Skip to main content

Evidence

7 Jul 2026, 15:02 UTC

Over 4 paired rounds, the model too few rounds to say (last race order).

BaselineRaceRoundsModelBaselineGainVerdict
Last race orderfeature46.356.44+0.09too few rounds to say
Last race ordersprint47.228.20+0.98too few rounds to say

Paired bootstrap on the round-by-round difference: 0.090, 95% CI [-0.588, 0.570] in positions gained. Positive means the model is closer to the real finishing order than last race 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.

  • Metrics are only comparable within this series. A position error over this field size means nothing next to another series' number.

Formula 3 · 2026

Model accuracy

How the F3 model’s leakage-safe pre-race forecasts have scored against the actual results, over 5 completed rounds of 2026. Every number is scored finishers-only, using only data available before each race.

Winner hit rate

0%

Podium hit rate

13%

Mean position error

7.16

NDCG@5

0.69

Model health at a glance

Healthy

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

Not enough graded rounds yet.

Rounds monitored

5

recent rounds in the rolling check

Input drift

5/5

model inputs that moved vs baseline

Input drift by feature

Predicted pace· ShiftedWin probability· ShiftedPodium probability· ShiftedMean finish· ShiftedFinish 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 still holding up.

Per-round accuracy

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

Per round (feature race)

RoundWinnerPodium hitsMean error
R1 · Australiamiss0/310.391
R2 · Monacomiss0/37.077
R3 · Spainmiss1/34.429
R4 · Austriamiss1/36.769
R5 · Great Britainmiss0/37.111

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

Walk-forward validation

Model vs the “last race repeats” baseline

Every completed round is re-forecast using only earlier rounds, then scored against a trivial predictor that just replays the previous result. Gold marks the better side. Beating this baseline is the bar the model has to clear.

Sprint race

5 rounds · model vs last-race
MetricModelLast-race
Mean position error6.908.20
Top-5 ranking0.6600.661
Order agreement0.4700.225
Podium hits / round0.600.50

Feature race

5 rounds · model vs last-race
MetricModelLast-race
Mean position error7.166.44
Top-5 ranking0.6940.728
Order agreement0.3700.456
Podium hits / round0.400.75

Candidate model

A shadow model runs alongside the live one

position-head candidate · gated behind F3_USE_POSITION_HEAD

Comparison basis: pooled (sprint+feature) mean_position_error

Production model still ahead

Production error

6.25

mean positions off

Candidate error

6.37

mean positions off

Gap

+0.13

candidate minus production

insufficient overlap (3 common rounds; need >= 5). The candidate only gets promoted once it beats the live model on enough real rounds — until then the site keeps serving the production forecast.

Probability calibration

How trustworthy the probabilities are

Calibration applied

A well-calibrated model assigns probabilities that match how often things actually happen. F3’s forecasts are tuned against the real classified results so a stated 30% podium chance means roughly 3-in-10 over the long run.

Training rounds

5

real completed rounds

Status

Calibrated on real F3 results.

Generated 8/30/2026, 12:00:22 PM

Calibration samples per market

Win

265

observations

Podium

265

observations

Top 6

265

observations

Top 10

265

observations

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

Historical performance

Backtest across 130 driver-rounds

Predicted finishing order scored against the official classification for the 2026 season — every round, every driver. Each round was replayed using only signals available before lights-out, so nothing here is hindsight.

Rounds Evaluated

5

Mean Position Error

7.16 pos

Within 3 Positions

33.1%

Within 5 Positions

50.8%

Podium Hit Rate

13.3%

Winner Hit Rate

0.0%

Order Agreement

0.370

Top-5 Ranking

0.694

Model health

Win-market Brier trend

Lower is better · 5 rounds

Diagnostics

  • predictedValue: PSI 2.233 (significant drift vs baseline)
  • pWin: PSI 2.044 (significant drift vs baseline)
  • pPodium: PSI 1.819 (significant drift vs baseline)
  • meanFinish: PSI 1.983 (significant drift vs baseline)
  • finishRangeHigh: PSI 1.519 (significant drift vs baseline)

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