How We Rate Players in the First Month of a Season
2026-11-01 · methodology · GAR
A month of games is not enough to rate anyone, so we do not pretend it is. An early-season GAR is last season's rating, discounted, plus the little that has happened since. Here is what the number means in October, why defencemen look quiet, and what changes at 20 games.
Figures computed 2026-11-01, against the models as they stood that day. We change them — see the methodology page for where they are now.
A month into 2026-27, the median NHL team has played 14 games. Connor McDavid leads our board at 10.0 Goals Above Replacement. Last season he finished at 37.5. Both numbers are right. They answer different questions.
This is what a GAR or WAR figure means when the season is a month old, why we show it anyway, and what the "Early" badge beside it is telling you.
- An early rating starts from last season. Each player's per-minute impact begins at 60% of last season's offensive rating and 70% of his defensive one, then moves as new shifts come in.
- Totals count only what has been played. After 14 games the whole league sits between -0.30 and +1.80 WAR.
- Right now it mostly tracks scoring. Offence, finishing and the power play account for roughly 95% of the differences between players. Defence barely registers yet.
- At 20 games the picture fills in. Percentiles appear, the defensive model switches on, and the badge goes away by itself.
Where the number starts
Our skater ratings rest on RAPM, a regression that asks how much the shot and chance picture changes when a player is on the ice, after accounting for the four teammates beside him and the five opponents across from him. It needs thousands of shifts to pull one player apart from his linemates. Fourteen games give a regular about three and a half hours of even-strength ice time, shared across more than 700 skaters, which is nowhere near enough.
So the fit does not start from nothing. It starts from each player's stored rating for 2025-26, discounted: 60% of the offensive rating and 70% of the defensive one. The discount is there because players change. A 21-year-old improves, a 34-year-old slows down, a winger moves to a better line. Starting below last season's number lets new evidence move a player faster than a full-strength carry-over would. A rookie has no stored rating and starts at zero, which is to say league average, not replacement level.
With a starting point that firm and this little new data, a per-minute rating a month into the season is still mostly last season's. That is the honest state of knowledge, and it is why the badge is there.
Why the totals are small
GAR is a running total, not a projection. It counts the value a player has delivered in the games played so far, so fourteen games of an elite season is a small number.
| Player | GP | GAR so far | 2025-26 GAR |
|---|---|---|---|
| Connor McDavid | 14 | 10.0 | 37.5 |
| Leon Draisaitl | 14 | 6.7 | 19.8 |
| Kirill Kaprizov | 15 | 6.7 | 19.0 |
| Auston Matthews | 16 | 6.5 | 12.3 |
| Nathan MacKinnon | 14 | 5.8 | 29.2 |
We do not show a full-season pace. A fourteen-game pace is fourteen games of noise multiplied by six, and it would put a number on the page that looks far more certain than it is.
Why defencemen look quiet
Five of the top 50 skaters today are defencemen. At the end of last season, seven were. That gap is mostly the model, not the players.
The part of the model that credits even-strength defence beyond the regression will not say anything about a player until he has 20 games. Until then, defence contributes very little to the spread and the board leans toward whoever is scoring. The two near the top, Quinn Hughes in sixth and Evan Bouchard in seventh, are there on offence. Expect shutdown defencemen to climb from mid-November, as each one crosses 20 games.
What changes at 20 games
- Percentiles appear. A player needs 20 games before we rank him against his position.
- The defensive model switches on for each player as he gets there.
- Tier labels use this season. Until then a player card carries the tier from his last completed season, and says so.
- The badge comes off once the median team has played 20.
Nothing is calculated differently on that day. The same fit runs twice a week all season. Last season's share of each rating simply fades as this season's shifts pile up.
How to read an early number
Read the order as "who has started well, weighted by who was already good", not as a ranking of the best players in the league. A veteran near the top with a modest 2025-26 is having a hot few weeks: Artturi Lehkonen is eighth, with 4.9 GAR in 14 games after 9.7 across all of last season. A star in the middle of the pack is not broken. The current board is on the Explorer, and every number on it will keep moving.
Figures as of 1 November 2026: 729 skaters, median team 14 games played, regular season only.
Frequently Asked Questions
Is WAR meaningful after five games?
Only loosely. A month in, a WAR figure is still mostly last season's rating, discounted, plus a small amount of new evidence. It is a fair summary of who has started well among players who were already good, and a poor ranking of the best players in the league. We mark these numbers as early until the median team has played 20 games.
Why do defencemen rank low in early-season WAR?
The part of our model that credits even-strength defence needs 20 games from a player before it contributes. Until then, offence, finishing and the power play account for roughly 95% of the differences between players, so the board leans toward scorers. Defencemen move up from mid-November as each one reaches 20 games.
How much of an early-season rating comes from last season?
The fit starts each player at 60% of last season's offensive rating and 70% of his defensive rating, then updates as new shifts come in. A rookie starts at zero, which means league average. Last season's share fades through the year and is small by the end of it.
More from the Notebook
How Good Is Our WAR Model? We Ran Four Tests and Lost Two
We tested our WAR model against HockeyStats and Evolving Hockey on the four questions an honest player-value model has to answer, matched to each benchmark's published protocol. Summed to a team it loses both tests - it forecasts next season worse than simply reusing last season's standings. Measured per player it wins both, repeating at 0.77 against 0.62 and 0.46.
Which Hockey Stats Are Skill, and Which Are Luck?
Line up every player's value in a category this season against next season and you get a brutal skill-vs-luck test. Staying out of the penalty box repeats more than four times better than finishing - and it changes how you should read a stat line.
What's Actually Inside a GAR Number
GAR compresses a season into one figure - so the parts underneath had better add up. What our GAR is actually made of, why it is anchored on counting stats rather than built purely from RAPM (measured: pure RAPM drops standings R-squared from 0.75 to 0.51), and the bug that left our own component breakdown failing to sum to the number above it.