Value board model's confident disagreements with Vegas
💎 NYI@TOR away — model 59% vs DK +114 (45%) · gap 14%
💎 VAN@TBL away — model 33% vs DK +310 (23%) · gap 10%
💎 SJS@DET home — model 64% vs DK -135 (55%) · gap 9%
💎 SEA@FLA away — model 45% vs DK +154 (38%) · gap 7%
💎 CAR@WPG home — model 43% vs DK +154 (38%) · gap 5%
Value board record: 2-5 (29% · avg prob 46%)
Only games where our win prob beats DK's no-vig implied prob by 5+ pts, ranked by gap. Dogs and favorites both qualify — confidence is the size of the disagreement. Experimental until ~100 graded plays.
Line movement DK moneyline trail, snapshotted every ~2h to puck drop
NJD@WSH (-1.5pp to home since 08:35)+105/-125 08:35 → -102/-118 12:35away/home moneyline at each snapshot
CAR@WPG (-2.2pp to home since 08:35)-185/+154 08:35 → -205/+170 10:35 → -205/+170 12:35away/home moneyline at each snapshot
VAN@TBL (+0.0pp to home since 08:35)+310/-395 08:35 → +330/-425 10:35 → +310/-395 12:35away/home moneyline at each snapshot
SEA@FLA (+0.0pp to home since 08:35)+154/-185 08:35 → +154/-185 10:35away/home moneyline at each snapshot
NYI@TOR (-0.9pp to home since 08:35)+114/-135 08:35 → +110/-130 12:35away/home moneyline at each snapshot
BUF@MTL (+0.0pp to home since 08:35)+120/-142 08:35away/home moneyline at each snapshot
SJS@DET (+3.2pp to home since 08:35)+114/-135 08:35 → +130/-155 10:35away/home moneyline at each snapshot
OTT@PHI (+0.0pp to home since 08:35)-155/+130 08:35away/home moneyline at each snapshot
NYR@CBJ (+0.0pp to home since 08:35)+114/-135 08:35away/home moneyline at each snapshot
DAL@CHI (-2.1pp to home since 08:35)-198/+164 08:35 → -218/+180 10:35 → -218/+180 12:35away/home moneyline at each snapshot · total 5.5→6.5
STL@NSH (+0.0pp to home since 08:35)-115/-105 08:35 → -115/-105 10:35 → -115/-105 12:35away/home moneyline at each snapshot
EDM@UTA (+0.0pp to home since 08:35)+114/-135 08:35 → +114/-135 12:35away/home moneyline at each snapshot
PIT@VGK (-3.0pp to home since 08:35)+195/-238 08:35 → +180/-218 10:35 → +170/-205 12:35away/home moneyline at each snapshot
Season record logged daily, graded on finals
Moneyline: 11-2 (85% · avg prob 57.3%)
Puck line: 5-8 (38%)
Model's side pick per game, logged before puck drop and graded on finals. For calibration, not betting advice.
Recent games
| Date | Game | ML pick | | PL pick | | Score |
|---|
| 10-15 | BUF@TOR | away 58% | W | away 34% | W | 5-4 |
| 10-15 | NYI@OTT | home 52% | W | home 32% | L | 1-2 |
| 10-15 | SEA@TBL | home 58% | W | home 35% | L | 3-4 |
| 10-15 | VAN@FLA | home 64% | W | home 39% | L | 2-3 |
| 10-15 | NYR@NJD | home 52% | W | home 32% | W | 1-7 |
| 10-15 | PHI@DET | home 62% | W | home 38% | L | 1-2 |
| 10-15 | WPG@CHI | away 55% | L | away 32% | W | 3-4 |
| 10-15 | SJS@NSH | home 59% | L | home 36% | L | 3-2 |
| 10-15 | STL@MIN | home 56% | W | home 34% | L | 4-5 |
| 10-15 | UTA@EDM | away 53% | W | away 31% | L | 4-2 |
| 10-15 | CGY@VGK | home 64% | W | home 39% | W | 2-6 |
| 10-16 | PIT@COL | home 59% | W | home 36% | W | 2-5 |
| 10-16 | BOS@ANA | away 53% | W | away 31% | L | 4-2 |
NHL Model v1
Expected goals (xG) is the efficiency core — the hockey analogue of EPA. Honest probabilities, walk-forward tested before wiring. Totals excluded after backtesting showed no signal.
P(home win) = sigmoid(0.133 − 4.177·xGdiff + 0.347·goalie − 0.175·B2Bhome + 0.324·B2Baway)
Walk-forward 2022–26 (n=3,854): 56.6% winner accuracy vs 54.3% always-home baseline. Puck-line cover = P(win) × historical 2+ goal margin rate (home 60.9%, away 58.9%). Inputs: current-season 5v5 xG% regressed to prior season, team goalie GSAA shrunk to prior until ~15 games, rest flags. Starting goalies shown from DailyFaceoff; model uses team goalie aggregate until starter-specific GSAA is backtested and approved.
Early-season caveats
October–November inputs are heavily regressed toward last season — expect conservative probabilities and more coin-flip games until ~15 games played. Goalie quality takes longest to stabilize.
Value board
Flags games where model prob beats DK no-vig implied prob by 5+ pts. Tracked as a forward experiment — the NFL backtest showed model-vs-market gaps were coin flips, so this stays experimental until ~100 graded plays. Closing line value (did the market move toward our number?) is the earliest honest signal.