"Clean" and "Super Clean" Starts

What my happy addiction to these notes has taught me is that I am a person who loves exploration. Exploration should be protected and a place for it carved out in society, but one of the few risks is that one might take his explorations too seriously and forgets that they are exactly that. Some do transcend, but you don’t want to habitually confuse an exploration for “a thing.” (One place where this has happened with me, I’m afraid to say, is with my “line drive of the year.” I created it, ergo it must be good.) So, keeping this injunction in mind, here’s the thing I am throwing together today.

They are not the only perpetrators, but I have come to suspect there is an essence to a 2025/2026 Jack Flaherty or Walker Buehler start (who, ironically, were both members of the World Championship 2024 Dodgers). Going back a couple of weeks, my feeling was that, with walks often the culprit, their line scores never seemed to be clean. But when I checked their log of games, I found that inevitably they had actually gotten passable results in one or more categories. It wasn’t really that they had stunk game after game that distinguished them, or even that they had walked batters game after game. It was more that one game it would be walks that offended, and the next strikeouts. It was more like they never had “clean” line scores.

The search to operationalize this particular brand of pitching fecklessness was just part of a greater void I felt when I relied on James’s game scores to do the number crunching of starts for me, as wonderful and useful an invention as I think they were. James’s game score tends to capture the bottom line, which is runs allowed. Dating itself, it also assume a high innings total (from a modern perspective) and assesses a harsh penalty if this isn’t met. So they still give a very good idea of effectiveness in an outing, if with the disproportionate weighting of length of start. But the scores don’t work as well if the goal is to identify truly admiral outings and separate them from the run-of-the-mill ones, where a guy might have been very successful but really just have gotten by. Performance is better identified by taking the measure of the whole line.

So what I had in my head to do (and I know this is stealing from “quality starts” in its approach of comparing each category to a standard¹) was to ask if the pitcher had been above average in all categories. Then, I thought we could ratchet it up a notch and ask if he had been not merely above average but good in all categories.

I designed it so that my definitions were technically accurate and faithful to the MLB averages for starters. First, it’s a bit complicated by openers, but in my recent work on Shota Imanaga, I found that, despite his rocky recent starts, he’d been going longer than the average start, which I found to be just over 5 innings. So, allowing for openers, using 6+ innings seemed a realistic embodiment of above average.

For starting pitchers, other categories in 2026 per inning come to the follow rates.

Hits 0.919

BB 0.361

SO 0.927

Runs 0.491

Earned Runs 0.459

“Stathead” is generally designed to filter based on ratios and not differentials, which isn’t ideal for a yes/no screening that you want people to be able to do in their heads. Given the ~1 average for something like hits, I was more looking for fewer hits than innings than applying some rigorous ratio. As it turns out, when we’re in the narrow range of starts that are 6 to 8 innings, and very occasionally 9, ratio meshes with differential, anyway. To wit, 5 hits over 6 innings isn’t better ratio-wise than 7 over 9. So because of Stathead, I used ratios, but you don’t have to. The results come out the same. The numbers above told me I wanted as many strikeouts as innings and less hits than innings, and I was able to set the ratios to achieve this.

There are so few 9-inning complete games that I wasn’t too worried about the 9-inning, 3-walk case, even though these pitchers have technically walked fewer than average over their start. Otherwise, among 6+ inning starts, 2 walks serves as the dividing point between good- and bad-walk starts, and was simplest just to use.

The runs element did bring more nuance. Run totals are generally the same as earned runs totals in individual games. But if you follow statistics and therefore have the same standard for them, if a pitcher is going to get tripped up on one of them and not the other, it’s going to be on runs. Earned runs can never exceed runs. So I dispensed with earned runs as a criterion.

With where the run rate conveniently is now, very close to 4.5 per game, it’s easy to check without resorting to a calculator which side of 0.5 runs-per-inning a start falls on. And here, an “innings minus runs” framework would obviously mean we had materially drifted from the intended focus on just runs. So, Stathead’s requirement of ratio was just what we wanted here. Using 0.5 as the chopping block, conforming starts are therefore those with 6 innings and 2 runs, 7 or 8 innings and 3 runs, or 9 innings and 4 runs. Three runs is also o.k. if you pitch 6 and change, and 4 is o.k. if you pitch 8 and change.

Although they’ve acquired such significance among pitchers in contemporary analysis, and rather now form a non-fringe category, until I was a good ways into this, I didn’t consider whether I should use home runs, to tell you the truth. From the perspective that they don’t enter into the James scores, either, I don’t mind that I left them out.

