It adds context that save percentage misses. Stopping a routine shot and stopping a dangerous chance both count as one save, but they ask different things of the goalie.
A simple example
Imagine two goalies who each face 100 shots and allow 10 goals. Both have a .900 save percentage. The first faces shots worth 12 expected goals; the second faces shots worth 8. These are hypothetical numbers.
| Goalie | Expected goals against | Goals allowed | Raw GSAx |
|---|---|---|---|
| First | 12 | 10 | +2 |
| Second | 8 | 10 | −2 |
Raw GSAx = expected goals against − goals allowed.
The first goalie stopped two more goals than the chances suggested. The second allowed two more. Their save percentages match, but the quality of the shots changes the comparison.
Why wRink's estimate can differ
The raw subtraction describes the result. wRink also adjusts for how much evidence supports the goalie's performance above or below expectation.
We estimate how the goalie changed the scoring probabilities, pull that effect toward average, then apply the adjusted effect to the shots he faced. Preventing goals produces a positive total.
This shrinkage adjustment limits extreme estimates built on little evidence. We don't simply keep a fixed percentage of every goalie's raw GSAx. The +2 in the example therefore isn't automatically +2 in wRink's adjusted estimate.
Which shots count?
The goalie comparison uses saved shots and goals with an identified goalie. Misses aren't saves, and empty-net goals aren't charged to a goalie. Penalty shots and shootouts are excluded.
The probabilities account for the shooter and recorded context, using a baseline for shots that reached the net. They aren't interchangeable with expected goals calculated across all skater attempts.
The event record doesn't capture every screen or movement before a save. GSAx adds useful context to shot stopping; watching the goalie can help explain what that record misses.
Read the total beside the workload
GSAx accumulates over shots faced, so compare the period and workload before treating totals as a ranking of ability. We convert adjusted GSAx into wins above average, then add one replacement adjustment to obtain goalie WAR.
Published goalie WAR begins in 2018–19. Earlier goalie history failed its stability test and is withheld. Recent listings pool three seasons and show WAR per 60 covered games only with at least 30 covered games. Goalie pages do not show WAR $.
GSAx is measured in goals; WAR is measured in wins. Neither is a prediction of next season.