Track Record
Every prediction we publish is scored against reality. This page is our permanent, public accountability record. No retroactive adjustments. No cherry-picking.
Every prediction is scored against the actual outcome and surfaced on the chart below. We share both hits and misses — the regulation reset taught the model new patterns and our rookie predictions are still calibrating, so each scored race shifts the picture.
Brier score
Lower is better, 0 is perfect
Skill score
vs grid baseline
15
Live races scored
+4 reconstructed · 19 in the history
Brier evolution
Brier score evolution
Per-race score over the season — hover for details
Race results matrix
Race results matrix
Hover any cell for race details
* Reconstructed by walk-forward backfill, not a live pre-race prediction. Headline figures count live races only.
Calibration
Live calibration
Hover any point for details
Drag to read the curve
The model was well calibrated here — prediction and reality agree. Based on 269 scored win probabilities in this probability band.
When we predict a 30% chance, it happens ~30% of the time. Points on the diagonal indicate perfect calibration.
Brier decomposition
The Brier score, decomposed
Reliability
calibration error — lower is better
Resolution
how much the model separates outcomes
Uncertainty
inherent unpredictability of F1
Baselines comparison
| Model | Brier | Skill Score |
|---|---|---|
| The Data Driver | 0.039 | reference |
| Grid baseline | 0.047 | +18.0% |
| Championship baseline | 0.060 | +35.7% |
| Random uniform | 0.054 | +28.6% |
Skill Score = improvement over baseline. BSS = 1 - (model / reference).
Sharpness
Mean favourite probability
Higher sharpness = model is more decisive
28.8%
Trend: stable — model confidence is stable across races
Log loss by race
Log loss penalises confident wrong predictions more heavily than Brier score
log loss
brier
log loss
brier
log loss
brier
log loss
brier
log loss
brier
log loss
brier
log loss
brier
log loss
brier
log loss
brier
log loss
brier
log loss
brier
log loss
brier
log loss
brier
log loss
brier
log loss
brier
log loss
brier
log loss
brier
log loss
brier
log loss
brier
Historical validation
Walk-forward backtest
Validated on 114 races (2021-2025)
Model retrained before each race using only past data — no future leakage
0.041
Brier score
0.558
Spearman rank
Transparency
Predictions published before each race. Track record computed automatically after each race using Brier score. We never modify predictions retroactively. Every probability is timestamped and immutable.
Scoring methodology: Brier score = mean squared error of probabilistic predictions. Skill score = 1 - (model Brier / baseline Brier). Grid baseline uses qualifying order.