How good are these forecasts?
Every forecast scored on seasons the model never saw while its settings were chosen.
Against the betting markets
Each market's price before the start, scored on the same finished games as the model. A negative figure means the market was more accurate.
| Market | Since | Priced | Model log loss | Market log loss | Model vs market |
|---|---|---|---|---|---|
| Kalshi | 1 Mar 2026 | 15 | 2.299 | 1.868 | −23.1% |
A driver's rating is the driver plus the team's car that season. ESPN's feed gives the finishing order; the team comes from each season's entry list on Wikipedia, matched on the driver's name and round. ESPN doesn't say who retired, so every car is ranked in order.
The title odds play the remaining races with IndyCar's points (50 for a win down to 5), leaving out bonus points for pole and laps led.
A one-off entry that the entry list names differently from ESPN goes unmatched and races without a team rating.
Ovals and road courses are not treated differently yet.
Questions and answers
How accurate are PodiumOracle's IndyCar predictions?
On 52 races from held-back seasons (2024 to 2026), which the model never saw while it was tuned, its favourite won 31% of the time in fields of about 27, and it gave the eventual winner 11% on average. Its log loss was 2.467, against 3.284 for treating every driver as equal and 2.527 for recent form. Against the betting markets on the same finished events, its log loss was 2.299 against Kalshi's 1.868 over 15 races (lower is better).