NHL · 2026–27
NHL Methods
Last successful check: 2026-09-21T22:22:10.330082Z. Enable JavaScript to view current quote availability.
Keep price comparisons, historical references and predictions distinct.
Fresh regular-season markets
The Odds API supplies book prices. Each event must match the home team, away team and start time on the official NHL schedule, with game type 2 in the current season. Unmatched events, preseason, playoffs, postponed games and started games are excluded. We retain provider quote timestamps and hide prices older than 24 hours. A failed refresh hides saved offers until the feed recovers.
Consensus and best price
Margin is removed only when both sides exist at the same book, game, player, market and line. Puck-line pairs use opposite handicaps. Each distinct book gets one vote in the median fair probability. Best price compares the exact same line and outcome.
Market Watch excludes the quoted book from its reference and requires at least three other paired books. Disagreement is a research signal, not proof of value or a guaranteed profit. Whole-number lines can push; market probabilities are conditional on a non-push result.
Player references
We refresh NHL regular-season summaries for the current and previous season. Until a player has 20 current-season appearances, we use the previous season only if it has at least 20 appearances. Once the threshold is met, the current-season rate replaces it. Every reference names its season and sample size. Ambiguous identities and missing players stay unavailable.
The average rate is total shots, goals, assists or points divided by games played. A Poisson reference estimates over, under and push probabilities from that mean. Under an integer line excludes the push; it is not simply one minus the over probability. Shots can be more variable than Poisson assumes. These are uncalibrated references, not recommendations, and no default 50% probability substitutes for a missing player.
Game-total references
The reference mean averages each team’s goals per game with its opponent’s goals conceded per game, then adds the two team estimates. Both teams must have at least 20 games from the same reference season. We do not infer a moneyline, puck-line probability or profitable bet from this mean. Overtime, shootouts and bookmaker settlement rules require separate treatment before totals probabilities are published.
What is still under review
The retired totals training path included same-game box-score inputs and full-sample opponent/split averages. That leaks information unavailable before puck drop. The previous player path could substitute market consensus for a missing model estimate. Those outputs no longer feed these public pages.
Production picks require corrected pregame features, chronological testing, probability calibration and comparison with timestamped odds. Starting goalie, line assignments, power-play role, ice time, injuries and rest need verified inputs. Historical references do not include those adjustments.
Updates and sources
Markets refresh twice daily. Statistics are checked at least daily and include completed dates only, excluding today. Historical references expire if their fetch is more than 36 hours old. Empty markets are a valid state; old December cards never serve as a fallback.