Today's puzzle Remember Some Guys — name the skater from the shape of his career. Three clues, one guess a day. Play →

Players

2026-27 · regular season + playoffs · 627 skaters with at least 29 minutes of 5v5 ice time (TOI is that ice time, in minutes) · click a column header to sort

Two seasons on one row. The observed columns — Team, TOI, G, xG and the on-ice rates — are 2026-27 to date and update daily. Every modeled column (RAPM, the Shooting and Blocking ratings, G+/100 and Value) is the 2025-26 fit and stays fixed until 2026-27 has enough games to refit. A skater the fit never saw shows a dash, never a zero.

observed — counted from the play-by-play modeled — the player's own share, isolated offense defense per-shot skill special teams

How to read this table

The table reads left to right, from what happened to what the player drove: first the observed record — his own shots, then his team's results with him on the ice — then the modeled reads that strip the context away, ending in GAR, the season put in one number. The marks on the headers say which is which, and tie the same end of the ice together across techniques.

Shooting — season totals is the player's own attempts, all situations: G goals scored, xG what those attempts were worth. The gap between them is his raw finishing this season — readable, but mostly noise over one year, which is why the modeled shooting columns exist.

On-ice is what the whole team did while the player was on the ice at 5v5, per 60 minutes: expected goals for (xGF/60) and against (xGA/60), and the resulting share (xGF%). Real results, context and all — linemates, competition and deployment are inside these numbers. Everything to the right of the heavy rule is the model taking that context back out.

Shooting — modeled is the shooter rating from the joint shooter/goalie fit: multi-season, time-decayed, and decontaminated of teammates and goalies faced, positive = good. It deliberately replaces the raw season G − xG residual. Shooting a puck is not one skill, and the model no longer treats it as one: an attempt has to get through the defenders in the lane, then hit the net (accuracy), then beat the goalie (finish). Each column is that stage's own margin, so they are three separate readings and never a sum — each sits on a different conditional probability. G+/100 is what all three are jointly worth, composed properly and put on this season's own attempts: goals added per 100 attempts versus what an average shooter would have managed with the same chances. Per-shot quality only: it says nothing about how many attempts a player generates.

Blocking — modeled is the defensive mirror of those columns: the rating for suppressing conversion on the attempts a player is on the ice for — getting a body in the lane so the shot never arrives. It is measured on the same stage as Through, from the other side, which is why it sits with the per-shot ratings rather than with the on-ice defense columns: those work at the expected-goals level and are blind to who is defending. Positive = good. It is a small effect — the league's best blocker is worth about a goal a season at 5v5 — and it covers only conversion per attempt faced, never stopping the attempt from being taken at all. That second thing is real defense too; it is the Def column in the RAPM groups, and the two never double-count.

RAPM is the regression-isolated version of the on-ice rates — the player's own contribution once teammates, competition, zone starts, score state and home ice are controlled away. Same per-60 units, but relative to league average: an Off of +0.20 in the xG trio means the team generates ~0.2 more expected goals per hour of 5v5 with this skater out there than with an average one. Off = effect on the team's xGF/60, Def = effect on its xGA/60 (sign-flipped so positive = good on both), Net = Off + Def. The xG/60 trio measures chance quality; the attempts/60 trio (Corsi) measures shot volume, on a ~15× coarser scale — and is the more reliable of the two. The window selector above the table swaps only these six columns, in place — row order and your current sort stay put. 3-season pooled is the honest default; single-season RAPM is much noisier and is a rough read, not a ranking.

