Skip to main content

Evidence

10 Aug 2026, 13:01 UTC

Over 10 paired rounds, the model no difference demonstrated (last rally order).

BaselineRaceRoundsModelBaselineGainVerdict
Last rally orderrally108.128.65+0.54no difference demonstrated
Standings orderrally108.127.81-0.31no difference demonstrated

Paired bootstrap on the round-by-round difference: 0.535, 95% CI [-0.946, 2.739] in positions gained. Positive means the model is closer to the real finishing order than last rally order is; the interval covers zero, so no difference has been demonstrated.

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.

  • Some comparisons straddle zero: a difference has not been demonstrated, in either direction.
  • Metrics are only comparable within this series. A position error over this field size means nothing next to another series' number.

WRC · 2026

Model accuracy

How the WRC model’s leakage-safe pre-rally forecasts have scored against the actual results, over 10 completed rounds of 2026. Every number is scored finishers-only, using only data available before each rally.

Winner hit rate

20%

Podium hit rate

37%

Mean position error

8.12

NDCG@5

0.67

Model health at a glance

Watch

A self-check on the live model: whether recent forecasts are still landing as well as they should, and whether the numbers feeding the model have drifted from what it was tuned on.

Forecast quality

13%

higher error than the season benchmark

Rounds monitored

10

recent rounds in the rolling check

Input drift

5/5

model inputs that moved vs baseline

Input drift by feature

Predicted pace· ShiftingWin probability· ShiftedPodium probability· ShiftedMean finish· ShiftingFinish range (high)· Shifted

Some model inputs have drifted from their reference range — normal early in a season with a small sample of rounds so far. It is flagged here for transparency, but the rolling forecast-quality check shows the predictions themselves are being watched closely.

Per-round accuracy

Podium-weighted rally accuracy per round. Tap a cell for the breakdown.

Per round

RoundWinnerPodium hitsMean error
R1 · Rallye Monte Carlomiss1/36.231
R2 · Rally Swedenmiss1/36.417
R3 · Safari Rally Kenyamiss1/38.429
R4 · Croatia Rally✓ hit2/315.067
R5 · Rally Islas Canariasmiss1/37.909
R6 · Vodafone Rally de Portugalmiss0/37.909
R7 · FORUM8 Rally Japanmiss2/35.545
R8 · EKO Acropolis Rally Greecemiss1/36.077
R9 · Rally Estoniamiss1/38.5
R10 · Secto Rally Finland✓ hit1/39.083

Win Brier scores the model’s win probabilities against who actually won — lower is sharper and better calibrated.

Vs the baselines

Does the forecast beat championship form?

A rally has no qualifying to lean on, so the fair test is whether the forecast is sharper than simply ordering crews by their championship standings. Over 10 completed rounds the model is sharper than championship form on both the win and podium probability scores. A last-rally momentum baseline is a touch sharper on the win score this season — shown here honestly rather than hidden.

MeasureOur forecastChampionship form
Win probability score(lower is sharper)0.8750.876
Podium probability score(lower is sharper)0.7340.765
Winner called(higher is better)20%10%

The skill model alone scores 0.901 on the win score and 0.792 on podium — weaker than championship form on its own. Blending it with championship form is what edges the combined forecast ahead.

Green marks where the forecast beats championship-form order.

Walk-forward validation

Model vs the championship-form baseline

Every completed round is re-forecast using only earlier rounds, then scored against a baseline that simply ranks crews by their championship standings. Gold marks the better side. Beating championship form on the ranking metrics is the bar the model has to clear.

Rally

10 rounds · model vs championship form
MetricModelChampionship form
Mean position error8.127.81
Top-5 ranking0.6650.608
Order agreement0.5300.501
Podium hits / round1.101.30

Probability calibration

How trustworthy the probabilities are

Calibration applied

A well-calibrated model assigns probabilities that match how often things actually happen. WRC’s forecasts are tuned against the real classified rally results so a stated 30% podium chance means roughly 3-in-10 over the long run.

Training rounds

10

real completed rounds

Status

Calibrated on real WRC rally results (one classification per round).

Generated 8/30/2026, 12:01:23 PM

Calibration samples per market

Win

124

observations

Podium

124

observations

Top 6

124

observations

Top 10

124

observations

Each figure is how many prior crew-outcomes fed the calibrator for that market. More samples means a steadier probability estimate.

Model health

Win-market Brier trend

Lower is better · 10 rounds

Diagnostics

  • pWin: PSI 0.405 (significant drift vs baseline)
  • pPodium: PSI 0.671 (significant drift vs baseline)
  • finishRangeHigh: PSI 0.447 (significant drift vs baseline)
  • predictedValue: PSI 0.169 (moderate drift vs baseline)
  • meanFinish: PSI 0.208 (moderate drift vs baseline)
  • rolling Brier regression +12.7%

Feature drift and rolling-Brier are tracked round-to-round; a spike flags where the field behaved unlike the rounds the model learned from.