This is the index to everything we have published for 2026-27 drafts: nine posts covering forwards, defensemen, goalies, category leagues, the schedule and the method behind all of it. Every number in every one of them comes from the 2026-09-21 projection run — 735 skaters and 97 goalies with enough NHL history to project — and every total is for 84 games, because that is what this season is.
If you read nothing else, read the next two sections. They are the two things that change how you use the rest.
How to read every board in the kit
Read tiers, not ranks. The model's typical miss on a season of points is about 9.5. Any two players inside 9 points of each other are, as far as this projection is concerned, the same player. The boards are sorted, but the sort is not a ranking you should follow to the decimal.
Full season or expected. Every total in the kit is full-season: the projected per-game rate times 84. The fantasy page also has an "expected" mode that multiplies by injury-adjusted games instead, which is the better number for a keeper or best-ball format and the worse one for comparing rates.
The boards are live. The tables embedded in these posts are not screenshots — they re-rank every time the projection is rebuilt, so a post read in November shows November's numbers. The prose around them is written against the 2026-09-21 run and says so.
The model has three tools and no opinions. It averages three seasons, applies an age curve, and adjusts for the team a player moved to. It has never watched a game. Where it disagrees with you, it is not seeing something — it is averaging.
How accurate this is
No one else publishes this next to their rankings, so it is worth being blunt about. These are last season's results: we held out 2025-26, projected it from the seasons before, and compared. Typical miss is the mean absolute error on the season total. Order is how well the projected ranking matched the real one, 0 to 1. Both sit next to what you would get by simply writing down what the player did the year before.
Points, goals, shots and blocks all beat the naive baseline on both measures, and blocks beat it by the most. Goalies are where it gets honest: the model misses wins by 7.0 against 8.0 for carry-forward, which is an improvement, but it ranks starts and save percentage worse than carry-forward does. A goalie's volume is a coach's decision made in October, and his save percentage is close to unpredictable at this range. Treat the goalie board as a starting point and the tandem-split table as the real information. The full backtest has every category.
The kit
Start with the boards
- Forwards: where the 2026-27 projections put centers and wingers — Where the centers and wingers land, where the real tier breaks are, and how far down each pool you can go before the points stop.
- Defensemen: points, shots and blocks in the 2026-27 projections — The same for defensemen — a short top tier, then a very flat middle, plus the blocks and power-play boards that reorder the position.
- Goalies: projected starts, tandem splits and last season's GSAx — Projected starts and wins, how every team actually split its starts last season, and goals saved above expected by shot danger.
Then the boards that are not points
- Category leagues: shots, hits, blocks and faceoffs in the 2026-27 projections — How far each category moves the board away from points, and why the hits and blocks boards share almost nobody with the points top 100.
- Who the projection trusts less than last season, and who it trusts more — The players the model disagrees with last season about, in both directions, and the three reasons it has for each.
- Shooting luck: who finished above and below expected in 2025-26 — Who beat their chances last season and who could not buy one, and which of them has ever done it before.
Before you set a lineup
- The 84-game season: games per week, off-nights and back-to-backs — What the 84-game schedule does to a fantasy week: games per week, which nights are off-nights, back-to-backs, and the February hole.
If you want to know why to trust any of it
- How the fantasy projection works, and how wrong it was last year — The method, step by step, and the backtest by category — including where the model loses to just carrying last season forward.
- How Heater xG works — What our expected-goals model counts, what it ignores, and how to read a goals-minus-expected number.
The top of the board
Read it in tiers. McDavid is alone at 129. Kucherov, MacKinnon, Celebrini and Draisaitl are the next tier, 124 down to 115, and the order inside it is noise. Then a 10-point drop to Hughes and Pastrnak. After that the board flattens hard: the eight players from 99 down to 91 are separated by 8 points in total, well inside the error bar. The forward board splits this by position, which is the version that matters on draft day.
Opening week
Week one runs September 28 to October 4 but the season opens on the Tuesday, so it is a short week and an uneven one.
NYR and VAN play four times; OTT plays once. In a weekly-lineup league that is the first waiver decision of the year, and it is the only week of the season where the gap is this wide. The schedule post has all 28 weeks.
Who has no projection
1,286 players are on an NHL roster right now. 830 of them have a projection. The other 456 do not, and they split into two groups: 314 have no NHL history for the model to read at all, and 142 have played fewer than 20 across the three-season window, which earns a thin_history flag and keeps them off every board in this kit.
That second group is the one to know about, because it is full of names you will hear in the first week of camp.
We would rather show nothing than project a full season off nineteen games. If one of these players wins a job, the number appears after the rebuild — it does not appear now because a handful of games is not evidence.
What is not in it yet
Camp lines. Ice time and power-play units in the kit are last season's, so a forward who moves up a line or onto the first unit in the next two weeks is not reflected anywhere yet; the projection gets rebuilt when those land and every board and embed here moves with it. There is no ADP or roster-percentage data anywhere on the site, so the kit tells you who the model rates higher, never who your league will take earlier. And there is no plus/minus board, because the backtest says we cannot project it.