Skip to main content

Evidence

29 Aug 2026, 17:27 UTC

Over 25 paired rounds, the model does NOT beat the baseline (grid order).

BaselineRaceRoundsModelBaselineGainVerdict
Grid orderrace259.629.11-0.51does NOT beat the baseline
Last race orderrace249.4010.09+0.69beats the baseline

Paired bootstrap on the round-by-round difference: -0.510, 95% CI [-0.975, -0.010] 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.

  • The model does not beat grid order on every measured comparison. That is stated rather than hidden.
  • 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.

NASCAR Cup Series · Season 2026

Model accuracy

How the NASCAR model’s pre-race forecasts have scored against the actual results, over 25 completed rounds of the season. Every number is scored against the full official classification (every car is classified in Cup racing), using only data available before each race.

Winner hit rate

20%

Podium hit rate

21%

Mean position error

9.62

NDCG@5

0.73

Per-round accuracy

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

Per round

RoundWinnerPodium hitsMean error
R1 · Daytona International Speedwaymiss0/315
R2 · Atlanta Motor Speedway✓ hit1/311.474
R3 · Circuit of The Americas✓ hit1/310.486
R4 · Phoenix Racewaymiss0/38.432
R5 · Las Vegas Motor Speedwaymiss1/36.778
R6 · Darlington Raceway✓ hit1/39.73
R7 · Martinsville Speedwaymiss1/36.595
R8 · Bristol Motor Speedwaymiss0/38.757
R9 · Kansas Speedway✓ hit1/37.243
R10 · Talladega Superspeedwaymiss0/314.2
R11 · Texas Motor Speedwaymiss1/38.579
R12 · Watkins Glen Internationalmiss0/310.789
R13 · Charlotte Motor Speedwaymiss1/311.077
R14 · Nashville Superspeedwaymiss1/39.526
R15 · Michigan International Speedwaymiss1/311.838
R16 · Pocono Raceway✓ hit2/37.579
R17 · San Diego Street Coursemiss0/310.154
R18 · Sonoma Racewaymiss0/39.667
R19 · Chicagoland Speedwaymiss1/38.526
R20 · Atlanta Motor Speedwaymiss0/39.211
R21 · North Wilkesboro Speedwaymiss1/37.73
R22 · Indianapolis Motor Speedwaymiss0/39.692
R23 · Iowa Speedwaymiss2/310.333
R24 · Richmond Racewaymiss0/38.27
R25 · New Hampshire Motor Speedwaymiss0/38.889

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.

Cup Race

25 rounds · model vs naive baselines
MetricModelLast-race
Mean position error9.6210.09
Top-5 ranking0.7280.672
Order agreement0.3640.286
Podium hits / round0.640.54

Candidate model

A shadow model runs alongside the live one

position-head candidate · gated behind NASCAR_USE_POSITION_HEAD

Comparison basis: race mean_position_error

Production model still ahead

Production error

9.34

mean positions off

Candidate error

9.45

mean positions off

Gap

+0.11

candidate minus production

candidate within ±2% of production (mean change -1.9%). 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

25

real completed rounds

Status

superspeedway stratumroad stratumshort stratumintermediate stratum

Calibrated on real NASCAR Cup results, stratified by track type (superspeedway / intermediate / short / road).

Generated 8/30/2026, 12:00:50 PM

Calibration samples per market

Win

941

observations

Podium

941

observations

Top 6

941

observations

Top 10

941

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 6,045 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

36

Mean Position Error

9.49 pos

Within 3 Positions

26.5%

Within 5 Positions

37.7%

Podium Hit Rate

14.8%

Winner Hit Rate

5.6%

Order Agreement

0.340

Top-5 Ranking

0.619

Model Health

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

14%

higher error than the season benchmark

Rounds monitored

25

recent rounds in the rolling check

Input drift

6/6

model inputs that moved vs baseline

Input drift by feature

Predicted finish score· ShiftingWin probability· ShiftedPodium probability· ShiftedDNF risk· ShiftedProjected finish· ShiftingFinish range (high)· Shifting

Win-market Brier, Round by Round

How far the model's win probabilities sat from who actually won — lower is a better-calibrated forecast.

Some model inputs have drifted from their reference range — normal across a season as the field moves between superspeedways, short tracks, intermediates and road courses. It is flagged here for transparency, but the rolling forecast-quality check above shows whether the predictions themselves are being watched closely.

Diagnostics

  • pWin: PSI 0.288 (significant drift vs baseline)
  • pPodium: PSI 0.287 (significant drift vs baseline)
  • pDnf: PSI 0.572 (significant drift vs baseline)
  • predictedValue: PSI 0.135 (moderate drift vs baseline)
  • meanFinish: PSI 0.135 (moderate drift vs baseline)
  • finishRangeHigh: PSI 0.241 (moderate drift vs baseline)
  • rolling Brier regression +14.4%