NHL Prediction Model Performance & Calibration

See how our NHL prediction model performs: accuracy, Brier scores, calibration, and error metrics across thousands of games.

To understand each metric, read Understanding Performance Metrics . To apply this in practice, view today's NHL predictions, our NHL playoff odds, and in-game win probability charts.

How to Read These Metrics

Accuracy: The percentage of games where the predicted winner (team with >50% win probability) actually won. Simple but incomplete — it ignores how confident the model was.
Brier Score: Measures the mean squared error of probability predictions (0–1 scale, lower is better). A coin-flip baseline yields 0.25; our model targets values below 0.24. Brier score rewards well-calibrated confidence levels, not just picking the right side.
Log Loss: A logarithmic scoring rule that heavily penalises confident wrong predictions. Assigning 90% to a team that loses costs far more than assigning 55%. This keeps the model honest about uncertainty.
Calibration: Shows whether stated probabilities match real outcomes. In a well-calibrated model, games given a 70% win probability should be won about 70% of the time. The calibration tables below group predictions into decile bins so you can verify this directly.
RMSE (Root Mean Squared Error): For goal total predictions, RMSE measures how far predicted totals are from actual totals on average, in goal units.

Why calibration matters most: For probabilistic predictions, calibration is more important than raw accuracy. A model that says "55%" every game can be 55% accurate but useless for decision-making. A well-calibrated model tells you how much to trust each prediction. Learn more in our analytics guide and methodology.

Game Predictions (Multi-Window)

Window	Start	End	Games	Accuracy	Brier	Log Loss	Avg Winner Prob	RMSE Total
last 30	2026-03-16	2026-04-15	244	56.6%	0.2360	0.6631	53.2%	2.637
season to date	2025-10-01	2026-04-15	1306	56.3%	0.2440	0.6808	52.1%	2.588
multi season	2023-10-10	2026-04-15	3930	61.9%	0.2270	0.6447	53.9%	2.202

Totals (Over 5.5)

Window	Games	Accuracy	Brier	Log Loss	Avg Outcome Prob
last 30	244	50.4%	0.2685	0.7341	50.6%
season to date	1306	54.5%	0.2614	0.7210	51.5%
multi season	3930	62.7%	0.2260	0.6423	55.2%

Playoff Game Performance

Start	End	Games	Accuracy	Brier	Log Loss
2024-04-20	2025-06-17	174	63.2%	0.2304	0.6530

Current Matchups

UTA vs STL model odds NSH vs ANA model odds WPG vs SJS model odds

Daily Performance

Show:

Date	Games	Accuracy	Brier	Log Loss

Prediction Recap Highlights

Definition: High confidence means the model assigned the predicted team a win probability well above 50%. Edges that hit are the highest-confidence correct calls, misses are the highest-confidence incorrect calls, and surprise results show the largest absolute gap between win probability and the actual outcome.

Show:

Biggest Model Edges That Hit

Date	Matchup	Win Prob	Outcome

Biggest Misses (High Confidence)

Date	Matchup	Win Prob	Outcome

Surprise Results

Date	Matchup	Surprise	Outcome

Calibration (Win Prob Deciles) — Last 30

Bin	Count	Mean Pred	Observed
2	2	25.2%	0.0%
3	22	37.3%	31.8%
4	66	45.5%	48.5%
5	72	54.6%	44.4%
6	49	65.2%	59.2%
7	26	73.4%	76.9%
8	7	83.3%	85.7%

Calibration (Win Prob Deciles) — Season To Date

Bin	Count	Mean Pred	Observed
2	10	26.8%	30.0%
3	104	36.8%	42.3%
4	311	45.7%	44.7%
5	447	54.7%	49.7%
6	287	64.6%	59.6%
7	131	73.5%	69.5%
8	16	83.5%	75.0%

Calibration (Win Prob Deciles) — Multi Season

Bin	Count	Mean Pred	Observed
1	1	16.6%	0.0%
2	57	26.7%	7.0%
3	368	36.3%	31.5%
4	969	45.5%	44.0%
5	1270	54.8%	54.3%
6	833	64.4%	66.0%
7	366	73.7%	77.6%
8	66	83.2%	92.4%

Calibration (Over 5.5) — Last 30

Bin	Count	Mean Pred	Observed
1	1	16.8%	0.0%
2	7	26.5%	71.4%
3	48	35.4%	60.4%
4	32	45.1%	59.4%
5	78	55.1%	55.1%
6	40	64.0%	57.5%
7	28	75.4%	57.1%
8	8	84.3%	50.0%
9	2	91.3%	100.0%

Calibration (Over 5.5) — Season To Date

Bin	Count	Mean Pred	Observed
1	10	18.7%	70.0%
2	48	25.9%	47.9%
3	187	35.6%	52.9%
4	145	44.5%	57.9%
5	445	55.4%	58.4%
6	202	63.8%	55.0%
7	204	74.5%	59.8%
8	56	84.8%	60.7%
9	9	94.3%	88.9%

