Track record
Accuracy
How the forecasts have actually scored against real results — measured across every scored class-round, with nothing left out. Endurance grids are large, so a car landing within a few places of its prediction is a strong result.
Evidence
23 Aug 2026, 19:05 UTCOver 8 paired rounds, the model no difference demonstrated (last race order).
| Baseline | Race | Rounds | Model | Baseline | Gain | Verdict |
|---|---|---|---|---|---|---|
| Last race order | GTD | 8 | 4.93 | 4.99 | +0.07 | no difference demonstrated |
| Last race order | GTDPRO | 8 | 3.38 | 3.11 | -0.27 | no difference demonstrated |
| Last race order | GTP | 7 | 2.86 | 2.99 | +0.13 | no difference demonstrated |
| Last race order | LMP2 | 5 | 2.79 | 2.87 | +0.09 | no difference demonstrated |
| Season form order | GTD | 8 | 4.93 | 5.06 | +0.14 | no difference demonstrated |
| Season form order | GTDPRO | 8 | 3.38 | 3.58 | +0.20 | no difference demonstrated |
| Season form order | GTP | 7 | 2.86 | 2.55 | -0.31 | does NOT beat the baseline |
| Season form order | LMP2 | 5 | 2.79 | 2.79 | +0.01 | no difference demonstrated |
Paired bootstrap on the round-by-round difference: 0.067, 95% CI [-0.245, 0.336] 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.
- The model does not beat last race order, season form 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.
By class
Mean err.
2.86
Podium
48%
Winner
29%
Mean err.
2.84
Podium
47%
Winner
20%
Mean err.
3.38
Podium
33%
Winner
0%
Mean err.
4.93
Podium
25%
Winner
13%
Round by round
| Round | Cars | Mean error | Podium | Exact | Within 3 | Winner |
|---|---|---|---|---|---|---|
| R1Daytona International Speedway | 11 | 3.64 | 1/3 | 1/11 | 4/11 | ✗ |
| R2Sebring International Raceway | 11 | 1.64 | 3/3 | 0/11 | 11/11 | ✗ |
| R3Long Beach Street Circuit | 11 | 1.82 | 2/3 | 0/11 | 10/11 | ✗ |
| R4Weathertech Raceway Laguna Seca | 11 | 2.91 | 1/3 | 1/11 | 7/11 | ✗ |
| R5Detroit Street Course | 11 | 2.91 | 1/3 | 4/11 | 7/11 | ✓ |
| R6Watkins Glen International | 11 | 2.18 | 2/3 | 4/11 | 8/11 | ✓ |
| R8Road America | 11 | 4.91 | 0/3 | 0/11 | 3/11 | ✗ |
Does the forecast beat guesswork?
Combined win + podium forecast error across 28 scored class-rounds — lower is better — measured against two naive baselines: simply repeating the last race’s order, and ranking cars by their season points.
Last-race form
baseline error 0.252
Season form
baseline error 0.267
“Matches it” means the model is statistically no worse than that baseline over the rounds scored so far. On a young season with few rounds, matching a strong season-form baseline while beating last-race form is an honest, expected result.
Calibration
Probabilities are calibrated on 9 real rounds of results, per class. Calibrated on real IMSA results, per class stratum.
Continuous evaluation
Every round is re-scored against simple baselines as results come in, so the track record above updates honestly over the season — no cherry-picking, nothing dropped.