Pitchers PlanetWhat the calculator measures
Pitchers Planet runs the Trackman Analyzer models supplied for this app. Stuff+ estimates pitch value from physical pitch characteristics. Pitching+ adds the pitch’s location at the plate. These are independent model outputs, not official TrackMan ratings or a reproduction of another provider’s Stuff+ scores.
The deployed version is trackman-analyzer-2026-10. Each model uses 16 LightGBM event models, followed by a trained regression model that combines their outputs into a pitch run-value estimate. The final estimate is converted to a plus scale.
Inputs: more than velocity
Stuff+ uses 13 inputs: release speed, induced vertical break, horizontal break, spin rate, extension, release height, release side, pitch-type code, throwing hand, and four differences from the pitcher’s fastball baseline. Those differences cover velocity, vertical break, horizontal break, and spin rate.
Pitching+ uses those same inputs plus PlateLocSide and PlateLocHeight. Pitch count, batter handedness, and game situation are not model inputs. See the TrackMan data guide for column names and units.
How Stuff+ is calculated
The model predicts pitch value first. The conversion to a readable score is:
Score = 100 + 10 × (reference mean − predicted run value) / reference standard deviationLower predicted run value is better for the pitcher, so it produces a higher score. Stuff+ and Pitching+ each use their own saved reference parameters. Scores are not clipped or recalibrated to the players in your upload.
| Score | Interpretation on this saved scale |
|---|---|
| 90 | One reference standard deviation below the center |
| 100 | At the model’s saved reference mean |
| 110 | One reference standard deviation above the center |
This is not a percentile, an ERA projection, or a percentage improvement in strikeouts. Do not assume the reference population is equivalent to today’s MLB, a particular conference, or your own roster.
Why your fastballs matter
The calculator builds a separate baseline for each pitcher using recognized fastball-family pitches in the selected sessions: fastballs, four-seams, two-seams, sinkers, and one-seams. Each baseline measurement averages its available numeric values. The model uses baseline minus pitch measurement for its four difference inputs.
Pitch-type filters share that baseline. Changing the selected sessions can change it, even when an individual pitch’s raw measurements stay the same. For meaningful before-and-after comparisons, record the session selection and model version with your exported report.
Missing data and score averages
Scoring requires an identified pitcher, recognized throwing hand, supported pitch type, all seven pitch-shape and release measurements, and a usable fastball baseline. Missing plate location leaves only Pitching+ unavailable. No default release point or guessed fastball profile fills the gaps.
Report averages include only scored pitches. Check the scored-pitch count alongside the score: a report based on a small or incomplete subset does not describe every pitch in the session. Players are grouped by normalized name and team so reused source IDs do not merge their reports.
What has—and hasn’t—been validated
The deployed inference implementation was compared with the supplied models using an independent Python reference across 7,722 feature vectors, including session pitches and location grids. Those checks establish that the implementation reproduces the supplied model’s calculations.
They do not establish predictive accuracy on unseen pitchers. Training splits, held-out accuracy metrics, and a complete training-population audit were not supplied with this model bundle. We therefore do not advertise a validated accuracy percentage or guaranteed performance improvement.
Using location targets responsibly
The Location Lab evaluates average pitch shape at 3,721 locations and suggests a target using the selected model metric and location spread. Observed spread includes intentional target changes as well as execution variation; it is not a direct measurement of command error. Count and batter handedness are not included. Use the map as a pitch-design aid alongside video, scouting, and coaching judgment.
Why other Stuff+ scores can differ
Different providers can use different training data, inputs, outcomes, and reference scales. Compare trends within the same model rather than treating identical numbers from different systems as equivalent. For background on a separate system, read FanGraphs’ Stuff+, Location+, and Pitching+ primer; its model descriptions should not be assumed to apply to this calculator.