Wednesday, July 24, 2019

ZoneCheck - Twins' Ump De Jesus' Ball 4 Call

When home plate umpire Ramon De Jesus ejected Twins Manager Rocco Baldelli Tuesday night after pitcher Tyler Duffey's 3-2 slider to Yankees batter Luke Voit was called ball four, Minnesota TV audiences saw a graphic suggesting the ump's call was in error...but in the end, it was FoxTrax that was wrong and the umpire that was correct.

Our new feature ZoneCheck examines borderline ball/strike calls at the bottom or top of the strike zone (sz_bot or sz_top values) hours after the controversial call is made, after MLBAM has had time to process the individual pitch and batter strike zones involved.

Executive Summary: MLB's technology is presently unable to adjust the vertical values of each batter's strike zone in realtime, so what MLBAM does for the purposes of graphic representation (PitchCast, and thus, FoxTrax, ESPN K-Zone, Gameday, etc.) is insert placeholder values for each batter's typical or average strike zone (placeholder sz_bot or sz_top). After the game is over, overnight processing of the data replaces these placeholder values with actual observed values from the game. On borderline pitches near the bottom or top of the strike zone, this sometimes has the effect of changing the umpire's Quality of Correctness—generally from "incorrect" or borderline (in realtime) to "correct" (post-processed), as umpires generally do adjust in realtime to variable batter height, while the computer does not.
Related PostPodcast - Truth About Baseball's Electronic Strike Zone (6/5/19).

Baldelli and De Jesus disagree about ball four.
Prior Example: We've discussed the fallibility of MLB's PitchCast (visual) system and its susceptibility to error relating to the vertical strike zone before, and we saw an example of this error earlier in 2019 when Ron Kulpa ejected Astros coach Alex Cintron and Manager AJ Hinch over a strike one call on April 3 that the PitchCast graphics application initially depicted as an incorrect call.

A day after the game, MLBAM's routine postgame processing changed the value of the bottom of Astros batter Tyler White's strike zone, turning Kulpa's previously incorrect call into a correct one.
Related PostMLB Ejections 007-08 - Ron Kulpa (1-2; Cintron, Hinch) (4/3/19).
Related PostPodcast Minisode 13M - AJ Hinch's Mea Ron Kulpa (4/4/19).
Related PostBad Computer Umpire - Faulty Pitch Data Defames Kulpa (4/6/19).

A Mea Culpa for Ron Kulpa in April 2019.
Sidebar for Fairness and Objectivity, Gil's Call: As is usually the case, this broadcast-first system at the expense of accuracy tends to paint the umpire in the light of "he's wrong" in front of thousands while only a handful of people will actually follow up the next day and figure out that MLBAM had it wrong during the game, and that the umpire's call was actually correct.

That's why we're starting ZoneCheck, to hold the robots just as accountable as we hold the umpires whom they grade.

For the number-oriented folks amongst us, the Astros postgame processing broke down thusly:
Real-Time (in Game) Values, Kulpa's Strike 1 Call: px 0.2, pz 1.23 (sz_bot 1.51 / RAD 1.39).
Post-Game Processed (Corrected) Values: px 0.2, pz 1.23 (sz_bot 1.34 / RAD 1.22).
Net Change: sz_bot decreased by 0.17 feet (2.04 inches), changing QOC from incorrect to correct.

Back to Target Field: A similar situation occurred in regard to Luke Voit's at-bat in Minnesota, on a ball four call from HP Umpire Ramon De Jesus. During the game, FoxTrax graphics—fed by PitchCast—led us to believe the pitch was located in borderline-strike territory. After the game when MLBAM corrected its numbers, as was the case with Kulpa in Houston, the batter's strike zone changed to such a degree as to solidify the umpire's realtime on-field ruling as correct.
Related PostMLB Ejections 130-131 - Ramon De Jesus (1-2; MIN) (7/23/19).

Real-Time (in Game) Values, De Jesus' Ball 4 Call: px 0.6, pz 1.71 (sz_bot 1.75 / RAD 1.62).
Post-Game Processed (Corrected) Values: px 0.6, pz 1.71 (sz_bot 1.88 / RAD 1.76).
Net Change: sz_bot increased by 0.12 feet (1.44 inches); De Jesus' QOC = Correct.


Post a Comment