F1Telemetry
← Prediction history

Prediction Credibility

Every race prediction is locked before the result is known, then scored against the actual outcome. This page shows the running season score and how our model compares to naive baselines — the qualifying grid order and championship standing. A credible model must consistently beat both.

Season avg ρ0.669Moderate
Events scored11races + 4 sprints evaluated
Beats grid position3/10events where model > qualifying order

Model vs Baselines — Spearman ρ per Round

Model accuracy compared to simply using the qualifying grid order or championship standing as a prediction.

Pit Window Confidence Calibration

Stated confidence vs actual within-2-lap hit rate per confidence bucket. A well-calibrated model's bars should be close in height.

Qualifying Prediction Accuracy

% of drivers whose predicted qualifying time was within 0.3 s of their actual best lap. Green ≥ 60%, yellow ≥ 35%.

EventWithin 0.3s
R1 Australian Grand Prix10.0%
R3 Japanese Grand Prix0.0%
R4 Miami Grand Prix4.5%
R5 Canadian Grand Prix0.0%
R6 Monaco Grand Prix45.5%
R7 Barcelona Grand Prix0.0%
R8 Austrian Grand Prix0.0%
R9 British Grand Prix9.1%
R10 Belgian Grand Prix0.0%
R11 Hungarian Grand Prix14.3%
Methodology

Spearman ρ measures rank correlation between predicted and actual finishing order (1.0 = perfect, 0 = random, −1 = inverted). Grid baseline is the ρ achieved by simply using the qualifying grid order as the prediction — the minimum bar any useful model must clear. Championship baseline is the ρ achieved by predicting in championship-standing order. All predictions are locked 90 minutes before session start — no retroactive changes.