About the model
How the hockey model works
Start from the betting market, and only bet where the model knows something the price doesn't.
Usually that's something you can name: a team on its third game in four nights, a goalie on a back-to-back, a power play missing its guys. The books adjust for it, just not all the way.
What it bets
- Game totals. The model's best backtested market. Power plays, chances allowed in tied games, and the other goalie's rest.
- Moneylines. Mostly fading tired teams. Nothing shorter than −160; a heavy favourite it loves gets bet in regulation instead.
- Team totals, away puck lines, regulation favourites. From the goals model, which projects every final score. It's where the projected scores on the site come from.
An underdog bet doesn't mean the model thinks they win, only that they win more often than the price says.
The track record
Four seasons run back, each predicted only from the seasons before it and graded at the best sportsbook price: 2,681 bets, +489 units, 12.1% return. Every season was up. At no-vig fair prices it would be 15.8%.
| Season | Bets | Return | Market | Bets | Won | Return |
|---|---|---|---|---|---|---|
| 2022–23 | 287 | 15.2% | Game totals | 461 | 59% | 16.7% |
| 2023–24 | 864 | 7.9% | Away puck line | 367 | 40% | 15.4% |
| 2024–25 | 822 | 12.6% | Moneyline | 434 | 48% | 14.0% |
| 2025–26 | 708 | 14.0% | Team totals | 953 | 55% | 8.7% |
| Regulation favourites | 466 | 60% | 3.5% |
How it was tested
- Every season was predicted using only the seasons before it.
- Every bet is graded at the best price available when it locks, not the no-vig fair price.
- The live card is checked against the backtest, bet for bet.
Expect some losing nights, about 4 in 10. The record is built over weeks.
How bets are sized
Each market has its own staking rule, based on where it has made money in past seasons. No more than 5 units on a game or 10 on a night.
Reading the card
Locked at is the best real number when the bet locked, and it's what the record is graded at. Model is our chance as a fair price, and Edge is how many percentage points our chance beats the market's, with the book's cut removed. Good to is the worst price the bet is still worth taking at. A bet locks once two hourly updates agree, then never changes. ★ is the team's power rating.
— Sushi