2026-27 season · picks return soon

WE CALL IT
BEFORE PUCK DROP.
WE TRACK IT
AFTER.

Every NHL bet we make goes on the record before the puck drops. Shots, points, assists, and now goals, moneyline and totals. Wins and losses.

See the full ledger → How the model works
2026 season opener
…
0/0
2026-27 record · 0 graded picks
0.00% ROI
Tonight’s slate

Every NHL game tonight, one board.

Records, puck drop, the best moneyline/total price we can find across every major book, and the model's own grade and pick on every game once the season's underway. This is the full board, not just our picks. See the Ledger for what we actually bet.

Last night’s results

What we posted, graded.

Every pick from the most recent night we posted, win or lose. No edits after the fact.

Free pick by email

Get our top free pick by email.

Drop your email and we’ll send our best free pick when we’ve got one. No spam, no other lists.

Get the Free Pick →

Picks post before puck drop, every night of the season.

Follow along on X and Reddit for the daily card, or come back here for the full ledger. Win or lose, it's all on the record.

Follow on X → Follow on Reddit
The public ledger

Every pick, every result.

The full record across every season: 164 player-prop bets across shots, points, assists, and goals, filterable by market. Moneyline and totals picks are new for 2026-27 and will show up here too, updated as real games get graded this season. Nothing here has been edited or removed after the fact. Every bet tracked here is a real Power Play pick, the same ones our paying members see live, before puck drop.

2025-26
87/77
Record
−0.51%
ROI
2026-27
0/0
Record
+0.00%
ROI
Staking changed this season
All-time
+0.33%
Avg. closing line value (164 picks)
2025-26 only, for now
2025-26 by market
+2.81%
Shots ROI (110 picks)
−10.45%
Points ROI (49 picks)
+24.00%
Assists ROI (5 picks)
Small sample
2026-27 by market
+0.00%
Moneyline ROI (0 picks)
+0.00%
Totals ROI (0 picks)
+0.00%
Shots ROI (0 picks)
+0.00%
Points ROI (0 picks)
+0.00%
Assists ROI (0 picks)
+0.00%
Goals ROI (0 picks)
Cumulative units, every graded pick in order
Moneyline & totals picks, 2026-27
DateMatchupMarketPickOddsGradeResultUnits

Player prop picks, every season
DatePlayerMatchupMarketLineOddsResultGradeUnits

What we price

Six markets, one model.

Player props built the 2025-26 track record. Moneyline and totals are new for this season. Same model discipline, just no results yet because there aren't any.

Shots on Goal

Tracked

Reworked ahead of 2026-27 after this market carried the book. It was the only one of the four original prop lines that finished the season profitable.

55.0%Win rate
+2.29%ROI
109Picks

Points

Rebuilt

Losing money in every odds range last season, favorites and underdogs alike, which pointed at the projection math, not variance. Rebuilt on a proper Poisson model for 2026-27.

46.9%Win rate
−10.45%ROI
49Picks

Assists

Rebuilt

Only 5 graded picks last season, too small a sample to mean anything, good or bad. New shrinkage-based model built for real volume this year.

60.0%Win rate
+24.0%ROI
5Picks

Goals

Rebuilt

Never actually posted a graded pick last season. A broken data reference silently zeroed out every projection. Found it, fixed it, rebuilt the model clean.

No picks logged yet this season.

Moneyline

New

Team-level win probability: a Poisson score model built off each team's offense/defense rate and real, measured home-ice edge. Expanding past player props for the first time this season.

No picks logged yet this season.

Totals

New

Game-level over/under, also new for 2026-27. Same live, public grading standard as everything else on this page from pick one.

No picks logged yet this season.
ROI vs. breakeven
Shots on Goal
+2.29%
Points
−10.45%
Assists
+24.0%

0% is breakeven. A market only counts as beaten if it clears that line, not just if the win total looks decent. Goals, Moneyline, and Totals don't have a season of results yet, so they aren't shown here.

Under the hood

How a projection gets made.

  1. 1

    Shrink the rate

    Season and last-10-game rates get blended, then regressed toward league average by sample size, so a hot 3-game stretch doesn't get trusted at face value.

  2. 2

    Adjust for the matchup

    Opponent defense and power-play strength scale the rate up or down, measured against real league-average allowed rates, not an arbitrary multiplier.

  3. 3

    Price it correctly

    The adjusted rate becomes a true Poisson probability for the actual line, P(X ≥ line), not a linear guess dressed up as a percentage.

  4. 4

    Compare to the market

    Model probability vs. the book's own price. A pick only posts when the gap clears a real threshold, not padded out to hit a quota.

