Saturday, September 21, 2019

Postgame Processing Changes Gibson's Strike EJ QOC

Dodgers Manager Dave Roberts became the sixth person ejected in 2019 arguing an umpire's ball/strike call that looked incorrect in real-time, but was later changed to 'correct' when MLB's postgame pitch processing changed computer zone numbers to flip Greg Gibson's QOC from N to Y. When Gibson ejected Roberts Friday night in Los Angeles, I cautioned this exact scenario might happen (as it has happened many times before), which is precisely what occurred as MLBAM's vertical StatCast/PitchCast problem struck again and transmogrified yet another big league ejection well after the game's final out.

To recap, Gibson tossed Roberts following a 2-0 pitch ruled a strike that passed over the center of home plate and vertically near the hollow of LA batter Gavin Lux's knee.

At the time, the computer (that'd be TrackMan near what could become the final phase of its Major League Baseball existence) deemed the pitch an unequivocal ball, low: sz_bot 1.6 and pz 1.35. Accounting for margin-of-error, the pitch was less than half an inch away from borderline "call stands" territory: QOC = Incorrect. In fact, the preceding ball call was located higher off the ground than the subsequent strike (ball: pz 1.41, strike: pz 1.35).
Related PostMLB Ejection 212 - Greg Gibson (2; Dave Roberts) (9/20/19).

Watch out for inaccurate real-time FoxTrax.
Postgame Processing Changes Strike Zone: Hours after all the fans and players at Dodger Stadium had cleared out, MLB's secretive postgame processing went to work to calibrate each player's strike zone for each individual semi-borderline pitch the hitter faced.

For Gavin Lux in the 4th inning, the numbers were changed thusly: sz_bot for the ball call increased to 1.69 ft, while sz_bot for the strike call decreased to 1.56 ft—a difference of 0.13 feet, or 1.56 inches.

The vertical phenomena happened before, too.
Notice that the ball thrown higher off the ground than the strike (pz 1.41 vs 1.35) was captured at a point 0.06 feet, or 0.72 inches higher than the strike. Upon normalizing the strike zone (e.g., accounting for the 1.56 inch-difference in sz_bot), we see that, in relative units, the pitch ruled "ball" was actually located lower than the pitch ruled "strike," relative to Lux's hollow-of-knees.

Lessons: The robot umpire concept simply doesn't work if we have to wait for overnight processing to conclude before knowing whether a given pitch was a ball or a strike. It's not good enough to have an automated system call "strike three," only to find out several hours later that, "sorry, it should have been ball four."

Second, this isn't even the first time this season that postgame processing has exposed the vertical strike zone blind spot of the pitch-tracking computer. Earlier in 2019, it happened to Ron Kulpa in Houston, it happened to Ramon DeJesus in Minnesota, it happened to Jeremie Rehak in Anaheim, and now it has happened to Greg Gibson in LA. All four times, it took until the next day for the umpire to be vindicated by revised pitch tracking numbers.
Related PostMLB Ejections 130-131 - Ramon De Jesus (1-2; MIN) (7/23/19).
Related PostMLB Ejection 077 - Jeremie Rehak (4; Brad Ausmus) (6/9/19).
Related PostMLB Ejections 007-08 - Ron Kulpa (1-2; Cintron, Hinch) (4/3/19).

Third, post-game processing solely goes to work, generally, on sz_bot and sz_top, meaning it tackles the vertical zone problem but not other sources of error, including calibration or tracking. For instance, postgame processing does not account for the three-dimensional strike zone problem posed by a tracking system that sees things in two dimensions (at the front edge of home plate).

For more information about UEFL ZoneCheck and MLB's habitual vertical strike zone problem, see the following related articles.
Related PostZoneCheck - Twins' Ump De Jesus' Ball 4 Call (7/24/19).
Related PostBad Computer Umpire - Faulty Pitch Data Defames Kulpa (4/6/19).

Video as follows:

Alternate Link: Postgame Processing Strikes Again - Greg Gibson's Ejection Revisited (CCS)


Post a Comment