Calibration (Over 5.5) — Multi Season

Bin	Count	Mean Pred	Observed
1	34	17.8%	26.5%
2	157	26.0%	24.2%
3	572	35.6%	38.5%
4	464	44.4%	46.8%
5	1251	55.3%	56.4%
6	622	63.6%	62.4%
7	614	74.5%	72.8%
8	176	84.8%	80.1%
9	40	93.5%	97.5%

Team Calibration (Home, Top 15 by Volume)

Team	Count	Mean Pred	Observed	Bias
FLA	146	60.9%	62.3%	-1.4%
EDM	144	61.1%	61.8%	-0.7%
DAL	143	55.9%	63.6%	-7.8%
CAR	137	66.2%	70.1%	-3.9%
TOR	133	53.5%	54.1%	-0.6%
WPG	132	56.4%	62.9%	-6.5%
VGK	132	58.5%	59.8%	-1.4%
WSH	131	54.8%	58.8%	-4.0%
NYR	131	53.8%	51.9%	+1.9%
VAN	130	53.0%	43.1%	+9.9%
BOS	130	51.1%	57.7%	-6.6%
COL	130	63.2%	66.2%	-2.9%
TBL	128	58.3%	63.3%	-5.0%
LAK	128	59.2%	54.7%	+4.5%
STL	126	50.5%	57.1%	-6.6%

Team Calibration (Pred vs Observed)

Mean Pred Observed

Starter Calibration (Home)

Window	Starter Status	Games	Accuracy	Brier	Log Loss
last 30	Starter	244	56.6%	0.2360	0.6631
season to date	Starter	1306	56.3%	0.2440	0.6808
multi season	Unknown	18	72.2%	0.2276	0.6475
multi season	Starter	3912	61.9%	0.2270	0.6446

Cross-Validation (Expanding Window)

Summary: 3 folds | Brier: 0.2495 | Log Loss: 0.6925 | RMSE Total: 2.394

Show fold details

Fold	Train N	Val N	Brier	Log Loss	RMSE
Fold 1	701	2,097	0.2542	0.7024	2.432
Fold 2	1,399	1,399	0.2491	0.6915	2.376
Fold 3	2,103	695	0.2453	0.6838	2.373

In-Game Checkpoints — Last 30

Checkpoint	Games	Accuracy	Brier	Log Loss
end_p1	244	63.1%	0.2082	0.5982
end_p2	244	82.8%	0.1140	0.3590
ot_start	53	62.3%	0.1999	0.5650
p3_10	244	86.9%	0.0830	0.2619
p3_5	244	88.5%	0.0773	0.2435
pregame	244	54.9%	0.2422	0.6772

In-Game Checkpoints — Season To Date

Checkpoint	Games	Accuracy	Brier	Log Loss
end_p1	1306	66.5%	0.2080	0.6019
end_p2	1306	79.3%	0.1423	0.4362
ot_start	326	63.8%	0.1924	0.5496
p3_10	1306	84.0%	0.1027	0.3231
p3_5	1306	85.8%	0.0879	0.2758
pregame	1306	53.4%	0.2478	0.6888

In-Game Calibration — Pregame (Last 30 Days)

Bin	Count	Mean Pred	Observed
3	3	39.8%	0.0%
4	51	46.4%	45.1%
5	152	53.8%	50.0%
6	36	62.9%	69.4%
7	2	70.8%	100.0%

In-Game Calibration — End P2 (Last 30 Days)

Bin	Count	Mean Pred	Observed
0	40	4.5%	0.0%
1	19	14.2%	0.0%
2	19	24.2%	26.3%
3	29	34.4%	31.0%
4	11	46.1%	63.6%
5	18	54.6%	44.4%
6	22	65.9%	72.7%
7	17	76.5%	82.4%
8	25	83.8%	92.0%
9	44	93.8%	100.0%

In-Game Calibration — P3 10 (Last 30 Days)

Bin	Count	Mean Pred	Observed
0	55	3.1%	1.8%
1	29	15.0%	6.9%
2	8	23.1%	25.0%
3	3	38.9%	33.3%
4	11	44.8%	36.4%
5	27	54.2%	51.9%
6	10	62.1%	40.0%
7	1	77.6%	100.0%
8	26	86.6%	88.5%
9	74	97.2%	100.0%

xG Holdout — Contextual

Train: 2023-10-10 – 2025-11-16 | Test: 2025-11-17 – 2026-04-13

Games (test): 991 | Shots (test): 85192 | ROC AUC: 0.786 | Log Loss: 0.2216 | Brier: 0.0601

