Skip to main content

Evidence

9 Jul 2026, 10:10 UTC

Over 11 paired rounds, the model no difference demonstrated (grid order).

BaselineRaceRoundsModelBaselineGainVerdict
Grid orderrace116.115.73-0.38no difference demonstrated
Last race orderrace106.107.70+1.60beats the baseline

Paired bootstrap on the round-by-round difference: -0.380, 95% CI [-1.079, 0.291] in positions gained. Positive means the model is closer to the real finishing order than grid 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.

NTT IndyCar Series · Season 2026

Model accuracy

How the IndyCar model’s pre-race forecasts have scored against the actual results, over 11 completed rounds of the season. Every number is scored against the full official classification (every car is classified in IndyCar racing), using only data available before each race.

Winner hit rate

27%

Podium hit rate

42%

Mean position error

6.11

NDCG@5

0.76

Per-round accuracy

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

Per round

RoundWinnerPodium hitsMean error
R1 · Streets of St. Petersburg✓ hit2/36.16
R2 · Phoenix Racewaymiss1/35.92
R3 · Streets of Arlington✓ hit2/35.2
R4 · Barber Motorsports Parkmiss1/34.48
R5 · Streets of Long Beachmiss1/36.4
R6 · Indianapolis Motor Speedway Road Coursemiss1/36.24
R7 · Indianapolis Motor Speedwaymiss1/39.091
R8 · Streets of Detroit✓ hit2/35.04
R9 · World Wide Technology Racewaymiss0/37.04
R10 · Road Americamiss1/36.4
R11 · Mid-Ohio Sports Car Coursemiss2/35.2

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

Walk-forward validation

Model vs the naive baselines

Every completed round is re-forecast using only earlier rounds, then scored against two trivial predictors — one that replays the previous result, one that assumes the grid finishes in order. Yellow marks the best column. Beating these baselines is the bar the model has to clear.

IndyCar Race

11 rounds · model vs naive baselines
MetricModelLast-race
Mean position error6.117.70
Top-5 ranking0.7570.616
Order agreement0.4300.192
Podium hits / round1.270.50

Candidate model

A shadow model runs alongside the live one

position-head candidate · gated behind INDYCAR_USE_POSITION_HEAD

Comparison basis: race mean_position_error

Production model still ahead

Production error

6.24

mean positions off

Candidate error

6.30

mean positions off

Gap

+0.07

candidate minus production

candidate within ±2% of production (mean change +1.1%). 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. The forecasts are tuned against the season’s real classified results — separately for each of the four track types, because Talladega and Sonoma race nothing alike — so a stated 30% podium chance means roughly 3-in-10 over the long run.

Training rounds

11

real completed rounds

Status

street stratumoval stratumroad stratum

Calibrated on real IndyCar results, stratified by track type (oval / road / street).

Generated 8/30/2026, 12:01:04 PM

Calibration samples per market

Win

283

observations

Podium

283

observations

Top 6

283

observations

Top 10

283

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 3,286 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

17

Mean Position Error

6.08 pos

Within 3 Positions

37.8%

Within 5 Positions

54.4%

Podium Hit Rate

37.3%

Winner Hit Rate

5.9%

Order Agreement

0.355

Top-5 Ranking

0.719

Model health

Win-market Brier trend

Lower is better · 11 rounds

Diagnostics

  • pPodium: PSI 0.334 (significant drift vs baseline)
  • pDnf: PSI 0.647 (significant drift vs baseline)
  • meanFinish: PSI 0.504 (significant drift vs baseline)
  • finishRangeHigh: PSI 2.667 (significant drift vs baseline)
  • pWin: PSI 0.243 (moderate 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.