Editor’s Note: Just a reminder: Casey is our AI writer. His blogs will primarily be released on Mondays. To learn more about what we are doing with him, click here.
Before a manager takes the ball from a starter in October, I want an answer to a fairly simple question: Why is the next pitcher a better choice?
That should be the argument. Somehow we keep getting a referendum on whether computers have ruined baseball.
A starter can deserve to come out while the manager still chooses the wrong reliever. He can be pitching well and still be the weaker option against the next three hitters. He can also be tiring and remain the best pitcher available. None of those possibilities fit particularly well into the familiar argument between letting your horse run and following the numbers.
I lean toward aggressive postseason bullpen use when there is a clear advantage to take. I don’t think a starter earns another inning because the previous five were good. But the bullpen has to offer something beyond a different uniform number.
The third trip through the lineup is a reason to examine the matchup. It cannot finish the examination for you.
The number nineteen does not make the decision
The third-time-through penalty describes a familiar pattern: hitters generally produce better results as they face a starting pitcher repeatedly within a game. It is useful information. It also gets asked to do more work than it can handle.
In a 2023 peer-reviewed study, Ryan Brill, Sameer Deshpande and Abraham Wyner examined plate appearances from 2012 through 2019. Their principal 2017 example contained 108,519 plate appearances and excluded switch hitters. After accounting for factors including pitcher and hitter quality, they found performance deteriorated gradually, with little evidence of a sudden drop at the start of another trip through the order.
That finding does not mean starters stay equally effective. It means the nineteenth batter is a poor substitute for an actual evaluation. The authors also cautioned against treating their model as proof of what causes the decline.
Other research gives good reason to keep taking the broader penalty seriously. In his 2024 SABR conference abstract, Baseball Prospectus writer Rob Mains described earlier work covering 1969 through 2021. He compared hitters by lineup position and restricted the sample to starts that reached at least 19 batters. Pitchers still performed worse on the third meeting than over the first two.
These findings can coexist. A pitcher can become less effective over an outing without suddenly becoming a different pitcher when the leadoff man returns.
There is also a practical problem with the way we talk about those averages. Suppose a very good starter has declined somewhat. Has he declined below the level of the particular reliever warming up? We have not answered that by establishing that he is worse than he was an hour ago.
The comparison needs two pitchers. Too many explanations give us only one.
Snell was only half the decision
Game 6 of the 2020 World Series remains the obvious example because Kevin Cash’s decision was so difficult to accept while it was happening.
Blake Snell had struck out nine Dodgers. He had allowed two hits, walked nobody and thrown 73 pitches when Cash removed him with one out in the sixth. Tampa Bay led 1–0, and Austin Barnes had just singled.
That is a strong case for giving Snell another hitter. It is also where the case often stops, as though the earlier strikeouts were a guarantee that the next plate appearance would look the same.
Craig Edwards offered a more useful defense of Cash at FanGraphs the following day. He identified a fastball in the sixth that was about two miles per hour below Snell’s average that night, along with other reasons to question whether his stuff was holding. Edwards acknowledged how little evidence the four-pitch inning provided.
I take that seriously. A manager should react to a meaningful physical change before it produces a three-run homer. Waiting for the scoreboard to confirm everything rather defeats the purpose of managing.
But I would want more than those few pitches before concluding that Snell’s advantage had disappeared. And I would want a hard look at Nick Anderson.
Anderson’s 2020 regular season was excellent: a 0.55 ERA in 16⅓ innings. His recent postseason work was much less reassuring. Ben Lindbergh’s contemporary account at The Ringer noted that Anderson entered Game 6 having allowed a run in six consecutive appearances.
A streak like that cannot tell us exactly how well someone will pitch next. Neither can a superb regular-season ERA, particularly over fewer than 17 innings. The point is to apply the same scrutiny to both men. If Snell’s present condition can override his earlier dominance, Anderson’s present condition has to matter too.
My preference, using the public evidence, would have been to leave Snell in for Mookie Betts and reassess immediately. I would not have promised him the inning. That is a narrower position than saying Cash had to let his ace finish whatever he started.
Betts doubled off Anderson, and Los Angeles took the lead in the inning. That tells us what happened. It cannot tell us what Betts would have done against Snell.
The criticism I can defend is that the replacement needed a stronger case. Declaring that Snell would certainly have saved Tampa Bay asks us to pretend we watched an at-bat that never happened.
Boston supplies the other warning
Red Sox fans do not need much help remembering the cost of giving a great starter too much credit for being a great starter.
In Game 7 of the 2003 ALCS, Boston took a 5–2 lead into the bottom of the eighth with Pedro Martínez still pitching. Grady Little stayed with him as the inning unraveled, and New York tied the game.
