Select an NHL matchup
One selected game requests up to three markets (up to 3 provider credits on a cache miss). Cached for 10 minutes. Model testing uses free NHL histories, with no subscription upgrade.
Loading games…
Experimental goal model
Experimental estimates. Starting goalies and lineup changes are not modeled.
Model assumptions, tests & settlement
Independent regulation goal counts use recent offense, opponent goals allowed and venue splits, with a fixed 2.8-goal prior. A regulation tie is split 50/50 for overtime/shootout modeling; that split is stress-tested at 40/60. Starting goalies, injuries, lineups, empty-net effects and correlated scoring are not explicitly modeled.
Up to 100 held-out league games are predicted using only earlier game dates. Each team needs 20 prior regular-season games. The benchmark is equal-strength Poisson goals (2.8 each); it is not the sportsbook market. Lower Brier loss is better. Parameters are fixed, not optimized on this displayed test.
Research candidates require 100 held-out tests, error no worse than baseline, positive expected value under ±15% goal-rate and 40–60% overtime-share sensitivity, fresh quotes, a verified pregame matchup and a complete team archive under 48 hours old. These are exploratory checks, not statistical significance or proof of profitability.
Moneyline and standard puck line use the official final score including overtime/shootout (one shootout winner goal). Regulation three-way uses home win/draw/away win after 60 minutes. Missing regulation-end metadata stays ungraded; it is never inferred from an overtime winner alone. Book-specific rules may differ from these paper rules.
Estimated return is per unit staked, accounting for pushes. Historical sportsbook prices are not available for the backtest; forward paper records preserve actual quoted prices before play.