Every version of RaceModel's AI-powered prediction model, what changed in it, and how each version scored race by race. When the model is retrained the version changes; frozen predictions always record the version that made them, so an accuracy claim can be tied to a specific model.
Scoring is the same for every version: mean absolute error in finishing positions across the full classified field, retirements counted, plus podium hits. The multi-season backtest with every race graded is on the accuracy ledger; how those numbers compare with figures published elsewhere is at /f1-prediction-accuracy. Terms are defined in the glossary.