<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Alex Devorsetz on Medium]]></title>
        <description><![CDATA[Stories by Alex Devorsetz on Medium]]></description>
        <link>https://medium.com/@devorsetza?source=rss-f05a276ef161------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*13eQSt91e0IEorin</url>
            <title>Stories by Alex Devorsetz on Medium</title>
            <link>https://medium.com/@devorsetza?source=rss-f05a276ef161------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 01:20:02 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@devorsetza/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com" rel="hub"/>
        <item>
            <title><![CDATA[The Art (and math) of Predicting Pitches]]></title>
            <link>https://medium.com/@devorsetza/the-art-and-math-of-predicting-pitches-8b27053e80fd?source=rss-f05a276ef161------2</link>
            <guid isPermaLink="false">https://medium.com/p/8b27053e80fd</guid>
            <category><![CDATA[mlb]]></category>
            <category><![CDATA[baseball]]></category>
            <category><![CDATA[data-analytics]]></category>
            <category><![CDATA[predictions]]></category>
            <category><![CDATA[pitching]]></category>
            <dc:creator><![CDATA[Alex Devorsetz]]></dc:creator>
            <pubDate>Thu, 12 Nov 2026 00:37:56 GMT</pubDate>
            <atom:updated>2026-11-12T00:42:26.774Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*1f48izC3ajf-xV5G.jpg" /><figcaption>Denis Poroy - Imagn Images</figcaption></figure><p>May 21st, 2026.</p><p>It’s the bottom of the fourth inning in a scoreless game between the Diamondbacks and Rockies. Arizona’s starting pitcher is Eduardo Rodríguez, at the start of an incredible bounce back season. He’ll end 2026 with a career-best 2.89 ERA, the first All-Star Game selection of his career, and a couple down-ballot Cy Young votes.</p><p>Eduardo Rodriguez will throw seven scoreless innings today. Willi Castro could’ve changed that in the bottom of the fourth inning.</p><p>E-Rod started the inning with an outside changeup, a fairly uncompetitive pitch that Castro spits at. The second pitch is another changeup in nearly the same spot, and Castro takes again. Third is a low fastball, and Castro barely flinches.</p><p>We all <em>“know”</em> what Eduardo Rodriguez is going to throw next, or at least we feel like we do. It’s a 3–0 count against a soft-hitting righty by a pitcher with good command that primarily uses a four-seam fastball. It’s easy enough for us to say that, but not good enough for Willi Castro. From the start of 2025 to May 21st, 2026, Rodriguez entered 24 3–0 counts against righties. He threw 18 four-seamers, five cutters, and one changeup. Only 13 of those four-seamers were in the strike zone. Those aren’t strong enough odds for a leadoff man to swing on a 3–0 count.</p><p>So Willi Castro took the pitch. It was a four-seamer almost perfectly down the middle, maybe a little outside. It’s easy to look at that take and say “why didn’t he swing”, but hindsight is 20/20. Just a month prior, Rodriguez made Everson Pereira look silly by throwing a changeup on a 3–0 count. As much as we feel like we <em>“knew”</em>, we didn’t know.</p><p>But I almost knew. That pitch had a 94% chance of being a four-seam fastball, and when it was, it was likely going to be down the middle, maybe a little outside.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*z70weDdBJCQYO1XwyqkV3g.png" /></figure><p>This plot was created by training a machine-learning model on every pitch thrown from 2024 to the 2025 All-Star Break, tuned on the rest of 2025, and tested on a 200,000-pitch sample from 2026. The pitcher’s repertoire, the hitter’s strengths and weaknesses, the pitches previously thrown this at-bat/game/season, the game state, the catcher’s tendencies, and the weather are all taken as inputs, and a model predicts the likelihood of each pitch type being thrown. Each pitch type then gets its own location-probability chart created, with 4-inch cells denoting the likelihood that the pitch ends up there.</p><p>A circle with a 6-inch radius is then placed on the type/location plot that will provide the highest likelihood of occurring. Circle 2 is then selected in the spot that will add the most probability, with a 6-inch overlap allowed with Circle 1. Circle 3 is the same, with a limited overlap allowed for the first two circles. Many plots look like Rodriguez’s with all three circles on one pitch type, but most predictions have circles on multiple pitch types.</p><p>If the ensuing pitch ends up in any circle, then we’ve successfully predicted it. In order to accurately evaluate the model, I removed any pitches with a swing probability below 1%. I then used my model to place the three pitch type/location circles, and tested its accuracy based on every 2026 pitch.</p><p>My model predicted the correct pitch type 46.6% of the time. About 25% of pitches end up in one of the three circles, with 11% ending up in the primary circle. Using this method, we can aggregate our 2026 predictions in order to find the predictability of every pitcher. Here are the top-10 and bottom-10 from 2026:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*L8QHeQ--nYP-rCCEQHlZjw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*--3KZ8tGxI9ihp0UsMtKIw.png" /></figure><p>As one could guess, pitchers with two pitches, one of which they are overly-reliant on, are the easiest to predict. The overall leader is Yennier Cano, whose pitches were accurately predicted an absurd 43.5% of the time. On the other hand, the least-predictable pitchers were the ones with a larger repertoire. Caleb Thielbar leads this category, with only 15.0% of pitches falling into one of the three circles.