So, putting everything in one place, my definition for this thing,² for the “clean” start was

IP GE 6

H LT 1-per-inning

BB LE 2 Total

SO GE 1-per-inning

R LT³ 0.5-per-inning

There have been exactly 200 such starts this year that met this criteria, divided among 97 pitchers. There are two pitchers who’ve done the deed 6 times, four who’ve done it 5 times, nine who’ve done it 4 times, thirteen who’ve done it 3 times, and twenty-four who’ve done it twice. Looking at that account, in truth, the threshold for cutting off my reporting and recognizing excellence should probably be 5+ starts, where I could confine myself to six names. But I started working with 4+ conforming starts as my quasi-leaderboard, and listing those 4+ guys here does help when we look at previous years, and see who is more or less continuing where he left off.

The “Clean Starts” leaderboard for 2026:

(1) Jacob Misiorowski 6

(1) Cam Schlittler 6

(3) Davis Martin 5

(3) Chris Sale 5

(3) Cristopher Sanchez 5

(3) Bryan Woo 5

(7) Chase Burns 4

(7) Nathan Eovaldi 4

(7) Tyler Glasnow 4

(7) Emerson Hancock 4

(7) Nolan McLean 4

(7) Shohei Ohtani 4

(7) Joe Ryan 4

(7) Tarik Skubal 4

(7) Yoshinobu Yamamoto 4

It certainly is not just a straight redux of E.R.A. The E.R.A.s of Eovaldi and McLean hover around 4.00, but their count suggests they’ve pitched really well in particular games. Kyle Harrison and Jose Soriano are perhaps surprises for just having 2; Spencer Arrighetti, Dylan Cease, Max Fried and Parker Messick, perhaps surprises for just having 1.

Some of the low counts can just reflect pitchers who typically don’t make it through the 6th inning. I’m sure that requirement got Freddy Peralta a bunch in 2025. Those cases are less interesting than underperformers where the issue is something else.

My original conception of this measure, that it would diagnose Flaherty and Buehler⁴ and find others like them, was misbegotten and isn’t really worth pursuing further. It turns out it’s damned hard to hit all of these markers. The rate comes out to 1 “clean start” per 8.87 in the major leagues. It moves the goal post for a quality start some ways. In contrast, the idea of a Flaherty or a Buehler is that you know (or think you know) going in that they are not going to get the job done, that their outing will leave something to be desired, if not by the bottom line, than at least aesthetically. My conceit is that line scores can pick this up, and I still suspect they can. But it doesn’t make sense that you find bad pitchers by asking if they’ve ever been great. It’s better to measure what you’re trying to measure directly. The question is if in the wake of this failed idea something else remains.

The leaders in “clean starts” for 2025:

(1) Yoshinobu Yamamoto 14

(2) Zack Wheeler 13 (from 24 starts)

(3) Paul Skenes 12

(3) Tarik Skubal 12

(5) Jacob deGrom 11 (I thought he was babied in terms of workload in 2025. Maybe that’s right, maybe it isn’t, but he rates in 2025, while he has only two “clean starts” this year.)

(6) Garrett Crochet 10

(6) MacKenzie Gore 10 (just 1 this year)

(6) Bryan Woo 10 (he is the man!)

Now, 2024:

(1) Tarik Skubal 14

(2) Chris Sale 13

(3) Bryce Miller 11

(3) Zack Wheeler 11

(Also note Dylan Cease with 9, belying the 2026 idea that he is necessarily weaker in “clean starts” than in his general performance.)⁵

This seemingly is not a statistic that accentuates Skubal’s strengths, but it is only by the relative standard of some other statistics that he could be said to be disappointing in it. He had four conforming starts from seven this year, so with “clean starts,” is about the best guy going, the same as he is in everything else. You could also argue for Chris Sale as the best. With Misiorowski and Schlittler, it’s an evolving assessment as hot as they are.⁶

As far as what possessed me to define and track “super clean”⁷ starts as well, I’m not sure (hey, this is exploratory). But before I knew that insisting on hitting all targets just at an above-average level would mean eliminating 8 starts of every 9, I probably was cognizant of the criticism that “quality start” peddlers have generally come in for, which is that that are championing a statistic that rewards mediocrity and excellence alike. In any event, I decided to find out how the picture would change if I ramped up the standards for all categories a level.

In my rendering this meant taking innings from 6 to 7, hits from innings - 1 to innings - 2, and totals walks from 2 to 1. Instead of equaling innings, strikeouts now have to exceed them by 1.⁸

I switched the run rate from less-than-or-equal-to .49 an inning to less-than-or-equal-to .334 an inning. The practical effect of which was that 3-run outings only qualify if the guy pitches 9 innings. With the mere “clean” standard, the pitcher could allow 3 if he pitched 7 or 8 innings. Changes in permissible runs for starts of fewer than 7 innings don’t matter and are only theoretical, because those starts are disqualified on account of innings regardless, no matter how many runs the pitcher does or doesn’t allow.

