Every projection we publish is graded against what actually happened on the court. When our model says a prop has a 60% chance to hit, it should hit about 60% of the time — here's the proof, good or bad.
We sort every graded prop into buckets by the hit-probability our model gave the side we projected, then check what fraction actually hit. A perfectly honest model sits on the diagonal: the dots should hug the dashed line. Where a dot falls below the line, the model was over-confident in that band.
| Predicted | Hit (actual) | Miss | Props |
|---|
Brier score is the average squared error between the probability we gave and what happened (lower is better; 0.25 = a coin flip, 0 = perfect). Straight-up accuracy is how often the side our model leaned (≥50%) actually hit. Calibration is the most important: across the whole range, our stated probabilities should match real-world hit frequencies.
These are measured on the same minutes-driven projection model we use to price every prop on the board — not a curated subset. NBA player-prop lines are a sharp, efficient market, so honest calibration sitting near a coin flip is expected; we'd rather show you the real curve than a flattering one.