Goals & Wins Above Replacement (GAR / WAR)

What GAR measures

GAR rolls everything a skater adds on the ice into one number, counted in goals. It counts the goals he helps create and the goals he helps prevent. Both are measured against what a replacement player would have done in the same minutes. That is the call-up or waiver-wire skater any team can get for free. WAR is the same value in wins, using a goals-per-win rate set by that season's league scoring. GAR is the measurement, and WAR is worked out from it. Higher is better on both, and a replacement-level player sits at zero.

The seven components

GAR is built from seven separate pieces. Each is measured in goals, and they add up to the total. Even-strength offense and defense are the backbone. The other five pieces round it out. Power-play and penalty-kill numbers rest on far fewer minutes. Treat them as a rough guide.

Replacement level, stated honestly

Each piece is measured against its own baseline, and the baselines differ. The table shows which, for GAR and WAR and for the companion Wins Above Average.

So 'above replacement' is literal for even-strength play and the penalty kill. It is close to literal for penalties and faceoffs. It is not literal for the power play or finishing. Those two are measured against league average on both numbers.

Single-season, and how we check it

The headline GAR is a single-season number. Its even-strength core is fit separately for each season, on that season's own data. Each fit is pulled toward league average, never toward the player's own past. How hard it is pulled is set from the data. Its WAR uses a goals-per-win rate set by that season's league scoring. Only one shape setting in that rate is fit on team results. So comparing team WAR with team wins is a fair check. It is not a fully independent one.

Three checks run before anything publishes. The first keeps goalies out of the skater WAR. The second refuses an update if the league-wide ordering drifts outside its expected bands. The third proves that a single-season update changed only that season. These catch mistakes in how the numbers are put together before they ever reach a card.

Sustained and Actual

The season card shows this number twice, under two names. One figure could not answer both questions a reader brings to it. Actual GAR counts every goal above expected at face value. Whatever the skater did to the expected value of his chances that season counts in full. It is the record of the year as it was played. Sustained GAR keeps the share of that finishing that tends to carry into the next season. It sets the rest aside. That share is measured on forwards. A defenseman's finishing cannot be told apart from chance. For him, Sustained keeps more than actually carries over. For most defensemen that extra is only a fraction of a goal. Everything else in the total is identical between the two. That covers chance creation, chance suppression, the power play, the penalty kill, penalties and faceoffs. So the whole gap between them is the finishing piece and nothing else.

How much of the finishing carries over depends on how much shooting stands behind it. A shooter with a full season of volume behind a hot year keeps most of it. The same hot year on a handful of attempts is mostly set aside. A short run of shooting says very little about the next one. That weight is read off the data, not a judgment about the player. It is the same treatment the career figure has always used. What is new is that the season card now names both views instead of showing only one.

Read them together rather than picking one. When the two are close, the season was built on parts of his game that hold up year to year. When Actual sits well above Sustained, a big slice came from finishing. The evidence behind that finishing does not support it yet. That describes how the season happened. Neither number is the corrected version of the other.

Two WAR numbers, and why they do not add up

A skater can show two WAR figures. They are not meant to match. The single-season number fits each year on its own data. It answers what a player did that season. The multi-season number pools the whole window into one steady estimate. It answers how good the player has been across that window. One pooled estimate is a different calculation from several separate seasons. So adding up a player's single seasons will not reproduce the pooled figure, by design. The gap lives almost entirely in finishing, the least repeatable piece. The single-season number credits finishing at face value. That means the goals that beat the expected value of his chances that season. A single-season figure records what happened. So it takes those goals as they came. The multi-season number does not. It pulls each season's finishing toward a neutral, league-average level. The less evidence behind a season, the harder the pull. So one hot season is mostly discounted. A full career of hot seasons survives largely intact. Adding up a high-volume shooter's single seasons therefore lands above his pooled multi-season finishing. That gap is exactly the season-to-season noise the multi-season number is built to remove. The season number is the honest record of one year. The career number is the steadier read of the same player, with one season's noise taken out. Both describe what has already happened.