Adopting these higher standards and ranking shows how brilliant Gavin Williams can be. All of his 2026 “clean” starts have been “super clean” as well, it turns out (plus, I guess what is at play is that GW is just a horse who can go 7). Misiorowski ties Williams in 2026 with 3 “super clean.” Reid Detmers is the same idea as Williams —just 2 “clean,” but both “super clean.” The only other pitcher with 2 “super clean” in 2026 is Cristopher Sanchez.

In all, there have been exactly 40 “super clean” this year, meaning exactly 1 of every 5 “clean” starts has been “super clean.” And overall, that comes to just 2.3% of starts being “super clean.”

The “super clean” leaderboard for 2024 and 2025 put together (i.e., the most conforming starts combined over those years):

(1) Chris Sale 9

(2) Zack Wheeler 8

(3) Garrett Crochet 7

(3) Sonny Gray 7

(3) Tarik Skubal 7

No one else had more than 5.

Contrary to what I might have expected, Skubal doesn’t gain muscles with the increased standard. Again, I suspect the issue is the length requirement, which at 7 innings, certainly is exacting. But permitting only 1 walk also is tough, giving little margin for error. So the whole thing is a haul.

If you don’t adjust it for the era norm, this is not a good measure if you really go back in time. Except for, I don’t know, a half-dozen pitchers or so, pitchers of old didn’t strike out more guys than innings reliably, no matter how good they were. And in compensation, I’m not sure any category was that much easier in the old days than it is today except for the 7 innings part. Runs allowed was easier without the designated hitter, all other things being equal, and was certainly easier for the small slice of baseball that was Deadball.

There are worse ways to compare pitchers of yesteryear, you could use this, but the results would be hard to interpret, and you would be producing a list that catered to strikeouts more than getting a general indication of greatness. So the idea of doing something more comprehensive historically didn’t motivate me, and I couldn’t in good conscience foist the results upon you.

But going back a few years, as long as strikeouts were comparable to today, I’m o.k. with taking inventory. So, in the true spirit of exploration and half-assedom, I went back to 2013, arbitrarily, and took a couple of cracks at it. I guess really only caring about evaluating the true greats, and thinking using the higher standard made for a better way of doing that, I looked just at “super clean,” not “clean.”

The most “super clean” starts in a season, from 2013 onward:

(1) Corey Kluber, 2017, 9

(1) Max Scherzer, 2018, 9

(1) Jacob deGrom, 2019, 9

(4) Max Scherzer, 2016, 8

(4) Chris Sale, 2017, 8

(6) Felix Hernandez, 2014, 7

(6) Clayton Kershaw, 2016, 7

(6) Justin Verlander, 2018, 7

(6) Walker Buehler, 2019, 7

(6) Max Scherzer, 2021, 7

(6) Chris Sale, 2024, 7

Those with 20 or more “super clean” since 2013:

(1) Max Scherzer 50

(2) Clayton Kershaw 43

(2) Chris Sale 43

(4) Jacob deGrom 32

(5) Justin Verlander 27

(6) Gerrit Cole 26

(7) Zack Wheeler 25

(8) Corey Kluber 23

(8) Aaron Nola 23

(10) Madison Bumgarner 22

(11) Yu Darvish 20

Thinking in terms of both performance in absolute terms and performance versus other statistics, I nominate Scherzer and Sale as the standouts. DeGrom, too, I suppose.

My theory is that it’s a strikeout-to-walk thing, that “super clean” traces to that heavily. That would also seem to explain Kluber and Nola making such good showings.

1. It really just expands the quality start to hits, walks and strikeouts, abandoning the earned runs and innings-only framework.

2. Or exploration!

3. Where GE = “Greater Than Or Equal To, LT = “Less Than,” LE = “Less Than Or Equal To.”

4. I was right at least on the narrow point: Flaherty and Buehler appear as 0 for 23 in clean starts combined.

5. To clarify, Cease was one of seven pitchers who had 9 clean starts in 2024. No one had 10.

6. But given the half-assed nature of this, I should not worry about putting the cart before the horse!

7. Agonizing over the nomenclature when I started the piece by saying this had better not become “a thing” would be a mistake, but a “clean”/”pristine” coinage is catchier than “clean” and “super clean”, for sure. “Clean” and “pristine” outside baseball are not mutually exclusive, but then again, my statistical categories of them are not, either.

8. In line with Stathead, this actually meant a 1.10 ratio of strikeouts to innings in the start. I forgot to think about the issue of partial innings, but in practice, 1.10 excludes the 8 SO, 7.1 IP case, which would be the highest ratio of any game with at least 7 innings and less than a whole number between strikeouts and innings.