — Historical Patterns

Our prediction pipeline transforms raw F1 data — timing sheets, results, calendar information — into calibrated probabilities for every driver at every race.

The process involves four stages: data collection from primary sources, feature engineering (52 predictive dimensions), model inference (Elo + ensemble + Monte Carlo), and probability calibration. Each stage is documented on our Methodology page.

Most prediction sites publish picks. We publish probability distributions. Most prediction sites hide their failures. We score every prediction publicly.

This approach costs us the ability to retroactively claim we were right. But it earns something more valuable: a track record that readers can verify for themselves.