How we measure whether our pitch grading models are successful — what “good” means, the metrics we track, and the results we hold ourselves to.
2023–2025, ~1,035 pitcher-seasons (50+ IP, regular season). Bold = higher absolute correlation.
| Outcome | Our Stuff+ | FG Stuff+ | Our DLoc+ | FG Location+ | Our Pitching+ | FG Pitching+ |
|---|---|---|---|---|---|---|
| ERA | -0.394 | -0.306 | -0.053 | -0.273 | -0.462 | -0.423 |
| FIP | -0.517 | -0.373 | -0.099 | -0.332 | -0.593 | -0.516 |
| K% | +0.612 | +0.489 | -0.110 | +0.079 | +0.538 | +0.454 |
| BB% | +0.176 | -0.034 | -0.639 | -0.626 | -0.207 | -0.424 |
| WHIP | -0.327 | -0.310 | -0.234 | -0.363 | -0.492 | -0.482 |
Pearson r between year N and year N+1 grade averages, 2017–2025 (skipping 2020).
| Grade | YoY r | Threshold | Assessment |
|---|---|---|---|
| Stuff+ | 0.863 | ≥ 0.80 | Excellent — physics are repeatable |
| Pitching+ | 0.744 | ≥ 0.70 | Solid |
| DLoc+ | 0.670 | ≥ 0.55 | Acceptable — command is inherently noisy |
2025 season, 1,490 pitchers. r ≥ 0.70 = trustworthy grade.
| Grade | 50p | 100p | 200p | 500p | 1000p | Stabilizes |
|---|---|---|---|---|---|---|
| Stuff+ | 0.952 | 0.962 | 0.975 | 0.985 | 0.993 | ~50 pitches |
| DLoc+ | 0.594 | 0.672 | 0.699 | 0.746 | 0.813 | ~500 pitches |
| Pitching+ | 0.679 | 0.735 | 0.773 | 0.839 | 0.897 | ~100 pitches |
Year N grades → Year N+1 outcomes. Ratio = forward ÷ same-year correlation. Higher ratio = more predictive.
| Outcome | Stuff+ Fwd | Stuff+ Same | Ratio | Pitching+ Fwd | Pitching+ Same | Ratio |
|---|---|---|---|---|---|---|
| FIP | -0.419 | -0.487 | 0.86(0.85 or higher) | -0.406 | -0.603 | 0.67 |
| ERA | -0.358 | -0.389 | 0.92(0.85 or higher) | -0.339 | -0.463 | 0.73 |
| K% | +0.556 | +0.599 | 0.93(0.85 or higher) | +0.449 | +0.554 | 0.81 |
| BB% | +0.156 | +0.132 | 1.51(0.85 or higher) | -0.112 | -0.279 | 0.41 |
Our highest correlations vs FanGraphs are in FIP, ERA, and K% prediction for both Stuff+ and Pitching+. Stuff+ FIP correlation (-0.517 vs FG -0.373) is 39% higher. Pitching+ FIP (-0.593 vs FG -0.516) is 15% higher. Stuff+ year-over-year stability (0.863) exceeds our threshold, and Stuff+ forward prediction ratio (0.86–0.93) shows grades carry strongly to the next season.