Value is where the table ends: a player's season in one number. It is an accounting of what actually happened, built from exactly six pieces that do not overlap: chances created (Off) + chances suppressed (Def) + special teams (ST, chances created and suppressed, on the power play and on the kill alike) + shooting (Sh, converting his own attempts above what an anonymous shooter would have) + blocking (Blk, the same for attempts he faced) + discipline (Pen, penalties drawn minus taken, priced at what a power play is actually worth). All six are in goals, so Off + Def + ST + Sh + Blk + Pen = GAA reads straight across the row.
Then the last two columns change the question from "versus average" to "versus the alternative". GAA is goals above what a league-average player would have produced. But a team that loses someone does not replace him with an average player — it replaces him with whoever is freely available, who is worse. Rep is that gap over the minutes he actually played, and GAR = GAA + Rep is the same season measured against that replacement instead. Rep is credit for having played the minutes, so it grows with ice time and position and never with how well the time was used — which is why it is its own column rather than being spread across the six terms, where it would invent an attribution the model never made. It is positive for everyone (replacement is below average everywhere), so GAR sits above GAA on every row. GAR is the number the player cards lead with, and it is the better one for "who was most valuable this season"; GAA is the better one for "was he above average". A player between the two — negative GAA, positive GAR — was below average and still worth more than the call-up.
ST is a skill estimate, but it is measured differently on the two sides — read it carefully. Both are regression-isolated the same way Off and Def are: the unit creates and concedes this much more with him on it, not the unit's output handed round to whoever was standing there. But the power play and the kill get there by different routes, because that is what the data supports, not a modeling choice. The power play uses a three-season window: a power-play unit's five players share almost all of their minutes together (the median player spends 83% of his power-play time with one particular teammate), which within one season is far too little independent variation to separate them, and tested directly a single season's worth of power-play deviation does not predict itself back — it reads as noise. Across three seasons the units dissolve enough to separate the players, so a PP player's ST number leans on the seasons either side of this one. The penalty kill is season-resolved instead: a player's kill rating is his career-to-date prior, updated by this season's own play, because tested the same way a season's worth of kill deviation DOES predict itself back — the signal a single power-play season lacks is present on the kill. So a PK number is this season's own evidence (steadied by, not replaced by, his history), where a PP number is a three-season read centered near this one. Both sides count both directions — a forward who creates shorthanded is paid for it, and a power-play unit that bleeds chances the other way is charged for it.
Because the unit is goals over a season, ice time is inside the number — playing 22 minutes a night is worth more than playing 12 at the same quality, which is what an accounting of value should say. The flip side: a player who missed half the year has a smaller GAA and GAR because he was not available, which is true about value delivered and misleading about how good he is. That is what the last column is for: GAA/60 is the same accounting as a rate, so it asks how good he was per minute rather than how much he delivered, and a part-season is no longer penalised. Sort by GAR for the season's most valuable players; sort by GAA/60 for the best per minute, and expect to see low-ice-time players climb. Two cautions: its clock is not the RAPM columns' clock — the rate is per 60 minutes of the time this accounting actually covers (5v5 plus special teams), where the RAPM cells are per 60 of 5v5 alone, so the two are not directly comparable; and a small denominator makes a loud rate, so read it alongside TOI. It is the same quantity the career page plots, on that page's single-season setting.
Coverage: shooting and blocking cover 5v5, power play and penalty kill; Off and Def are 5v5 only, with special teams' rate value arriving through ST instead. 4-on-4, 3-on-3 and empty net are outside the number entirely, and goaltending is its own object. Pen counts every penalty, including the ones that were the right hockey play — the hook that kills a breakaway is a cheap trade, and this accounting prices it as a straight loss.

TOI is 5v5 minutes with both goalies on; the on-ice and RAPM columns share that universe. RAPM columns come from fit rapm__0cd6d075 on nested_structure__nested_structure_xg__d2ec2662 (the joint-xg structural baseline); the modeled Shooting and Blocking ratings from joint__baseline_xg__f3e3111e (as of 2026-06-14, nested chassis) — separate surfaces from the box-score xG model badged in the footer. The rating columns are career-scoped and time-decayed; G+/100, Sh and Blk put those ratings on 2025-26's shots, so they move with a player's usage as well as their skill. Skaters under 29 5v5 minutes this season are fit but not shown. FO is a rating too, from the faceoff model's own fit — it is not priced in goals and is not part of the Value sum below (docs/handoffs/ faceoff-value-term.md §10: measured, gated, and kept as a rating rather than an accounting term).

The Value columns come from value__f4dc1877, built on the same RAPM fit above. They are in goals over 2025-26 as played — not per 60 — so the six terms share one unit and sum to GAA across the row no matter which RAPM window is selected above, with Rep carrying that sum on to GAR. Rep is not stored: it is exactly GAR − GAA, resolved per position and strength state, and is shown as its own column because it belongs to no term. GAA/60 is the artifact's gaa60_cov: the same goals over covered ice time (5v5 + power play + penalty kill), which is a wider clock than the 5v5-only one the RAPM and On-ice columns are per-60 of — the column is grouped separately for that reason, and equals the career page's single-season point. Off and Def are the single-season RAPM coefficients converted to goals on the player's own 5v5 minutes, so they need not equal the 3-season-pooled cells when that window is showing. Shooting and blocking are each centred within every season and strength state separately, which is what keeps the power play's higher scoring environment out of a player's number rather than inside it. The two differ in tense, gated independently and each measured rather than assumed (the same discipline as the ST split above): Blk is season-resolved — a season's rating is the career-to-date prior through the season before, updated by that season's own play — while Sh stays a single career-wide snapshot, so cross-era comparisons of Sh specifically carry a drift the centring cannot remove. Blocking was measured to gain more from being season-resolved than it lost to the smaller sample that update leans on; shooting was measured the other way and stayed put.

Goalies

The against end of the same on-ice machinery, regular season, 5v5: GA goals conceded while in net, xGA what the chances faced were worth, Saving = xGA − GA (goals prevented above expectation, positive = good). Sv rating is the modeled multi-season saving rating (per-shot logit margin, positive = good). Click a goalie for their card.