Read this before you follow a pick

No model here is guaranteed to be profitable. −0.49% ROI over 164 bets last season is the honest starting point, not a pitch. Sportsbooks price lines efficiently, so a real edge is hard to find and easy to talk yourself into on a small sample.

Every number on this page is real, not backtested-and-polished. If you follow these picks, start small and track your own results independently.

Does the model know what it doesn't know

Calibration, against 164 real bets.

When the model says a bet has a 60% chance to win, does it actually win about 60% of the time? This checks every 2025-26 bet against what actually happened, grouped by the model's own predicted probability at the time.

PredictedSampleActual win rate
~30%3 bets33.3%
~40%31 bets61.3%
~50%76 bets43.4%
~60%50 bets62.0%
~70%4 bets75.0%

The 50% bucket carries the most bets by far and runs cold (43.4% actual). The 30% and 70% buckets are too thin (3 and 4 bets) to read much into either way.

Do the letter grades mean anything

Real win rate by grade tier.

Every graded pick ever, bucketed by its letter grade at the time it was made. D and F are combined here since F alone is a single bet, too thin a sample to read on its own.

A-tier
65.0%20 picks
B-tier
49.3%73 picks
C-tier
57.1%56 picks
D/F-tier
40.0%15 picks

A clearly outperforms the rest. B and C are close enough at this sample size that the ordering between them isn't meaningful yet. More picks will sharpen this.

Season over season

What actually changed since last year.

This isn't a claim that the model got better. These are the specific things that were found broken and fixed, plus what's new.

Points was rebuilt, not retuned

Last season's Points projections stacked manual multipliers into a point estimate, then compared it to the betting line as if that were a real probability. It wasn't, and that's the direct cause of Points' −10.45% ROI. It now runs the same Poisson framework that's produced Shots' real edge (+2.81% ROI, 110 picks).

A real data bug, found and fixed

Player stat ingestion briefly mixed Olympic games into regular-season rolling stats for a roughly 10-day window last February, right before this site started publicly tracking picks. Fixed by filtering to real NHL games only, then reprocessing the affected window.

Moneyline and totals are new

Last season only tracked player props. A separate model now prices full-game moneyline and totals with the same approach, shown on the Ledger as its own category once real 2026-27 games start.

Smaller refinements

Teams now carry a measured penalty for playing on zero days' rest, and penalty-kill strength uses each team's real, point-in-time number instead of a season-average placeholder.

A sample-size bug, found and fixed

The moneyline/totals model is supposed to trust a team's own results more as real games pile up, and lean on the league average early when it doesn't have much to go on yet. A bug meant it always assumed exactly 20 games had been played, for every team, all season long, whether it was opening night or February. Fixed, and it measurably mattered most in exactly the games early in a season, right when trusting a small sample the wrong amount is most costly.

Real shot quality, not just goals scored

Moneyline and totals now weigh each team's real shot quality (distance, angle, shot type, rebounds) instead of raw goals scored, which treats a lucky bounce and a well-earned goal as the same thing. Built from the NHL's own public shot data, not a third-party feed, and checked against a full season of real results before it was trusted with a live pick.

Handling a new season honestly

Early in a new season there's no real current-year data yet for any team. The model used to quietly blend in last year's results with no real logic for when to stop leaning on them. It now treats a team's own last-season number as a deliberate starting point, and lets real current-season results take over within roughly the first 8 games, the same way a careful bettor should get more confident as a small sample grows into a real one.

None of this is a promise the record improves. See the calibration above and the disclosure on this page. It's what was actually wrong, and what's actually different now.

League context

Toughest matchups right now.

Every team ranked by goals allowed per game, the core input behind every opponent adjustment the model makes.

2025-26 season data. The 2026-27 season hasn't started yet, so this is last season's final numbers. Switches to live 2026-27 stats automatically once real games are played.
#TeamGF/GPGA/GPSF/GPSA/GPPP%PK%
League standings

Every team, every division.

Live NHL standings. Frozen on last season's final table until the new season has games played.

Head to head

Compare any two teams.

Same GF/GA/PP/PK numbers behind every model pick, side by side.

VS
Power Play

Every pick’s public.
Power Play is the advantage.

Every game we track is public, win or lose, forever. Power Play unlocks the plays we’d actually put money on: the model’s win probability, the edge over the market, and the stake size, before puck drop.

