The pregame model gives one win probability before the game starts. It combines team-strength ratings with other game context in a fitted model; Elo is one of the inputs, not the final formula.
P(win) = fitted model(team strength, form, schedule, venue, home ice)Elo-style ratings are one input: an Elo expectation turns a rating gap into an expected score, and those ratings feed the model along with the other features. The model isn’t just a two-team Elo formula, and we don’t publish its fitted coefficients.
Each team carries Elo-style rating tracks that update after each game using the standard Elo update rule, including a margin-of-victory-weighted track.
The rating tracks and other contextual features (schedule, venue, recent form, team identity) are assembled into one feature vector per game.
A fitted statistical model combines those inputs into a win probability. The current release is a stacked logistic model; its exact form and coefficients aren’t published.
The probability is computed once per game and stored. It’s the model’s best read of the matchup before any in-game information arrives.
The pregame probability seeds the in-game model. At puck drop with 0:00 elapsed and no goals, the in-game model converges toward the pregame output. As game events accumulate, the in-game model takes over — score differential and time remaining become the dominant factors and the pregame baseline fades in influence.
NHL-native pregame probability
Schedule-aware stacked probability model
Promoted · validated
Pregame release 35This model does not incorporate betting market odds, injury reports, or situational factors like back-to-back scheduling fatigue. It is a pure statistical estimate from NHL-native data. Treat it as a structured prior, not a sharp market line.
Choose an upcoming game to compare the baseline probability with matchup, goalie, and team-strength context before puck drop.