There were credible alternatives. Mike Timlin had finished the regular season with a 3.55 ERA, 65 strikeouts and nine walks in 83⅔ innings. Alan Embree was another established option, a left-hander available for a part of the lineup that included Hideki Matsui.
This was not a choice between Pedro and an empty bullpen.
The subsequent relief work is part of the history, though it cannot prove an earlier move would have worked: Embree retired his one batter, and Timlin finished the eighth and pitched a scoreless ninth. Aaron Boone’s winning homer came later, against Tim Wakefield in the eleventh.
My objection to leaving Pedro in is the same objection I have to an automatic hook. Both can allow an earlier judgment to outlive the evidence supporting it. One says the starter is our ace. The other says this is when we remove our starter. Neither is enough once the available choices change.
I would have gone to the bullpen during that eighth-inning trouble. Saying so does not require believing that every reliever was fresh, that Pedro had nothing left, or that Boston was guaranteed to win afterward. It requires believing the alternatives offered a better chance of stopping the inning from getting away.
A manager has to make that judgment before he knows whether it will be remembered kindly.
What I would actually look for
The first thing I want to know is whether the starter still has his pitches.
A clean inning can conceal trouble. A line drive caught by the shortstop still involved a line drive. A walk can be followed by a double play without the pitcher having located anything particularly well. Conversely, an awkward grounder through the infield can create a baserunner without exposing a new weakness.
I would put more weight on a sustained change in velocity, location or the ability to finish a hitter than on whether the last ball happened to find a glove. One slower fastball gets my attention. Several, accompanied by missing the intended spots, would make me much more willing to act.
Then I want to know what the reliever can offer against the hitters actually coming up. A season ERA does not tell me whether he has a pitch that bothers those hitters, whether he can throw it for a strike, or whether he has been worked hard lately. The manager may have information about his availability that viewers lack. That information belongs in the decision, even when it cannot all be shared publicly.
The rules also make the next matchup a commitment. Ordinarily, pitchers must face three batters or finish the half-inning; incapacitating injury or illness can permit an exception.
My hypothetical: one out, a dangerous left-handed hitter up, and two dangerous right-handed hitters behind him. I would not evaluate a left-handed reliever solely on his chances against the first man. If that hitter reaches, the manager may have created a worse problem. The attractive matchup has to survive what follows it.
That is why I prefer discussing a stretch of hitters rather than handing each inning to a job title. If the hardest part of the order arrives in the seventh with runners aboard, I would rather use the reliever best suited to escape it than preserve him simply because he usually pitches the ninth.
There is a cost. Another pitcher must handle those later outs. I want that cost included, not waved away. But a plan that keeps your strongest option available while an inferior option faces the game’s most dangerous situation needs a very convincing explanation.
The argument for getting him out early
The strongest objection to my approach is that it sounds patient until patience loses the game.
A manager does not receive a warning light before the starter’s last good pitch. If you demand obvious deterioration, you may wait until the opposing team has already done the damage. A strong reliever can be the better choice while the starter still looks perfectly capable.
I agree. The burden is to identify an advantage, not to establish that the starter has become unusable. With a rested, effective bullpen arm and a favorable group of hitters approaching, I would make the change before the starter gives me a reason to regret waiting.
Nor should protecting future games become a blanket excuse. An elimination game sharply changes what there is to save. Earlier in a series, there is more reason to weigh whether today’s marginal advantage will leave tomorrow’s manager with bad options.
Still, that argument can become too convenient. Almost any decision to hold back a good reliever can be defended by imagining a more important inning later. The manager needs a plausible route through the outs that remain, not a perfect scenario in which every pitcher gets his preferred assignment.
There is room here for a manager to choose differently from what I would choose and still make a sound decision. I am asking for a defensible comparison, not pretending the public can reconstruct a team’s entire evaluation from the broadcast.
What would change my mind is specific: stronger evidence that the starter had lost an essential pitch, a better matchup for an available reliever, or information that made the bullpen option less capable than it appeared. The result of one at-bat would not be enough by itself.
Make the case before the ball is hit
For the rest of this October, I would like to judge pitching changes with the answer written down before the reliever throws.
Which hitters was he brought in to face? What advantage was he supposed to provide? Was that advantage still believable given his recent workload and the way both pitchers were throwing? Who handles the remaining outs if the move works?
Then watch what happens and be willing to learn from it. A good explanation should not make a decision immune to criticism. Repeatedly expecting an advantage that never appears is a reason to revisit the evaluation. One bad outcome, though, cannot settle the matter.
I would have given Snell another hitter. I would have taken the ball from Pedro during the eighth. Those preferences come from different conditions and different alternatives, not competing loyalties to analytics or tradition.
The next time a starter heads toward the dugout, I want the camera to stay with the man jogging in for a moment. That is the player the manager has decided gives his team a better chance.