Every shot has a probability of going in, and adding those up over a season gives you the goals a player should have scored from the chances he actually took. The difference between that and what he really scored is the most useful number in a preseason conversation, because most of it does not carry over. This is that difference for 2025-26: every skater with 100 or more shots, 482 of them, from MoneyPuck shot data for the regular season. How our expected-goals model works if you want the mechanics.
The last column of each table is the part worth reading. It is the same number per 100 shots, last season against the five seasons before it, which is how you tell a finisher from a good year.
The fifteen who beat their chances
Cole Caufield is the headline: 51 goals on 33.7 expected, +17.3. Morgan Geekie and Anthony Mantha follow, and the names below them are mostly the ones you would guess from a goal leaderboard — which is the problem with reading a goal leaderboard.
Only one of them has done it before
Look down the last column. Leon Draisaitl finished +4.2 per 100 shots across the five seasons before last, better than the +3.8 he managed in 2025-26. He is the only player in the fifteen whose history is as good as his season. Everyone else spiked: Caufield +1.8 to +4.4, Geekie +1.8 to +6.1, Mantha +1.9 to +7.0 and Stamkos +2.3 to +4.3. Three of the fifteen — Macklin Celebrini, Cutter Gauthier and Bobby McMann — have thin shot histories behind them, so read their prior number as a hint rather than a record.
That distinction is worth about as much as this kind of analysis ever is, and it can be measured. Take every shooter with 100+ shots in both 2024-25 and 2025-26 and correlate their finishing rates: 0.29 across 396 players. One season tells you a little about the next, and not much. Do it again using the five seasons before last instead of just one, and it rises to 0.36 across 365. Finishing above expected is a real skill measured over years and mostly noise measured over one.
So the practical reading of the board above is not "these players are good finishers." It is "these players will almost certainly score fewer goals on the same chances next season," with Draisaitl as the exception.
The fifteen who could not buy one
This board splits in two, and the last column is again what splits it.
Good shooters who had a bad year. Jake DeBrusk (−0.4 per 100 before, −3.6 last season), Tomas Hertl (−0.4 per 100 before, −3.5 last season), Ross Colton (−0.4 per 100 before, −4.0 last season) and Timo Meier (+0.2 per 100 before, −2.2 last season). These are players whose finishing has been ordinary for years and then fell off a cliff for one. That is the shape luck actually makes.
Players who miss for a living. Anders Lee (−1.9), Andrew Copp (−1.7), Barrett Hayton (−2.6) and Brendan Gallagher (−2.4). Their 2025-26 was worse than their normal, but their normal is already below expected. Nothing about last season was unusual enough to expect a full correction.
What the projection does about it
It does not look at expected goals at all. The projection regresses a player's own per-game scoring rate across three seasons, so a career year gets pulled toward the two before it whether the reason was finishing, ice time or anything else. In practice that lands close to where the shooting numbers point.
Steven Stamkos takes the largest cut of the fifteen, 42 goals down to 30, which is the three-season average and an age curve at 37 working in the same direction. Leon Draisaitl is the only one of the fifteen projected up, and both reasons apply: his finishing is real, and he played 65 games.
Two cautions on that column. It is not a finishing verdict — Macklin Celebrini, Dylan Guenther and Cutter Gauthier also spiked last season and still project up, because at 20, 24 and 23 the age curve outweighs the regression. And in the other direction the model has a real blind spot: it regresses everyone toward the league, so it projects the chronic under-finishers up too — Barrett Hayton 10 to 15, Adam Lowry 5 to 10 and Brendan Gallagher 7 to 11. It does not know that some players simply miss.
The projected goal board, which re-ranks every time the projection is rebuilt:
What this does not know
Expected goals describe the shot, not the shooter: location, angle, what happened just before, and nothing about who was in the net or how the goalie was screened. A player in front of a bad goalie all season collects real goals the model calls lucky. Shot totals here are all situations, so a power-play specialist's chances look better than an even-strength grinder's for reasons that have nothing to do with finishing. And every number above is last season — ice time and line assignments move in camp, and they will move these players more than their shooting percentage will.
The shot maps have this by team and location, and each player's profile carries his own goals-versus-expected history.