The film room.
Your scoring turns a stat line into points. Then Puckhound goes to the tape: three seasons of NHL games, regression to the mean, age curves, health history, deployment and every fancy stat we could get our mitts on, each one tested before it was allowed to dress.
Your scoring sets the table
One night, run through one league's weights: goals 3, assists 2, power-play points 1, and shots, hits and blocks half a point each.
| Stat | Count | Weight | Points |
|---|---|---|---|
| Goals | 1 | 3 | 3.0 |
| Assists | 1 | 2 | 2.0 |
| Shots on goal | 4 | 0.5 | 2.0 |
| Hits | 2 | 0.5 | 1.0 |
| Blocks | 1 | 0.5 | 0.5 |
| Fantasy points | 8.5 | ||
Put that helper on the power play and the same night is worth 9.5.
| Stat | Count | Weight | Points |
|---|---|---|---|
| Win | 1 | 4 | 4.0 |
| Goals against | 2 | −1 | −2.0 |
| Saves | 30 | 0.2 | 6.0 |
| Shutout | 0 | 3 | 0.0 |
| Fantasy points | 8.0 | ||
A shutout win on 28 saves is worth 12.6.
Those are one league's weights. Puckhound reads yours. In that league a D-man's night of one assist, three shots, four hits and three blocks scores 7.0, and a winger's lone goal on two shots scores 4.0. A generic ranking gets that backwards.
Who gets the ice time
Every projected rate is a blend: this season's games, the two seasons before it, and what an average player at his position does. Thin sample? The position average eats more minutes until he earns his own. Pick a player type and slide through the season to watch the ice time shift.
The lineup card
Everything that goes into a call, where it comes from and how much it counts. The fancy stats had to earn a sweater like everyone else: each one was tested on whether it predicts next season once this season's scoring is already known. If it didn't move the needle, it sits.
First line
Sets the number.
- Your league's scoring, roster and sizeYahoo Fantasy once connected, or set by handTurns every stat into points and decides how deep the league runsEvery point
- Replacement levelYour league size and starting slotsValues each player against the best guy left on waivers at his positionThe baseline for every rank
- Three seasons of per-game statsNHL stats and game feeds, refreshed nightlyGoals, assists, PPP, shots, hits, blocks, PIM and plus/minus; wins, goals against, saves and shutouts for goaliesPer game: 1.5 this season, 0.65 last season, 0.35 two back
- Position average (regression to the mean)League-wide forward, D and goalie rates from 2025-26 regularsPulls small samples back toward normal until a player earns his minutes12 to 40 games of average, by stat
Second line
Adjusts it.
- Age curveNHL player biosYoung legs get a bump, aging vets a discount×1.10 at 20 down to ×0.90 at 35+; hits and blocks move half as much
- Games availableGames played against the schedule, two past seasons plus this oneHow many games he actually dresses forBlended 50/50 with a 93% norm, never under 60%
- Injuries and absencesPublic injury reports, nightly; every games-out call carries a source and a dateWho can't dress tonight, and for how longPer report
- Goalie workloadPreseason depth charts, then actual startsHow many starts a goalie getsThe depth chart counts as 20 team games, then his starts take over
- Goalie team strengthLast season's team recordsWins follow the team in front of him40% of a goalie's win rate
- Draft marketYahoo FantasyThe price behind buy-low and sell-high, and the starting point for rookies with no NHL gamesRookies: rank-implied points, less 10%
- ScheduleNHL scheduleGames per week, back-to-backs and light nights for streamingEvery night
Special teams
Raises the flags. Shown beside the projection, never baked into it.
- Shot volumeNHL stats feedsFlags thin and heavy shooters for their scoring rate+0.0091 lift, the best single signal tested
- Power-play roleNHL ice-time feedFlags who the coach trusts on the PP, and who he doesn't+0.0062 lift
- On-ice shooting %NHL possession feedFlags hot and cold runs that are due to come back to earth+0.0271 lift (5v5)
- Shooting luckNHL game dataBuy-low and sell-high cues once he's put 25 shots on netFires at 1.5× or 0.6× his own baseline
- Ice-time trendNHL game dataFlags a player moving up or down the lineupThe only one of seven streak signals that predicted the next ten games, over 1,999 streaks
- Form checkNHL game dataAfter 10 games, flags anyone running hot or cold against the model±15%
- NewsPublic NHL headlines, nightlySorted by who's on your rosterRead, not scored
The doghouse
Tested, benched. Yes, that includes Corsi.
- CorsiNHL possession feedShot-attempt share. Adds next to nothing once on-ice shooting is known, and runs backwards where it's supposed to matter+0.0018 (5v5)
- Expected goals (xG)Built in-house from 156,622 shot attemptsA real model (AUC 0.758). Counting shots still beats it+0.0010 on top of shots
- Finishing (goals above xG)Four seasons, 619,505 shot attemptsRight direction, tiny size: five times under the bar to dress+0.0012
- IPPNHL scoring feedsDescribes how involved he was. Doesn't predict how involved he'll be+0.0006
- Penalty differentialNHL stats feedsLooked good in the first sweep, went flat when we checked it properlyFailed the check
- Faceoff %, total ice time, 5v5 shooting %, primary assists per 60NHL stats feedsNothing left once this season's scoring is known+0.0000
Lift is how much a stat improves the prediction of next season's points per game (R²) once this season's scoring is already in the model; (5v5) means it was measured on next season's 5v5 points per 60 instead. Most signals were tested over 7,000+ season-to-season pairs since 2009-10; xG and finishing on shot-by-shot data from 2021-22 to 2024-25. The example players in the chart above are illustrations, not real players.
Seen enough tape?
The 2026-27 beta is live on iPhone through TestFlight, Canada first. Email contact@hashmarkanalytics.com with your league size and whether it's points or cats.