Report card
When we say 60%, it should happen about 60% of the time. We test each model on games it wasn't fitted on and grade how close its numbers land.
Who wins
game-0.3, 2,760 games
Close, a little off at the edges
Games from Friday, October 4, 2024 to Sunday, October 18, 2026, each priced with what was known that morning, including our guess at the starting goalies.
- Our favourite won
- 57%
- home team won 54%
- Average miss
- 2.0 pts
- between our % and what happened
When we said the home team would win
Each dot is a group of games. On the line, a 60% call won 60% of the time.
↑ It happened
Shots on goal
sog-0.1, 51,928 player-games
Says 60%, means 60%
Each skater's shots in each game, projected from the year before it and scored only on games the model wasn't tuned on.
- Shots per game
- 1.54
- we projected 1.58
- Average miss, over 2.5
- 0.4 pts
- between our % and what happened
When we said three or more shots
Player-games grouped by our chance of over 2.5. On the line, a 40% call came in 40% of the time.
↑ It happened
Expected goals
xg-0.3, 248,769 shots
Says 60%, means 60%
The chance a shot goes in, tested on shots from games the model didn't learn from.
- Picks the goal
- 75%
- of the time, a random goal against a random miss
- Average miss
- 0.2 pts
- between our xG and the real shooting %
When we said a shot would go in
Shots grouped by xG. On the line, a 20% chance went in 20% of the time.
↑ Went in
The tests rerun each Monday with the newest games. How the models are built is on the methodology page.