Our Projections vs the Betting Market
Before every game we lock each player prop line next to our own projection, then grade both after the final horn. Books are hard to beat. This page keeps score in public, including the days the market wins.
Season so far, 4,022 graded lines: log loss 0.640 for us vs 0.631 for the market (lower is better). Our chance was closer on 2038 lines, the market's on 1714.
Today: biggest disagreements
No lines locked yet today. Each game's lines lock about 75 minutes before puck drop.
Season scorecard
| Market | Lines | Closer: us / market | Log loss: us / market | Over: us / market / hit |
|---|---|---|---|---|
| Points | 1,436 | 683 / 657 | 0.653 / 0.642 | 41% / 42% / 42% |
| Assists | 1,404 | 771 / 540 | 0.585 / 0.573 | 29% / 32% / 30% |
| Shots on goal | 1,094 | 544 / 474 | 0.688 / 0.687 | 46% / 49% / 47% |
| Blocked shots | 88 | 40 / 43 | 0.698 / 0.687 | 46% / 51% / 56% |
| All two-sided markets | 4,022 | 2038 / 1714 | 0.640 / 0.631 | 38% / 41% / 39% |
Anytime goal (2,191 players): our average chance 16%, the market's 18% (margin included), and 15% scored.
Day by day
- Tuesday, October 20, 20261105 linesCloser: us 366, market 291
- Monday, October 19, 2026406 linesCloser: us 122, market 118
- Sunday, October 18, 2026197 linesCloser: us 49, market 65
- Saturday, October 17, 20261273 linesCloser: us 376, market 330
- Friday, October 16, 202696 linesCloser: us 28, market 25
- Thursday, October 15, 20261080 linesCloser: us 346, market 296
- Wednesday, October 14, 2026227 linesCloser: us 86, market 54
- Tuesday, October 13, 20261677 linesCloser: us 571, market 466
- Monday, October 12, 2026288 linesCloser: us 94, market 69
Before we went live: backtest
Rebuilt after the fact for Oct 6 to Oct 11: our stored pre-game projections against prices 15 minutes before puck drop. It is kept apart from the live record.
| Market | Lines | Closer: us / market | Log loss: us / market | Over: us / market / hit |
|---|---|---|---|---|
| Points | 981 | 521 / 402 | 0.633 / 0.634 | 39% / 42% / 39% |
| Assists | 962 | 553 / 346 | 0.552 / 0.556 | 28% / 32% / 28% |
| Shots on goal | 745 | 348 / 339 | 0.693 / 0.690 | 46% / 49% / 50% |
| Blocked shots | 71 | 32 / 35 | 0.696 / 0.673 | 46% / 50% / 49% |
| All two-sided markets | 2,759 | 1454 / 1122 | 0.623 / 0.623 | 37% / 41% / 38% |
How it works
- About 75 minutes before each game we lock every player prop line we can match to a projection: the market's consensus line, its chance of the over with the bookmaker margin removed (median of US books), the best prices, and our chance from the projection the site showed.
- Anytime goal chances come from Edgehalla's goal model (goals per 60 minutes times recent ice time) from October 13, 2026; earlier days and the backtest used LineupExperts' projected goals. Points move with the goals; assists, shots and blocks are unchanged.
- Locked lines never change. After the game each line is graded from the box score; pushes and players who did not play are left out.
- Scoring: log loss is the headline, lower is better, and it punishes confident misses. "Closer" counts the lines where each chance was nearer to what happened; it flatters whichever side leans toward the more common outcome (usually the under), so read it next to log loss. A market that already knows everything we know will match or beat us.
- Anytime goal prices only have a Yes side, so they still include the margin. We show our calibration there but don't score it head to head.
- This is a test of our projections against the betting market. It is not betting advice, and a gap is not a recommendation.