How good are these forecasts?
Every forecast scored on seasons the model never saw while its settings were chosen.
Held-back seasons
| Forecast | Forecasts | Log loss | Favourite won |
|---|---|---|---|
| The model | 3086 | 0.676 | 57% |
| Base rate | 3086 | 0.689 | 54% |
| Recent form | 3086 | 0.688 | 53% |
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 |
|---|---|---|---|---|---|
| Bookmakers (Pinnacle, DraftKings, bet365) | 20 Jan 2026 | 913 | 0.670 | 0.662 | −1.2% |
| Polymarket | 7 Oct 2025 | 1687 | 0.684 | 0.674 | −1.5% |
| Kalshi | 19 Apr 2025 | 1769 | 0.683 | 0.672 | −1.6% |
A game decided in overtime or a shootout counts as a win and a loss for the ratings, but the simulated table gives the loser a point, as the NHL does, at the league's recent overtime rate.
The playoff odds seed the real way: top three in each division, two wild cards per conference, best-of-seven series, and a series under way continues from its score.
The starting goalie is only a small adjustment, and for games to come it is a guess until the starter is named on the day; MoneyPuck models goaltending in far more detail.
Expected goals and shot data are not in this rating yet.
Questions and answers
How accurate are PodiumOracle's NHL predictions?
On 3,086 games from held-back seasons (2024–25 to 2026–27), which the model never saw while it was tuned, the side it favoured won 57% of the time. Its log loss was 0.676, against 0.689 for the base rate and 0.688 for recent form. Against the betting markets on the same finished games, its log loss was 0.670 against the bookmakers' 0.662 over 913 games, 0.684 against Polymarket's 0.674 over 1,687 games and 0.683 against Kalshi's 0.672 over 1,769 games (lower is better).