PitchingBot Visualizer

Updating DataExperimental Metrics
Changelog

PitchingBot grades, on a 20-80 scouting scale, overlaid with the pitch locations that produced those grades. Filters by pitch type, count, and batter handedness, for any season since 2020. Graphics and analysis based on Stuff, Command, or Overall grades.

PitchingBot is a model that uses pitch-level characteristics to produce an estimate of pitcher quality. Pitching models have gained popularity across baseball, both in the public sphere and on the team side, because they deliver information about pitcher talent level more quickly than on-field results. We think that seeing PitchingBot results displayed graphically helps make the model’s outputs more intuitive.

Command
Stuff
Overall
Off
On
All Batters
  • LHH
  • RHH
  • All Pitches
    0-0
  • All Counts
  • Two Strikes
  • Three Balls
  • 2026
  • 2026
  • 2025
  • 2024
  • 2023
  • 2022
  • 2021
  • 2020
  • 2019
  • 2018
  • 2017
  • 2016
  • 2015
  • Navigate forward to interact with the calendar and select a date. Press the question mark key to get the keyboard shortcuts for changing dates.

    Navigate backward to interact with the calendar and select a date. Press the question mark key to get the keyboard shortcuts for changing dates.

    Glossary
    PitchingBot is a system that uses the characteristics of individual pitches to estimate pitcher quality. It's based on a gradient boosting method that uses inputs like pitcher handedness, batter handedness, strike zone height, count, velocity, spin rate, movement, release point, extension, and location to predict the likelihood of various outcomes. Those likelihoods are then converted into a grade on the 20-80 scale. PitchingBot produces three grades for every pitch:
    • Stuff grades are based on physical characteristics of a pitch: velocity, movement, spin, release point, and the like. They ignore count and location, focusing entirely on how the ball moves and where it starts from. Stuff stabilizes more quickly than other metrics of pitch quality.
    • Command grades are based on location and count. They consider pitch type (optimal breaking ball location is different from optimal fastball location), but ignore the particular characteristics of each pitch, focusing only on where they cross home plate.
    • Overall grades use all of the available information about a pitch, including everything that goes into the Stuff and Command models.

    The PitchingBot visualizer includes optional count-by-count analysis. The three metrics currently available in our analysis suite are:
    • Danger Miss Percentage: The percentage of the selected pitches that have below-average command grades, below-average overall grades, and are located within the strike zone. In other words, these pitches are located in hittable areas that are unlikely to result in called strikes or whiffs (below-average command grade) and don't possess physical characteristics that offset that poor location (below-average overall grade).
    • Waste Miss Percentage: The percentage of the selected pitches that have below-average command grades, below-average overall grades, and are located outside the strike zone. These pitches are located in areas that are unlikely to draw a swing (below-average command grade outside the zone), and they don't have enough movement to entice poor swings (below-average overall grade).
    • Squandered Percentage: The percentage of the selected pitches that have excellent stuff but aren't thrown in a location that challenges the opposing hitter. A pitch counts as "squandered" if it ranks among the top third of the selected pitches in stuff grade while also receiving a command grade of 40 or lower on the 20-80 scale, outside the strike zone. Squandered pitches have the potential to miss bats, but they're located so poorly that they rarely draw a swing.