  • Picks every night
  • Posted before puck drop
  • Tracked publicly
You were invited by a friend. Mention code when you sign up for Power Play and you’ll both get a free week.
87-77
All-time record
53.0%
Win rate
164
Graded picks
Sample card · not a real pick
BOSBOS Moneyline −135
61%Model win %
+7.2%Edge
2uStake size
Sample card · not a real pick
EDMMcDavid Shots O3.5 −115
58%Model win %
+5.4%Edge
1uStake size
Monthly
$14.99/mo

Billed monthly, September through June. Pauses automatically once the season ends, no charges over the summer and nothing to cancel.

Start Monthly
Season Pass
$79.99/season

One charge covers the whole season, playoffs included, plus continuous access into the offseason. Renews automatically next September.

Get the Season Pass

By subscribing you agree to our Terms.

Refer a friend

Give a free week. Get a free week.

Share your link. When a friend signs up for Power Play and mentions your code, you both get a free week added to your plan.

Your code: · include it on the sign-up form so we can credit you both
FeatureFreePower Play
Every NHL game tracked publicly, win or lose✓✓
Tonight’s full slate: records, odds shopped across books✓✓
Head-to-head history on every matchup✓✓
Team defense rankings and model methodology✓✓
Live standings and head-to-head team comparison✓✓
Real track record: calibration and closing-line value, not just win/loss✓✓
Top free pick by email✓✓
The plays we’d actually bet, before puck drop—✓
Model win probability and edge on every play—✓
Stake sizing on every play—✓
Full player-prop projections for every prop on the board, not just posted picks—✓
Full moneyline and total model projections for every game, not just shopped odds—✓
Real power-play ice-time data behind every skater’s prop, not just a PP1/PP2 label—✓
Glossary

How this all works.

The terms used across this site, explained plainly.

Edge

The gap between our model's win probability and the market's implied probability from the posted odds. A positive edge means the model thinks the real chance of winning is higher than the price suggests.

Units

A unit is whatever stake size you choose, usually a fixed percentage of a betting bankroll. Results are reported in units rather than dollars so the record means the same thing regardless of bankroll size. Through the 2025-26 season, every graded pick was staked flat at 1 unit. Starting 2026-27, Power Play stake sizing scales from 1 to 3 units based on model edge, so the record going forward blends two different staking approaches across seasons.

Vig (juice)

The built-in cut a sportsbook takes on every bet, priced into the odds. It's why both sides of a true coin-flip bet aren't offered at even money.

ROI

Return on investment: units won or lost divided by units risked, as a percentage. It's the cleanest single number for whether a strategy is actually profitable over time, independent of how many bets were made.

Letter grade (A+ through F)

A read on how large the real model edge was on a given pick at the time it was made, not on how the bet turned out. A good grade doesn't guarantee a win; a bad grade doesn't guarantee a loss. It measures the bet, not the outcome. See the Model page for the exact cutoffs and real win rates by tier.

Moneyline

A bet on which team wins the game outright. No point spread or margin involved, just the final result.

Totals (Over/Under)

A bet on whether the combined final score of both teams lands over or under a set number.

Closing line value (CLV)

How the odds moved between when a pick was made and when the game actually started. Consistently beating the closing line is one of the better long-run signals that a model has real predictive edge, separate from whether any single bet won.

Terms

Power Play terms.

Plain-language terms for the paid Power Play subscription. Last updated September 12, 2026.

What you're paying for

Power Play is access to the full output of a statistical model: win probabilities, edges, and stake sizing on every game and player prop we track, before puck drop. It is informational and for entertainment purposes only. It is not financial advice, not gambling advice, and not a guarantee of any result. See the disclaimer at the bottom of every page.

Eligibility

You must be of legal age to wager in your jurisdiction to subscribe. Sports betting is regulated differently everywhere; it's your responsibility to know and follow the laws where you live.

Billing

Payments are processed by Stripe. Monthly is billed each month during the season and pauses automatically in the offseason. Season Pass is one charge for the full season and renews automatically the following September unless canceled first. We don't store your card details ourselves.

How you get access

After payment, we email the private Power Play link to the address you paid with, usually within a day. That link is for your use only; please don't share or repost it. There's no separate login, so treat the link itself as your access credential.

Cancellation

Manage or cancel your subscription yourself, any time, using the button below. Enter the email you subscribed with and we'll send you a secure link to your account. Cancellation takes effect at the end of your current billing period; you keep access until then, and you won't be charged again after that.

Manage or Cancel Subscription

Prefer email? You can also reach us directly to cancel.

No guarantee of results

Every number on this site, free or paid, is real and unedited, but past performance doesn't guarantee future results. Betting carries real financial risk. Never bet more than you can afford to lose.

Contact

Questions, refunds, or cancellations: email Proplinehockey@gmail.com.