Skip to main content

Evidence

Over 12 paired rounds, the model does NOT beat the baseline (grid order).

BaselineRaceRoundsModelBaselineGainVerdict
Grid orderrace123.613.39-0.22does NOT beat the baseline
Last race orderrace113.534.84+1.31beats the baseline
Standings orderrace113.534.19+0.66beats the baseline

Paired bootstrap on the round-by-round difference: -0.221, 95% CI [-0.418, -0.039] in positions gained. Positive means the model is closer to the real finishing order than grid order is; the whole interval is BELOW zero, so the shortfall is measured, not noise.

forward evaluation — each round was forecast before it ran and scored afterwards; not a backtest. Historical replays and the forward record are computed separately and never merged.

Loading accuracy data

Historical Evaluation

Loading historical evaluation…

Forecast Calibration

Probability Calibration

A well-calibrated model assigns probabilities that match observed frequencies. Bins close to the dashed reference line indicate good calibration.

Perfect calibration (y = x)Observed bins (size = sample count)

Calibration Error

0.0334

Lower is better

Forecast Sharpness

0.1118

Lower is better

vs Uniform Baseline

Baseline not provided

Sample size: 263 observations

Note: Each round is calibrated strictly out-of-sample: the calibrator is fit only on rounds that completed BEFORE it (expanding window), never on its own result. A round with fewer than --min-completed-rounds prior races publishes raw Monte Carlo probabilities (calibration.applied=false) — publishing raw is strictly better than over-fitting a handful of events. Published probabilities carry a small floor (no driver is ever 0.0) and are renormalized per market (win sums to 1, podium to 3, top6 to 6, top10 to 10). The metrics below score the FINAL PUBLISHED probabilities against actual results, not the raw inputs.

Generated: 8/23/2026, 4:17:31 PM