NHL game · live box score, betting context · The One Analytics

END OF 2ND+2BUF

MTL

48-24-10

BUF

50-23-9

Model ensemble · how the prediction is built

3 sub-models, blended.

Each sub-model uses a different rating substrate. Bayesian model averaging weights them by rolling Brier score so the ensemble inherits each model's strengths. Disagreement flags games where the sub-models don't see eye-to-eye — lower confidence, wider band.

50.0%

ensemble · BUF favored

Elo Static
fallback · inputs missing
50.0%
P(BUF win)
33%
weight
Goalie Starter
fallback · inputs missing
50.0%
P(BUF win)
33%
weight
Xg Blend
fallback · inputs missing
50.0%
P(BUF win)
33%
weight

Disagreement

0.00 pp

weighted σ across sub-models

Confidence

100% · high

maps from disagreement

Substrate count

0 / 3 active

ones with full inputs tonight

Weights are recalibrated nightly on a 90-day rolling window with strict point-in-time correctness — no model gets credit for a game it hasn't seen

3 sub-models, blended.

3 sub-models, blended.

Every player. Every line.