DT K

Returning to baseball research Twitter in 2026 be like:

@RTS_Owen Bit too ahead of the time with the anti framing thing

Just added Catcher Framing + Catcher Challenging to our full game MLB sim. Was doing some validation, and thought I made an error when we had MJ Melendez as the worst catcher in our data set. Turns out he caught 71 games in 2022 and amassed an impressive -16 framing runs and -26 blocks above average!

@aidan_throws @MLBONFOX DM sent

@tomdoyo @MLBONFOX Would love to connect. I am at UWaterloo working with the DIESEL lab and working on pitching command research

I spent 100s of hours grinding out manual labels this summer so that we can measure command! How? By measuring the distance between the catcher's target VS where the ball landed. There's more to come, but here's 10 pitches from the All-Star Game: (📽 @MLBONFOX)

@MattKugler @MLBONFOX Me too

@tomdoyo @MLBONFOX I’ve been dreaming about this for years

@Timelville @MLBONFOX Yeah it's definitely a common sight. There's a subset of pitchers who start their pitches at the glove. I'm looking to identify who they are

@tomdoyo @MLBONFOX Kudos to you for grinding on this! Curious if you saw any intended “start the pitch here” glove signaling? Instead of intended end position it’s use as a reminder to start the curveball above the zone for example.

@ChienLiHo1 @BachTalk1 @MLBONFOX Very interesting, could you elaborate?

@tomdoyo @BachTalk1 @MLBONFOX Instead of absolute miss is it possible to measure miss in terms of how tight the miss is related to glove? Like a Polar Coordinate System?

@bravesblood91 @8_Maxx @MLBONFOX Do you mean taking clues from the pitcher's mechanics somehow?

@tomdoyo @8_Maxx @MLBONFOX Is there a way to implement pitcher arm slot or release point, to catcher end point of catch? If you keep everything you have but try to implement pitcher position it may help with this.

@HutchNorris @8_Maxx @MLBONFOX So variance (location apread) given target?

@tomdoyo @8_Maxx @MLBONFOX I think that it is a fundamental miscalculation of command. Instead of the distance from the spot being a measure of command, it should be about the standard deviation of the distance of the pitches from the target location.

@druidofparanor @RTS_Owen @8_Maxx @MLBONFOX Interesting, what if it's a rookie pitcher?

@RTS_Owen @tomdoyo @8_Maxx @MLBONFOX You can use the historical distribution of the same pitch type for a pitcher/catcher combo. If something like 50pct fall in the yellow and only 20pct fall in the green, you can infer that is the target.

@4the__love @MLBONFOX Stay posted 👍

@tomdoyo @MLBONFOX Damn is anybody command good

@JermelFuego @MLBONFOX It'll be open, keep posted 📽

@tomdoyo @MLBONFOX Where can I buy this ?

@TmpstXyz @MLBONFOX Yolo11 👍

@tomdoyo @MLBONFOX this is awesome! what model(s) are you using? a friend is trying to do similar things for rowing form.

@T3HT0RCH @MLBONFOX Not yet!

@tomdoyo @MLBONFOX Does it account for breaking pitches? Catchers often set targets for where the pitch ought to start, not where it should finish.

@RTS_Owen @8_Maxx @MLBONFOX Interesting idea

Without thinking too hard about it, I think it's near impossible to say exactly the intention, but I think pitchers tend to fall in 1 of 2 camps, either the target is where they want the pitch to end up (Red) or the target is their aim point, and they let the break take it from there (Yellow). Perhaps you can use the initial target AND the break of a pitch the estimate an area the pitcher COULD be intending the pitch to end up. You don't know if they wanted the pitch to end up in Red or Yellow, but you know it's likely somewhere in the green, so perhaps distance outside of green.

@RealCamBrown @8_Maxx @MLBONFOX Nice idea. What if there's a pitcher who microadjusts his target?

@tomdoyo @8_Maxx @MLBONFOX Use aggregate data to infer intent. Look for clusters for a given pitch (maybe a guy throws his curveball for a strike and also to try and elicit chases, in separate situations) and treat some radius around the center of the cluster as accurate execution.

@CoachJackCheney @MLBONFOX 💯

@tomdoyo @MLBONFOX Just wanted to come up with something that gave better feedback to our pitchers. I know it’s not perfect, but for the high school level, and without having to spend over $10k on tech, this works for my guys.

@BaseballPoli @MLBONFOX Appreciate the insight!

@tomdoyo @MLBONFOX Anyway, regardless, this is a really cool project and really excellent work! Can’t wait to see where this project goes longterm.

@CoachJackCheney @MLBONFOX Wait this is actually amazing. You're ahead of the time

Like I said, poor man’s concept haha. Just me and my players in a high school gym. We set up a camera behind the pitcher’s release point like the overlay you see below. Tracking data, we go old school with a pitch tracking sheet. We don’t have a Rapsodo, so everything is done through video playback and charting. From there, we input the data into our overall bullpen tracking sheet, and it spits out our data from a coach’s viewpoint.

@BaseballPoli @MLBONFOX Where the pitcher is looking? 👀

@tomdoyo @MLBONFOX What we need is absurdly high def imaging of exactly where the pitcher is looking. And we need to catalog the average movement on each pitcher’s change, curve, splitter, cutter, etc — using only data from when they clearly have that pitcher working well.

@SamulskiNYC @MLBONFOX I'm considering looking into individual pitcher tendencies

@tomdoyo @MLBONFOX This is cool stuff but how do you factor in when the catcher shows a target as a visual reference but not the exact spot he wants the pitch to wind up?

@CoachJackCheney @MLBONFOX That's awesome! If there was a video on it it would be interesting to watch

This is sick. I actually developed a poor man’s version of this a few years ago just by working with percentages. Using video feedback during our pitchers bullpens, I’m able to go back and identify the pitch call and pitch location. From there, I calculate what I call Zone Percentage. Zone Percentage is based on the number of pitches that hit the intended target area, giving us a measurable way to evaluate command. Since we don’t have live hitter data during bullpens, I also developed a Quality Pitch Percentage. This metric estimates whether a pitch would likely produce a swing or weak contact based on its location, even if it’s just off the plate or out of the intended quadrant we’re trying to attack. After every bullpen, my pitchers receive three metrics: Quality Pitch % (Elite: 75%) – Measures the probability that a pitch would achieve the desired outcome against a hitter based on location. Strike % (Elite: 65%) – Measures overall control by tracking how often pitches finish in the strike zone. Zone % (Elite: 55%) – Measures command by tracking how often the pitcher hits the exact location that was called. Quality Pitch % bridges the gap between simply throwing strikes and executing pitches that actually win at-bats. Strike % measures control. Zone % measures command. This is way cooler then anything I’ve attempted to come up with.