NHL Model Accuracy

Out-of-sample backtests of the goalie projection engine — projections vs. what actually happened, graded per stat against the Marcel baseline — rate skills (SV%/GAA) directly, counting stats over a full starter workload
Goalie projections
Beat the Marcel baseline by 7% on out-of-sample accuracy · p10–p90 coverage 54% (2025-26 backtest, 248 preds)
Player Accuracy
per-game rate vs naive · 248 predsskill 7%p10–p90 cov 54%
StatNModel/gBase/gSkillCov
Rate
Save %620.01220.0132+8%52%
GAA620.30310.3265+7%58%
Volume
Saves62111.3248123.144+10%48%
Shutouts621.9141.9582+2%56%
Team Game Model vs Home-Ice Baseline
backtest
SeasonGamesBrierBaseAccMargin MAETotal MAE
2025-261,2520.25260.249453.1%2.171.84
2024-251,2420.24690.246057.3%2.181.88
2023-241,1700.24470.248857.4%2.191.84
2022-231,1640.23980.250059.9%2.131.83
2021-221,1690.23850.248361.6%2.141.86
2020-217490.24250.249158.2%2.111.86
2019-209420.25050.249055.0%2.091.84
Overall7,6880.24520.248657.5%2.151.85
Opponent-adjusted goal-differential ratings (SRS) → predicted margin, win probability, and total, scored walk-forward (ratings built only from games played before each matchup — no leakage). A lower Brier than the home-ice baseline means real predictive skill; NHL is a high-variance, near-coin-flip league, so honest edges are small. The model never sees the betting line.