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 | 1776 | 0.718 | 70% |
| Base rate | 1776 | 0.881 | 52% |
| Recent form | 1776 | 0.799 | 63% |
A regular-season game can end tied after overtime, so the model forecasts a win, a tie and a loss.
The title odds use an approximate tournament: each conference's best team by conference record gets in, standing in for its tournament winner, the rest of the eleven are the best left by simulated wins and rating. The top five wait while the other six play a first round at the higher seed's rink; the quarter-finals are at the higher seed, then the Frozen Four. The committee's selection and seeding are not modelled.
A handful of programs (Wisconsin, Ohio State, Minnesota and Minnesota Duluth among them) dominate the sport, and the ratings are spread wider than in the men's game.
Roster turnover between seasons is only a pull toward average; transfers, recruits and Olympic centralisation years are not in the rating.
Goaltending is not in the rating.
Live scores come from ESPN's scoreboard, matched to College Hockey News's games by date and teams; a game ESPN doesn't list has no live score.
Questions and answers
How accurate are PodiumOracle's NCAA women's hockey predictions?
On 1,776 games from held-back seasons (2024–25 to 2026–27), which the model never saw while it was tuned, the side it favoured won 70% of the time. Its log loss was 0.718, against 0.881 for the base rate and 0.799 for recent form.