Method

How the model works

The quick read

How it works in 30 seconds

We play every game 20,000 times

Instead of guessing a winner, a computer plays out each game thousands of times and counts how often each team wins.

  1. Rate every team. We look at how many good scoring chances each team creates and gives up, using every game played so far this season.
  2. Don’t overreact. Ratings are pulled toward league average, so a hot or cold start of a few games doesn’t fool the model.
  3. Set up the matchup. Each team’s attack is matched against the other team’s defense, with a small boost for home ice and an adjustment for the starting goalie.
  4. Play the game 20,000 times. A computer plays each game out moment by moment: penalties, power plays, pulled goalies, overtime and shootouts all happen along the way.
  5. Count the wins. If a team wins 12,000 of the 20,000 games, it has a 60% chance. Fair odds are that chance flipped around: 100 ÷ 60 = 1.67.
  6. Freeze it and check it. Each preview is locked at puck drop, then compared with the real result on the Track record page.

Probabilities, not promises: a 60% favorite still loses about 4 games in 10. The standings sim uses the same engine to play out the rest of the season 10,000 times.

The details

Every game is simulated 20,000 times. Each simulation plays 60 minutes in 10-second steps, then 3-on-3 overtime and a shootout if needed. The share of simulations each team wins is its win probability.

Team ratings

Each team's 5v5 expected goals for and against per 60 minutes (score- and venue-adjusted), power play and penalty kill expected goals, and penalties taken and drawn come from MoneyPuck. Every rate is regressed toward league average, so a hot two-week start doesn't dominate.

The matchup

A team's scoring rate is its offense times the opponent's defense, relative to league average, with a small home-ice boost. The starting goalie's goals saved above expected adjusts the rate, regressed even harder than team stats. Unconfirmed starters are shown as projected.

Inside the game

Penalties create power plays, which end early on a goal. Trailing teams push and leading teams sit back. Teams down by one or two late pull their goalie, which creates empty-net goals and wider margins.

Fair odds

Fair odds are 1 divided by the probability, with no bookmaker margin. They're a way to read the probabilities, not betting advice.

Locking and scoring

Previews update during the day as goalies are confirmed and lock at puck drop. Only locked previews count toward the track record.

Testing it on a past season

The backtest replays last season game by game. Before each game day the ratings are rebuilt from only the games already played, so the model never sees the future. Its predictions are scored against what happened and against simple baselines.