Slate accuracy report
2026-10-29 · 11/11 games final · goalie-v2-unified
October 30 games have not started. This report evaluates the October 29 slate that just finished, using predictions saved before puck drop—not regenerated postgame forecasts.
Probability accuracy · half-goal thresholds
Lower Brier and log-loss are better. Binary Brier ranges from 0 (perfect) to 1; log-loss is in natural-log units. A fixed 50% forecast scores 0.25 Brier and 0.6931 log-loss.
| Line | Mean P(Over) | Observed Over | Brier | Log-loss | Direction correct |
|---|---|---|---|---|---|
| Over 5.5 | 57.9% | 7/11 (63.6%) | 0.2350 | 0.6621 | 7/11 (63.6%) |
| Over 6.5 | 42.0% | 7/11 (63.6%) | 0.2781 | 0.7511 | 5/11 (45.5%) |
Over 5.5 was modestly better than the uninformative 50/50 reference on this slate. Over 6.5 was worse, with an average forecast of 42.0% versus 63.6% observed. All 11 Over-5.5 forecasts exceeded 50%, so its 7/11 directional result also equals an always-Over rule; this alone does not establish matchup discrimination or a betting edge.
Integer lines · pushes treated as a third outcome
These are three-class Under/Push/Over scores, not the binary scores above. Multiclass Brier here is the sum across classes (range 0–2). No pushes are silently graded as Under or as losses.
| Line | Actual U / P / O | Mean P(Under) | Mean P(Push) | Mean P(Over) | 3-class Brier | Log-loss |
|---|---|---|---|---|---|---|
| 6 goals | 4 / 0 / 7 | 42.1% | 15.9% | 42.0% | 0.5383 | 0.8680 |
| 7 goals | 4 / 4 / 3 | 58.0% | 14.0% | 28.0% | 0.7553 | 1.2568 |
Four games finished on exactly seven goals: pushes occurred in 36.4% of games versus 14.0% forecast on average. Strict Over 7 occurred in 27.3%, close to its 28.0% average forecast. On this sample, the Over-6.5 gap therefore includes more exact-seven results than projected, not simply more games strictly above seven.
Calibration bins
Five fixed probability bins, empty bins omitted. Each interval is a 95% Wilson interval for the observed frequency; these tiny bins cannot establish stable calibration.
| Threshold | Forecast bin | Games | Mean forecast | Observed | Observed 95% interval |
|---|---|---|---|---|---|
| Over 5.5 | 40%–60% | 8 | 55.5% | 62.5% | 30.6%–86.3% |
| Over 5.5 | 60%–80% | 3 | 64.5% | 66.7% | 20.8%–93.9% |
| Over 6.5 | 20%–40% | 4 | 36.8% | 50.0% | 15.0%–85.0% |
| Over 6.5 | 40%–60% | 7 | 45.0% | 71.4% | 35.9%–91.8% |
Game-by-game archived probabilities
P(Over) is scored against official final totals, including overtime and any shootout winner bonus. First and latest pregame snapshots are identical for all 11 games.
| Game | Predicted total | Final A–H | Actual total | P(Over 5.5) | P(Over 6.5) | Over 6.5 result |
|---|---|---|---|---|---|---|
| WPG @ TBL | 6.24 | 4–6 | 10 | 59.3% | 43.3% | OVER |
| UTA @ PIT | 6.62 | 4–3 | 7 | 64.8% | 49.3% | OVER |
| CBJ @ CAR | 6.20 | 2–3 | 5 | 58.7% | 42.7% | UNDER |
| CHI @ DET | 5.90 | 1–4 | 5 | 53.8% | 37.7% | UNDER |
| BOS @ STL | 6.00 | 1–4 | 5 | 55.5% | 39.4% | UNDER |
| WSH @ NSH | 6.16 | 4–3 | 7 | 58.0% | 41.9% | OVER |
| MTL @ DAL | 6.06 | 5–2 | 7 | 56.4% | 40.3% | OVER |
| NYI @ MIN | 5.74 | 4–3 | 7 | 51.1% | 35.2% | OVER |
| BUF @ ANA | 6.70 | 7–6 | 13 | 65.9% | 50.5% | OVER |
| OTT @ LAK | 5.73 | 5–6 | 11 | 51.1% | 35.1% | OVER |
| VAN @ SJS | 6.47 | 2–3 | 5 | 62.7% | 46.9% | UNDER |
Goal-mean accuracy and interpretation
Mean prediction 6.17 versus actual 7.45 goals per game. Total MAE 2.12, RMSE 2.86, team MAE 1.40. Scoring was underpredicted by 1.29 goals per game on average.
All 22 goalie inputs were estimated; none was recorded as independently confirmed. Eleven games are far too few to establish season-level probability calibration. The observed Over rate of 7/11 has a wide 95% interval of 35.4%–84.8%. Do not tune coefficients to this slate or interpret fixed-threshold side accuracy as sportsbook ROI.