How the odds have done.
Every game is predicted from what was known before puck drop. Since Oct 21, 2026 each night’s odds are saved before the first game and never changed, so the live record below can be checked against our public data files.
Live record
6 of 14 favourites won (42.9%) · log loss 0.734 · Brier 0.270
Four-season backtest
| Season | Games | Favourite won | Points % baseline | Home won | Log loss | Brier |
|---|---|---|---|---|---|---|
| 2022–23 | 1,312 | 62.0% | 60.7% | 52.4% | 0.661 | 0.235 |
| 2023–24 | 1,312 | 59.5% | 57.4% | 54.1% | 0.663 | 0.235 |
| 2024–25 | 1,312 | 59.7% | 55.7% | 56.3% | 0.669 | 0.238 |
| 2025–26 | 1,312 | 55.0% | 52.7% | 52.2% | 0.686 | 0.246 |
| All four | 5,248 | 59.1% | 56.6% | 53.7% | 0.670 | 0.239 |
Each season is replayed night by night with the ratings as they stood before each game.
Are the percentages honest?
| We said the favourite would win | It won | Games |
|---|---|---|
| 52.5% | 55.0% | 1,893 |
| 57.4% | 56.4% | 1,550 |
| 62.4% | 61.1% | 1,017 |
| 67.3% | 68.5% | 505 |
| 72.0% | 74.8% | 226 |
| 77.0% | 84.9% | 53 |
| 82.1% | 100.0% | 4 |
Playoff odds
We replayed 2022–23 through 2025–26, stopping each season at 10, 20, 41 and 60 games, and asked the playoff simulation who would get in. Its Brier score (lower is better) was 0.129, against 0.169 for the same simulation treating every team as equal and letting only the standings and the schedule decide. The percentages held up too: teams we put around 86% made it 89% of the time, and teams around 97% made it 97% of the time. Widening the ratings’ uncertainty made the scores worse, so we didn’t.
Things we tried that didn’t help
- Shot quality in the rankings. Blending each game’s expected-goal share into the ratings didn’t improve the four-season backtest overall. It helped a little in teams’ first 20 games and hurt later, so the ratings stay results-based. Expected goals are shown beside them instead.
- Starting goalies. A goalie term is real but small: with the actual starters known, it improved the backtest’s log loss by 0.0003 and didn’t change how often the favourite won. Since Oct 23, 2026 the odds include it, blended over each team’s likely starters because the NHL doesn’t confirm them in advance.
- The score in expected goals. Knowing the score when a shot was taken added nothing measurable, so our expected-goals model (v1.1) no longer uses it. That also keeps “record by the chances” from leaning on the real score.
Model versions
| Team Puckigree ratings and win odds | 0.2.0 |
| Expected goals (xG) | xg-v1.1 |
| High-danger chance rule | danger-v1.0 |
| Lines and on-ice numbers | lines-v1.0 |
| Player and Goalie Puckigree | puckigree-v1.1 |
| Goalie what-if | whatif-v1.0 |
| Player swap what-if | swap-v1.0 |
| In-game win probability | live-v1.0 |
| Imprint (single-game score) | imprint-v1.0 |
| Playoff and Cup odds | playoffs-v1.0 |
Every published number carries the version that made it. A new major version means the method changed and numbers aren’t comparable with older ones.
Model 0.2.0 · updated Oct 23, 2026