</p><p>This raw predictability is heavily dependent on repertoire size. However, if we isolate for this, we can evaluate each pitcher on how predictable they are relative to the average pitcher with their repertoire size. This allows us to measure each pitcher’s predictability based solely on pitch selection. Here are the top-10 and bottom-10 in repertoire-adjusted predictability:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9TsQv26YeM4KUSUf5DX6tg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZH9tvtU4bDQHHunWBWDg5A.png" /></figure><p>Many of the same names appear in the top-10 for predictability, with the rest of the list being filled out by pitchers that wildly change their pitch mix for handedness. For example, Jacob deGrom throws his changeup and curveball about 30% of the time against lefties, but his mix against righties is 40% four-seamers, 47% sliders, and 13% for the rest. He’s the most predictable pitcher against right-handed hitters as a “five-pitch pitcher”, which drags him down.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*B0yw8u4biKb9JbCBpkCKlw.png" /></figure><p>The pitcher with the least-predictable pitch mix is Riley Cornelio, a right-handed reliever for the Nationals with a small but deceptive pitch mix. Cornelio throws his slider, four-seamer, and cutter with a 50/35/15 mix against RHBs and a 45/40/15 mix against lefties. It seems to be a common trend for the least predictable pitchers to be guys that throw their primary pitch approximately 50% of the time regardless of batter handedness, as that describes Caleb Kilian, Fernando Cruz, Clayton Beeter, and Pete Fairbanks.</p><p>We can then expand this process even further, finding each team’s net pitch predictability. This produces a fairly neat staircase of teams ranked from predictable to unpredictable, with one huge outlier.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wF7uLlFiWEtBrO-FJxNFIA.png" /></figure><p>The Miami Marlins have the least-predictable pitching in MLB, with the Kansas City Royals as the only team within their stratosphere. A lot of this status is due to their organizational wealth of high-repertoire pitchers, and their overall pitch diversity.</p><p>Of their 20 pitchers to surpass 300 pitches in 2026, only six had a pitch that they threw over 40% of the time. The only team with fewer high-usage pitchers was the aforementioned Royals squad, who only saw a 40%-usage pitch on five of their 20 qualified pitchers. For reference, the Los Angeles Dodgers also had 20 pitchers surpass 300 pitches in 2026. 15 of them had a pitch they threw over 40% of the time, and 2 of them had multiple.</p><p>However, this doesn’t tell the full story. If we adjust for repertoire size, the Marlins still have a comfortable lead in unpredictability.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mDr2fvHVUatzIWDfxA08uA.png" /></figure><p>The way the Marlins call pitches is uniquely more difficult to decipher than any other team, and I don’t have a concrete answer as to why. I have hypotheses, such as <a href="https://www.mlb.com/news/marlins-and-rockies-calling-pitches-from-dugout">their decision to call pitches from the dugout</a> allowing them to make better pitch decisions than a catcher would be able to derive from memory alone, but we can’t know for certain unless we can find which teams are calling pitches from the dugout and which teams aren’t.</p><p>In theory, pitches that are easier to predict should produce worse results for the pitcher. In practice, that isn’t the case.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hMXPznTn-C4uh-0EiORZ3Q.png" /></figure><p>When facing pitches that should be more predictable, hitters whiff more and produce worse contact, with no major change in swing rate. This isn’t some side effect of small-repertoire pitchers with one uber-dominant yet predictable pitch. <a href="https://medium.com/@devorsetza/other-pitch-prediction-graphics-that-i-found-interesting-that-i-couldnt-fit-in-the-main-article-a8c25a72a4c7">Pitch predictability changes depending on count and location</a>, but both are adjusted for in this check. What’s even more conclusive is that zero MLB teams see an improvement on contact quality of predictable pitches with 90% confidence. Nobody has cracked the code yet.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oAHbpHn7TZv__XiFV4GIsQ.png" /></figure><p>I see this as a major inefficiency with both pitching strategy and hitting strategy. The average game in 2026 had about 300 pitches thrown between both teams. Of those 300, my model would correctly predict 140 pitch types, 74 pitches within one of three pitch type/location circles, and 32 pitches within the primary circle. The fact that those numbers are so high is shocking to me, and the fact that hitters do so poorly on them is even more surprising.</p><p>I believe that this is an untapped area of optimization. When we plot each pitcher’s predictability against their average pitch quality, we see a very weak correlation that matches the hitting plot, with more predictable pitchers counterintuitively having marginally better production. Despite the fact that we can predict pitch type and location a quarter of the time, it does not impact results.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Sh0UXP1FZMXX16dSB8PLwg.png" /></figure><p>My hypothesis is that the use of this information will not cause the unpredictable pitchers to see a jump in production. I’d expect the inverse, that the more predictable pitchers will see higher swing rates, lower whiff rates, and better contact quality when facing batters that focus on pitch-by-pitch predictions. Pitchers like Yennier Cano, Tyler Rogers, and Kenley Jansen thrive on their novelty, hoping their small batch of unique pitches will confound hitters. The ability to predict which one they will throw and where should give the hitter a strong advantage, but it seems none are using it to the fullest extent yet.</p><p>Whichever team capitalizes on this first will surely have a strategic leg-up on their opponents. The Marlins made the first move by building baseball’s most unpredictable pitching staff, but there are still 29 MLB teams whose pitch selection is too easy to anticipate, and 30 hitting staffs who simply cannot take advantage of it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8b27053e80fd" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>