Net Goals 101 · component 11 of 11
Goaltending
A goalie’s net goals are his goals saved above expected, each shot weighted by how much it mattered, net of the share his defenders explain. This page works one real season through from the raw events to the number on the card. Every figure is recomputed from the model’s own data and checked against the published value.
Step 01 · Every shot
1,485 shots on goal faced
Goalies do not get the skater components. A goalie’s value is the goals he saved compared with the expected goals of the shots he faced. For goalies the expected-goals model drops the rebound and rush flags (location, type, side and calibration only): a goalie’s rebounds are largely his own leakage, and repricing them would excuse the skill being measured. Empty-net shots are excluded; a few goals judged flukes on video are removed or discounted.
Step 02 · Weight it
Not every save matters equally
A goalie only prevents goals, so each shot is weighted by what conceding it would have cost his team in win probability, at the score before the shot: a save in a tied third period is worth far more than one at 6-1. The weight is in win-probability points (his shots averaged 17.4; the goals he allowed 17.0):
| Weight (WP pts) | Shots | Goals | xG | Σ w·xG | Σ w·goal |
|---|---|---|---|---|---|
| 0 to 5 | 106 | 9 | 11.73 | 26.9 | 15.3 |
| 5 to 15 | 426 | 37 | 44.34 | 501.2 | 439.3 |
| 15 to 25 | 718 | 68 | 74.88 | 1344.3 | 1211.8 |
| 25 to 40 | 165 | 9 | 16.40 | 509.7 | 270.3 |
| 40 and up | 62 | 7 | 8.73 | 392.4 | 311.5 |
| All | 1,485 | 132 | 156.82 | 2774.5 | 2248.2 |
The goals he allowed at the highest leverage, and the biggest saves (weight × xG):
| Goal against | Clock | Shooter | Type | xG | Weight |
|---|---|---|---|---|---|
| 2025021139 | OT 2:47 | Brandon Montour | wrist | 0.181 | 50.2 |
| 2025020113 | P3 19:06 | Ryan Donato | wrist | 0.097 | 46.0 |
| 2025020011 | P3 18:13 | Shane Pinto | snap | 0.290 | 43.5 |
| 2025020052 | OT 1:19 | Jakob Chychrun | snap | 0.345 | 43.3 |
| 2025020072 | OT 3:36 | Dylan Larkin | snap | 0.323 | 43.3 |
| Save | Clock | Shooter | Type | xG | Weight |
|---|---|---|---|---|---|
| 2025020322 | OT 1:34 | Jack Roslovic | wrist | 0.319 | 50.2 |
| 2025021282 | OT 0:18 | Alex DeBrincat | wrist | 0.305 | 50.2 |
| 2025020169 | OT 1:14 | Miro Heiskanen | snap | 0.283 | 50.2 |
| 2025020803 | OT 1:45 | Ilya Mikheyev | wrist | 0.304 | 43.3 |
| 2025020726 | OT 4:51 | Bryan Rust | wrist | 0.300 | 43.3 |
Step 03 · Normalise
Back to goals
The weighted sums are in win-probability points, so they are divided by the league’s average weight (n, xG-weighted, every shot of 2025-26) to read as goals at average leverage. A second constant k makes league GSAx net to exactly zero, because goals happen at slightly different leverage than chances:
Step 04 · Hand back
The defenders’ share and the shootout
Two zero-sum transfers finish the number. The Suppression regression estimates, with the goalie in the model, how much the skaters in front of him changed whether shots on goal went in; that share of his goals-against-versus-expected is handed to them (re-centred per shot faced, so goalies as a group keep their total). And every shootout attempt he faced gets the mirror image of the shooter’s value.
Step 05 · Result
Per 84 games
Goalie value is quoted per 84 games in net (any appearance counts as a game).
Recomputed +40.47; his 2025-26 row on the Players table shows +40.5. They match.
Step 06 · Context
How unusual is this?
GSAx is noisy: about 219 games in net before a career rate is half signal. Season rows are shown as they happened; careers are shrunk by that amount.
The numbers on this page were rebuilt on 2026-10-08 from the same files the cards are built from. The model behind every step is described on Net Goals 101; definitions are in the Reference.