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 | 1468 | 0.632 | 63% |
| Base rate | 1468 | 0.691 | 53% |
| Recent form | 1468 | 0.659 | 60% |
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) | 28 Jan 2026 | 345 | 0.623 | 0.605 | −2.9% |
Junior hockey turns over fast: players age out at 20 (with a few overage spots), so a team's strength swings from one season to the next. The model pulls every rating a long way back toward average over the summer for that reason.
The playoff odds take the top eight in each conference and re-seed every best-of-seven round, best against worst left, before the final.
Roster moves (trades before the deadline, import picks, players returning from NHL camps) are not in the model until they show up in results.
Goaltending is not in the rating.
Questions and answers
How accurate are PodiumOracle's QMJHL predictions?
On 1,468 games from held-back seasons (2024–25 to 2026–27), which the model never saw while it was tuned, the side it favoured won 63% of the time. Its log loss was 0.632, against 0.691 for the base rate and 0.659 for recent form. Against the betting markets on the same finished games, its log loss was 0.623 against the bookmakers' 0.605 over 345 games (lower is better).