xG Splits — Contextual Strength State

Split	Shots	Goal Rate	AUC	Log Loss	Brier
Even	68274	6.2%	0.780	0.2022	0.0540
PP	14330	10.6%	0.729	0.2942	0.0810
PK	1823	7.1%	0.822	0.2143	0.0605
EmptyNet	765	50.7%	0.743	0.6100	0.2120

xG Splits — Contextual Shot Type

Split	Shots	Goal Rate	AUC	Log Loss	Brier
wrist	36458	7.1%	0.814	0.2045	0.0554
snap	21423	8.3%	0.776	0.2461	0.0693
slap	10084	5.0%	0.712	0.1852	0.0461
tip-in	8203	6.6%	0.667	0.2332	0.0602
backhand	6300	9.5%	0.827	0.2447	0.0698
deflected	1372	11.7%	0.713	0.3243	0.0944
wrap-around	542	4.4%	0.642	0.1760	0.0414
bat	436	8.9%	0.840	0.2307	0.0631
poke	269	9.7%	0.738	0.2692	0.0704
between-legs	56	12.5%	0.803	0.3043	0.0891
nan	42	61.9%	0.728	2.3192	0.2680
cradle	7	14.3%	1.000	0.2044	0.0574

xG Holdout — Neutral

Train: 2023-10-10 – 2025-11-16 | Test: 2025-11-17 – 2026-04-13

Games (test): 991 | Shots (test): 85192 | ROC AUC: 0.783 | Log Loss: 0.2247 | Brier: 0.0613

xG Splits — Neutral Strength State

Split	Shots	Goal Rate	AUC	Log Loss	Brier
Even	68274	6.2%	0.781	0.2021	0.0540
PP	14330	10.6%	0.705	0.3133	0.0885
PK	1823	7.1%	0.822	0.2146	0.0605
EmptyNet	765	50.7%	0.744	0.6043	0.2089

xG Splits — Neutral Shot Type

Split	Shots	Goal Rate	AUC	Log Loss	Brier
wrist	36458	7.1%	0.809	0.2083	0.0569
snap	21423	8.3%	0.774	0.2486	0.0703
slap	10084	5.0%	0.713	0.1850	0.0461
tip-in	8203	6.6%	0.669	0.2331	0.0602
backhand	6300	9.5%	0.812	0.2558	0.0741
deflected	1372	11.7%	0.712	0.3249	0.0946
wrap-around	542	4.4%	0.655	0.1758	0.0415
bat	436	8.9%	0.835	0.2328	0.0639
poke	269	9.7%	0.747	0.2677	0.0705
between-legs	56	12.5%	0.777	0.3116	0.0909
nan	42	61.9%	0.713	2.3896	0.2752
cradle	7	14.3%	1.000	0.2025	0.0577

Monthly Performance Trends

Track how model performance varies month-to-month across the season.

Month	Games	Accuracy	Brier	Log Loss
2023-10	140	62.9%	0.2195	0.6268
2023-11	213	66.2%	0.2152	0.6191
2023-12	219	66.7%	0.2185	0.6270
2024-01	208	62.5%	0.2167	0.6219
2024-02	172	67.4%	0.2064	0.6016
2024-03	228	71.1%	0.2035	0.5949
2024-04	132	62.9%	0.2265	0.6434
2024-10	166	70.5%	0.2040	0.5965
2024-11	220	66.8%	0.2146	0.6182
2024-12	214	69.2%	0.2056	0.6007
2025-01	224	60.7%	0.2275	0.6461
2025-02	122	55.7%	0.2473	0.6862
2025-03	234	62.0%	0.2279	0.6475
2025-04	132	53.8%	0.2466	0.6861
2025-10	180	57.2%	0.2494	0.6940
2025-11	225	53.8%	0.2439	0.6795
2025-12	226	54.0%	0.2499	0.6936
2026-01	240	56.7%	0.2439	0.6809
2026-02	74	67.6%	0.2234	0.6378
2026-03	242	54.5%	0.2493	0.6913
2026-04	119	59.7%	0.2266	0.6442

Playoff Model Performance

Game-level and series-level accuracy across playoff rounds.

Playoff Games

Round	Games	Accuracy	Brier	Log Loss
All Rounds	86	52.3%	0.2506	0.6945
Round 1	47	44.7%	0.2525	0.6982
Round 2	23	56.5%	0.2530	0.6993
Round 3	10	80.0%	0.2340	0.6613
Round 4	6	50.0%	0.2547	0.7026

Playoff Series

Round	Series	Accuracy	Brier	Log Loss
All Rounds	15	66.7%	0.2312	0.6553
Round 1	8	62.5%	0.2444	0.6827
Round 2	4	50.0%	0.2215	0.6346
Round 3	2	100.0%	0.1999	0.5913
Round 4	1	100.0%	0.2269	0.6470

Playoff Calibration (Pred vs Observed)

Mean Pred Observed