Pitch Profiler has open-sourced a per-swing reconstruction of Baseball Savant swing-timing data that cuts error against Savant’s pitcher rates by 57.2%, according to the project’s README. The company announced the release in a post on X.
Baseball Savant publishes bat contact position, timing and swing-plane height as season aggregates, Pitch Profiler said. The project reconstructs those three measures for individual swings from public Statcast fields.
The project’s reported 57.2% reduction compares a null mean absolute error of 0.0475 with a model mean absolute error of 0.0203 for pitcher rates in the 2024 holdout season. The README calculates the reduction as the difference between null and model error, divided by null error.
The repository reports pitcher-rate mean absolute error of 0.0203 in the 2024 holdout season, with an R² of 0.876. It reports 0.0205 and an R² of 0.861 for the 2025 fit season, and 0.0209 and an R² of 0.864 for the 2026 holdout season.
The reconstruction uses exit-velocity inversion for 69% of swings, gradient-boosted whiff modeling for 26%, and geometry fitting for 5% of swings. The geometry method covers fouls that lack exit-velocity data. For batter rates, the project reported a 23% error reduction and attributed the smaller gain to greater per-player stance variation.
Pitch Profiler’s site describes the service as “Your Personal Front Office from Anywhere.” The repository’s author, GitHub user jmaschino56, has a profile bio reading 'Senior Data Analyst and Creator of Pitch Profiler.'
