Projected saves for every starting goalie on the slate, with the probability of going over or under each line. An XGBoost regression projects the mean from the goalie's recent workload and the opponent's shot volume; his own game-to-game spread sets the width.
The over/under lines on this page are the model's own ladder, not a sportsbook's — compare the projection to the number your book is hanging. Graded picks are settled against the box score after the night's last game is final, on the performance page.
Every starting goalie on the slate, every line on the ladder and the model's picks, updated three times a day.See plans, or read how the projections are built below.
See plans →A goalie's save total is mostly a function of how many shots he faces, so the model leans on volume first: saves and shots against over his last five and ten starts, his season averages, save percentage and goals against over his last ten, and how many starts he has made. It adds the game context — home or away, days of rest, whether he played last night — and the opponent's goals and shots per game over their last fifteen.
That gives a projected mean. The width comes from the goalie's own spread over his last ten starts, floored at 60% of his season spread so a quiet run of games cannot make the model overconfident. A normal distribution around the mean then gives the probability of going over or under each line from 22.5 to 29.5. A goalie needs five starts before he gets a projection.
Shot volume against is the lever. A goalie behind a team that concedes 33 shots a night has a higher floor than a better goalie behind a team that concedes 26, which is why a strong opponent is often good for the over: they shoot more. Back-to-backs matter twice — a tired team gives up more shots, and the second night is when a backup is most likely to start.
The starter is the one thing the model cannot know in advance. Most teams confirm their goalie at the morning skate; if the projected starter sits, the projection does not apply to his replacement. Check the confirmation before you use the projection.
Saves are a reference projection, not a graded play. Tested against simply taking the under on the same starts at the common lines, the model's side was right about as often, so there is no saves record to publish. Read the projection against the line, and check the confirmed starter first.
The graded NHL plays are the moneyline, the puck line road +1.5, shots on goal over 1.5, points, assists and anytime goal. Their records sit on the performance page, losses included.
The model projects a mean from the goalie's saves and shots against over his last five and ten starts, his season rates, rest and back-to-backs, and the opponent's shots per game. His recent game-to-game spread sets the width, and a normal distribution gives the over and under probability at each line.
A goalie can only make saves on shots he faces. Save percentage varies a few points around .900, while shot volume swings from the low twenties to the high thirties. That is why a goalie facing a high-volume offense usually projects over, even when he is not the better goalie.
No. The lines on this page are the model's own ladder from 22.5 to 29.5 saves. No sportsbook line is archived for this market, so compare the projection with the number your book is hanging. A projection two saves above the book line matters more than any single probability.
Most NHL teams confirm their starter at the morning skate, usually by mid-morning Eastern. Back-to-backs are when a backup is most likely to start. The projection belongs to the goalie named on the card, so if a different goalie starts, the projection and the probabilities do not apply to him.
Shots on goal, goalie saves, goals, assists, points and hits, plus live in-game signals. MLB, NFL and NBA included.
See plansFor information only. 21+. If gambling is a problem for you or someone you know, call 1-800-GAMBLER.