Goalie	Team	GP	TOI	GA	xGA	Saving	Sv rating
Linus Ullmark	OTT	7	299	5	10.9	+5.9	+0.09
Jeremy Swayman	BOS	7	346	8	13.8	+5.8	+0.12
Jake Oettinger	DAL	6	278	5	10.7	+5.7	+0.09
Ilya Sorokin	NYI	6	278	6	11.1	+5.1	+0.17
Leevi Merilainen	VAN	6	225	5	10.0	+5.0	-0.09
Alex Nedeljkovic	SJS	3	136	2	6.3	+4.3	+0.04
Jesper Wallstedt	MIN	8	375	10	13.7	+3.7	+0.09
Igor Shesterkin	NYR	5	213	6	9.4	+3.4	+0.21
Joseph Woll	PHI	6	275	11	14.2	+3.2	+0.08
Daniil Tarasov	DET	3	113	1	4.1	+3.1	-0.00
Jet Greaves	CBJ	5	212	6	9.0	+3.0	+0.15
Dylan Garand	NYR	7	283	7	9.8	+2.8	+0.06
Philipp Grubauer	SEA	5	209	8	10.8	+2.8	+0.03
Carter Hart	VGK	6	291	9	11.0	+2.0	+0.02
Stuart Skinner	WPG	5	248	9	10.9	+1.9	+0.07
Casey DeSmith	DAL	2	99	2	3.8	+1.8	+0.10
Dennis Hildeby	TBL	2	100	2	3.8	+1.8	+0.05
Colten Ellis	BUF	3	116	4	5.6	+1.6	+0.04
John Gibson	DET	6	261	11	12.6	+1.6	+0.11
Cam Talbot	CBJ	3	128	5	6.3	+1.3	+0.05
Matt Murray	NSH	1	43	0	1.1	+1.1	-0.01
Devon Levi	EDM	5	230	8	8.9	+0.9	+0.02
Logan Thompson	WSH	6	276	12	12.8	+0.8	+0.13
Michael DiPietro	BOS	2	92	3	3.7	+0.7	-0.05
Mackenzie Blackwood	COL	7	327	13	13.7	+0.7	+0.08
Lukas Dostal	ANA	6	262	11	11.6	+0.6	+0.05
Adin Hill	VGK	4	200	6	6.5	+0.5	+0.03
Karel Vejmelka	UTA	8	345	12	12.4	+0.4	+0.04
Ville Husso	ANA	3	103	6	6.3	+0.3	-0.02
Jacob Fowler	MTL	4	180	6	6.1	+0.1	+0.06
Semyon Varlamov	NYI	2	96	3	3.0	+0.0	+0.09
Joel Hofer	STL	5	229	8	8.0	-0.0	+0.06
Clay Stevenson	WPG	3	148	7	6.9	-0.1	+0.01
Juuse Saros	NSH	7	334	15	14.6	-0.4	+0.13
Arturs Silovs	PIT	5	223	10	9.6	-0.4	-0.06
Pyotr Kochetkov	CAR	5	238	9	8.6	-0.4	+0.06
Jacob Markstrom	FLA	6	234	7	6.6	-0.4	+0.06
Sebastian Cossa	UTA	3	89	4	3.5	-0.5	-0.00
Charlie Lindgren	WSH	2	89	5	4.4	-0.6	+0.04
Scott Wedgewood	COL	2	100	4	3.3	-0.7	+0.07
Arvid Soderblom	CHI	2	99	6	5.2	-0.8	-0.04
Nico Daws	NJD	3	135	6	5.2	-0.8	+0.00
Akira Schmid	FLA	4	163	7	5.8	-1.2	-0.01
Brandon Bussi	CAR	5	217	10	8.8	-1.2	+0.05
Anthony Stolarz	TOR	3	138	8	6.5	-1.5	+0.12
Joey Daccord	SEA	5	234	13	11.5	-1.5	+0.01
Anton Forsberg	LAK	3	132	5	3.4	-1.6	+0.05
Darcy Kuemper	LAK	4	153	8	6.3	-1.7	+0.06
Tristan Jarry	EDM	3	158	6	4.2	-1.8	+0.02
Jake Allen	NJD	5	249	11	9.1	-1.9	+0.05
Dan Vladar	PHI	5	239	14	12.1	-1.9	+0.01
Andrei Vasilevskiy	TBL	6	272	12	10.0	-2.0	+0.13
Spencer Knight	CHI	7	344	19	17.0	-2.0	+0.06
Calvin Pickard	MIN	1	53	4	2.0	-2.0	-0.05
Sergei Murashov	PIT	4	202	12	9.3	-2.7	-0.01
Samuel Ersson	OTT	3	107	9	5.0	-4.0	-0.11
Jakub Dobes	MTL	4	177	12	7.9	-4.1	+0.13
Sergei Bobrovsky	TOR	6	284	18	13.8	-4.2	+0.08
Dustin Wolf	CGY	5	247	16	11.8	-4.2	+0.07
Devin Cooley	CGY	3	144	11	6.6	-4.4	+0.08
Ukko-Pekka Luukkonen	BUF	7	318	16	11.3	-4.7	+0.03
Yaroslav Askarov	SJS	5	246	17	11.6	-5.4	-0.03
Jordan Binnington	STL	3	148	12	6.6	-5.4	+0.03
Kevin Lankinen	VAN	7	259	22	16.0	-6.0	+0.01

Goalie TOI is 5v5 in-net minutes; the GA/xGA/Saving columns are the same 5v5 slice, matching the card's opening view. The display floor is total in-net time — goalies under it are not shown.