We graded the NHL wallets two weeks in: 1,184 on the board, 57% farming-flagged, a clean cohort underwater, and four wallets that trade hockey alone

Two weeks into the NHL's expanded 84-game season, the graded NHL board holds 1,184 wallets, and 56.8% of them carry a farming-risk flag — the twelfth-highest share of the 26 category boards we grade, against 34.9% site-wide. The wallets that pass the farming check traded 56% of the board's hockey volume and finished −$13,977 on it; the two flagged cohorts took all of the board's $141,558 net profit, and the single top wallet, which carries a severe flag, took 49% of it on its own. Fourteen wallets carry a positive copy verdict. Between them they have traded $2,271 of hockey.

The finding that explains most of the rest: only four of the 1,184 wallets on the NHL board trade hockey alone. 904 of them are also on the NBA board, 1,073 on the soccer board, and the median NHL wallet qualifies for nine other boards. There is no hockey-specialist population on Polymarket to grade — an NHL record is a slice of a generalist's book.

All wallet figures are a point-in-time pull from CopyGrade's live scored set via the read-only database, dated 2026-10-14; market figures are from our Gamma sweep the same day. The board recomputes with every sync, so every figure here is a dated level, the baseline we will measure the season against, the way college football got one in September. "Real edge" is our post-fee, post-latency estimate of what a copier would have kept across a wallet's whole record, not its NHL record alone, per the published methodology.

How big is hockey on Polymarket right now?

The season opened on Tuesday, September 29, with the schedule expanded to 84 games per team (the NHL's schedule announcement) — and Polymarket lists every game as its own event under the nhl-2026 series:

Eighty-six game events settled in the two weeks from September 28, for $23.4M, with the full week of October 5 the biggest so far: fifty games, $13.7M. The largest single games were Blues–Senators on October 13 ($820k), Utah–Bruins ($768k) and Maple Leafs–Golden Knights ($765k) on October 8, and the opening-night Panthers–Hurricanes banner game ($565k). Each game event carries six to fourteen markets — the moneyline, puck lines, totals and period lines.

Hockey futures are thin by comparison. The 2027 Champion market has taken $1.23M across 38 team markets against $53.6k of open interest; the NBA's equivalent is twenty times larger. Hockey on Polymarket is a game-by-game product, which for a copy trader is the useful kind.

One scope note. The NHL board is built on the nhl and hockey tags, so other hockey series — KHL, Liiga, SHL and AHL among them — roll into the same board. The NHL series carries over nine-tenths of the settled hockey volume we track, so the rest moves no figure here materially.

Who is on the board?

56.8% of the board is flagged (672 wallets; 41.4% severe) against 34.9% across all 13,039 graded wallets — and 57.3% among wallets with twenty or more trades in the last 90 days, the population a board mostly draws from. Twelfth-highest of the 26 boards: WNBA (77.9%), college football (74.9%), cricket, UFC, esports, tennis, IPOs, boxing, MLB (59.6%), the NFL (59.3%) and commodities sit above it; the NBA (43.0%) sits fourth-lowest and Formula 1 (40.9%) lowest, and all 26 boards sit above the site-wide rate, because concentration on one market family is exactly the repetition the farming detectors read.

The last three columns carry the sign-versus-magnitude split every sports board has shown since the NFL cut: clean wallets are underwater more often than severe ones (88.9% against 79.0%) and by less (a median −1.00% against −2.70%). The board's negative-edge share, 83.7%, is the eighth-highest of the 26, behind Formula 1, golf, the NFL, the NBA, the Premier League, MLB and the Champions League.

The clean cohort is underwater, and one wallet holds half the profit

The wallets that pass the farming check are 43% of the board and 56% of its volume — $30.3M — and they finished $13,977 down on it. In the August category map the same cohort read $28.8M of NHL volume for $671 of profit; that was a dated 2026-08-05 level and this is a dated 2026-10-14 one, on a board that has since added two weeks of a new season and several hundred wallets, so the two are not a trend — but both say the same thing: tens of millions of dollars of clean hockey trading that nets nothing.

Where the profit sits:

One severe-flagged wallet holds half of the board's net result; the top ten, all flagged, hold more than all of it. Everyone outside the top fifty is, collectively, $48k underwater on hockey. We are not naming any of them — the concentration is the finding, and a fill-by-fill description of one position would be a name in all but form.

Hockey wallets are not hockey wallets

A tag board qualifies a wallet with five trades, and 1,180 of the 1,184 wallets with five NHL trades have five on something else too — a median of nine other boards. That changes how to read every NHL figure above. The cohort that is $14k underwater on hockey is not a group of hockey traders having a bad fortnight; it is the hockey slice of more than 500 generalist books. And it changes what to vet: a wallet's hockey record is a small, possibly incidental part of the record its score is computed on, which is why the verdict's category-edge breakdown — where a wallet's edge actually comes from — is the first thing to open on any NHL-board name.

Fourteen positive verdicts, and what they have traded

All fourteen also clear the canonical four-part copyable bar, so "positive verdict" and "worth starting diligence on" are the same fourteen wallets here. They have traded $2,271 of hockey between them, a median of twelve and a half NHL trades each, and six of them have been under observation for less than a month. They are whole-record verdicts with a hockey footprint, not hockey records — the same shape as the NBA board's 31 a week before that season.

The first two weeks' tape

We read every fill graded wallets placed on the nhl-2026 game events since September 28, split by cohort and by whether the fill landed before the puck dropped (the game-start time Polymarket publishes per market):

The severe cohort placed 90.6% of the graded fills and 56% of the buy volume, two-thirds of its fills in play. The clean cohort trades the other way: 70% of its fills and 78% of its buy dollars land before face-off. Pre-game positioning is the only part of a record a copier inherits at a comparable price, so the cohort a copier can actually follow is the one placing a twentieth of the fills (a quarter of the dollars) — and losing money at it.

The prices tell you why a hockey hit rate is close to meaningless on its own. Across 2,744 pre-game moneyline buys by graded wallets in the first two weeks, the median price paid was 51¢, 77.2% were between 40¢ and 60¢, and only 2.3% were at 70¢ or above. On NBA preseason games the same cut read a 47¢ median with 12.0% of buys at 70¢ or above; the regular-season NBA will be far more lopsided. Hockey moneylines are coin flips with a lean. A wallet that hits 55% of 200 hockey calls has beaten its prices by about ten games — worth reading, not yet proof — where 55% on NBA favourites priced at 70¢ is simply losing to the market. The recent-form guide walks the expected-hits arithmetic; it is the only honest way to read a moneyline record.

What this means for copying hockey

Hockey gives a copier the thing most sports cannot: 1,344 regular-season games of comparable, fast-resolving, roughly even-priced markets, so a wallet that trades the slate for a season can build a record large enough to mean something. The board two weeks in says almost nobody has done that yet in a form a copier can use — the wallets with hockey sample are flagged and trading in play, the wallets a copier could follow have thin hockey records and lose small, and nobody on the board is a hockey specialist. The season guide sets out the hockey-specific checks; the general rule is unchanged: vet the whole record, price every hit, and treat a positive verdict as where diligence starts.

A farming-risk flag is an algorithmic risk assessment, not an accusation of conduct, and a positive verdict is not a recommendation — our own validation of whether the score predicts copier returns is still null. CopyGrade is analysis-only — it never executes trades, holds funds, or custodies keys. Not financial advice.