Skip to content
HabsMetrics

Model accuracy

How accurate are HabsMetrics NHL predictions? Every projection the model published is graded after the final horn: winners picked, log-loss and calibration, compared with a coin flip and always picking the home team. Updated nightly. Analytics for educational purposes only — not gambling advice.

94 games graded since Oct 25

Winners picked

60%

Always picking the home team: 51%

Log-loss

0.649

Coin flip: 0.693 · lower is better

Brier score

0.229

Coin flip: 0.250 · lower is better

Goal error

1.36

Average miss per team, in goals

Accuracy, week by week

When the model says X%, does it happen X% of the time?

Games grouped by how strongly the model favoured one team. A well-calibrated model’s favourites win about as often as it said they would.

Favourite’s chanceGamesModel saidFavourite won
50–55%4153%49%
55–60%2657%62%
60–65%1462%57%
65–70%8Too few games
70–100%5Too few games

Latest results

DateGameModel pickFinalResult
Nov 7TBL @ MINMIN 52%4–3Wrong
Nov 7CHI @ CARCAR 79%2–9Right
Nov 7SJS @ DALDAL 69%3–4Right
Nov 7UTA @ OTTOTT 55%2–4Right
Nov 7NJD @ NYINJD 54%5–4Right
Nov 7MTL @ TORMTL 50%4–2Right
Nov 7NYR @ SEANYR 53%4–3Right
Nov 7STL @ NSHNSH 53%1–3Right
Nov 7FLA @ WSHWSH 54%3–4Right
Nov 6SJS @ MINMIN 71%1–4Right
Nov 6ANA @ CGYCGY 54%5–3Wrong
Nov 6CBJ @ BUFBUF 58%2–5Right
Nov 5NSH @ COLCOL 60%3–2Wrong
Nov 5ANA @ EDMEDM 65%5–4Wrong
Nov 5CHI @ TBLTBL 75%1–4Right
Nov 5DAL @ CBJDAL 52%1–2Wrong
Nov 5VAN @ WPGWPG 59%2–6Right
Nov 5BOS @ PITBOS 55%6–2Right
Nov 5CAR @ NYICAR 62%5–3Right
Nov 5VGK @ DETVGK 57%2–1Right

Log-loss: how far the stated probabilities were from what happened, punishing confident misses hardest (saying 50% every game scores 0.693). Brier score: the average squared gap between the probability and the result (0 is perfect, 0.25 is a coin flip). Calibration: whether 60% favourites win about 60% of the time.

For educational/analytics purposes only — not gambling advice. Model version poisson-dc-v1.