Top Shelf Analytics

For shot prop-style predictions, overall accuracy can be misleading. A model can look accurate by correctly identifying many players who finish below the betting threshold, but those are not the picks users are usually trying to act on.

This public preview focuses on 2+ shots predictions, where the goal is to identify players with a stronger chance of reaching the threshold. Because fewer players reach 2+ shots than fall below it, accuracy alone may overstate how useful the model is for finding actionable picks.

Precision measures how often selected picks hit. Recall measures how many of the available hits the model is able to capture. F1 Score balances both, which makes it useful when trying to keep pick quality high without shrinking the player pool too aggressively.

The default minimum F1 Score is set to 75. This threshold helps balance pick quality and player coverage, keeping the model focused on stronger predictions while still providing a meaningful pool of players across each slate.

Each row represents one player prediction for 2+ shots on a given prediction date.

Prediction probability and model performance metrics come from machine learning model results. Actual results are updated using the shots column from Skater Summary.

Distribution metrics include mean, median, mode, skew, and concentration. Concentration measures the share of prior games where a player’s shot total finished within ±1 of their median shot total as of the model cutoff date.

Pending rows indicate games where the prediction exists, but final shot results have not yet been joined.