Reading the Ice
How we estimate player shots, compare market prices and track published results before the puck drops.
Calibrated player probabilities
Version 4 combines a player shot-count forecast with a bookmaker reference. The probability adjustment was fitted on earlier seasons and is applied before ranking or sizing picks.
The rebuilt model improves probability accuracy over the original Poisson forecast. Its evaluated betting results have not established profitable edge. A displayed probability is an estimate, and a quoted recommendation is separate from a bet accepted by your sportsbook.
Inputs available before the game
Player profiles blend season-to-date and recent shot rates with projected ice time. Opponent shot suppression and game pace adjust the count forecast. Only prior-game observations enter these features.
All-situations shot rates already include power-play shots. There is no additional power-play multiplier. Missing, stale or ambiguous player and event data prevent a recommendation.
Reference probability and offered price
The reference removes the margin from an OVER and UNDER pair at the same book. A pair near the median of the available retail books anchors the estimate. The best supported offered price is selected separately.
For example, -130 corresponds to a 56.5% break-even probability before removing any bookmaker margin. Calculating a fair probability requires both sides of the market. A model advantage over a retail reference does not by itself prove a profitable bet.
Selection and exposure limits
| Rule | Current model policy |
|---|---|
| Contracts | UNDER 1.5 and 2.5 shots; American odds from -180 through +200. |
| Selection | At least 2 percentage points of calibrated edge and 2% modeled net EV after uncertainty and execution allowances. |
| Sizing | One tenth Kelly, calculated with an assumed 100-unit reference bankroll and rounded down. Up to 1 unit per recommendation. |
| Exposure | One contract per player, at most 1 unit per game and 4 units per day. |
Picks rank by modeled net EV. The CALIBRATED label identifies this sizing policy; it is not a guarantee of accuracy. Model units are not a personalized bankroll recommendation. Playoff publication remains under manual control.
Closing value and the public record
New CLV observations compare the closing fair probability with the break-even probability of the published entry odds, for the same player, event and line. A missing exact closing quote stays unavailable. Older raw price-movement values are not treated as verified fair CLV.
Performance includes recorded published wins and losses across regular-season and playoff picks. Synthetic results are excluded. Voided stakes do not count toward the amount wagered in ROI. Provisional injury news flags a participation check; official final results determine settlement.
Daily schedule, Eastern time
6:00 AM: Grade prior games using completed official NHL box scores.
8:00 AM: Refresh player histories and team profiles.
10:00 AM: Fetch prices, generate probabilities and publish qualifying picks.
Before each game: Capture closing prices for the exact published contract.
A rerun cannot replace the published slate or remove a settled loss. Prices can move before a sportsbook accepts a bet; check the actual contract and price when acting on a recommendation.
Full Access
- Less than $1 a day — one winning pick can pay for months of access
- +111 units of tracked profit — that's +$11,100 at $100/unit, verified in public
- Zero research needed — the exact pick, odds, and bet size handed to you every day
- Everything included — every sport, every model, every market. No tiers, no upsells