Ben Clemens

Pardon the mobile screenshot, but look, his swing is just perfect for it

Send this to our mailbag column so we get it in a pretty format. I'm down to answer though

I answered a few questions from the comments of my last $/WAR article, and moved replacement level around just for funsies while I was at it

I'm out skiing today, but I want to highlight this research on the cost of a win in free agency. I think it's both intuitive and useful

No time to chat today, too busy buying dog t shirts

The worst part of making small upgrades is learning that things you think are small are actually impossible. Come up with a new color scheme and calculate league average dynamically? Simplicity itself. Flip the axes of a graph without making it look ridiculous? Hope you have some free time.

This was supposed to be an article showing off the FG Lab, but instead I just spent a while analyzing Logan Webb's weird platoon splits

Updates: four improvements to data visualization tools

Huh. Updates. Cool. (www.fangraphs.com/lab/changelog)

That said, with the Lab checked off (for now, we have plenty of new tools in the pipeline) my big project for the year is to dig through the PitchingBot codebase and make new fun/useful stuff, so expect more on this front

Boy, we should probably come up with some kind of PitchingBot Visualizer to investigate this in greater depth 😂

and performed a count-adjusted version of PitchingBot grading so that 'this individual pitch has a 70 grade' would have meaning. But that's ever so slightly different from adding them all up and dividing by overall average, the real method, since that doesn't have count as a variable anywhere.

league throws 15%, so it's off a little, but it basically comes out in the wash. We measure every pitch relative to the all-count league average, add em up, divide. But since the 0-0 baseline is less nasty than the 2-strike baseline, you get weird effects. So we created count-level averages...

Hey you asked for it - for various reasons, the average pitch in each count (the normalizing factor that sets 50 on the scale) is different, and it's particularly different in modeled stuff. Taken across the whole sample, nbd; maybe this one pitcher throws 17% two-strike fastballs and the...

Don't expect this fast of service all the time but Sean and I are trying to get some of the first improvements out fast. Here's my dev-version mockup, should be out before too long:

In two-strike counts in 2025 where he threw a slider, Helsley recorded a strikeout on 18.9% of pitches. In his career before last year, that rate was 31%. Call it what you want - he didn't get strikeouts because he couldn't end at-bats with his slider, and it hurt him

Join the club, because I do that all the time too. I blame the way the human brain works

I think the real risk is like 'Zebby gets hurt, or another guy gets hurt, and now the minor leaguers are throwing 3 innings a start and the bullpen collapses'. It's kind of shocking how many pitchers can get hurt out of nowhere these days

Okay, I gotcha here. Let's say Lopez+ bullpen = 3.86 RA/game and Zebby+bullpen = 4.7 RA/G. Twins score 4.2 R/G. Plug that into Pythag: (RS^1.83)/(RS^1.83+RA^1.83). You'd get a .537 W% for the Lopez version and .448 for Zebby. Across 25 starts, that's the diff between 13.4 wins and 11.2 wins.

It's a long story related to the fact that PitchingBot overall grades are count-agnostic, whereas we have to change that for the individual count grades. We settled on sticking with the unweighted ones, which have the most predictive power - but if you weight by counts, there's a bit of noise

We had a blast talking about whether Jordan Walker can figure it out (probably not! But I hope so!), what we think of the new front office direction, long-term roster strategy, and much more

Oh look, a conveniently scheduled time to ask questions or demand new Lab features

Welcome to my new life, incidentally. Make a chart, find a cool conclusion, make another chart that contradicts that earlier conclusion, investigate.

I think there's maybe something to his splitter being different, and the two-strike locations look pretty good to me even if PitchingBot is mixed (it likes them net of stuff, of course). But yeah, the fastball feels like sample size. Pretty cool either way though.

Man, Davy rocks, read @davyandrewsdavy.bsky.social

Fascinating. One potential complication, though, courtesy of PitchingBot Visualizer. Yesavage doesn't have the biggest sample size on some of these:

FanGraphs Lab Explore experimental baseball tools and visualizations

All the screenshots of scatterplots and swing shapes I've been posting here for the past few weeks? You can now make those for yourselves - and much more. Come and check out the FanGraphs Lab, now live:

www.fangraphs.com/lab

I'm really excited about this new project. Come take a look!

Congrats R.J.! I hope to mistakenly pan many of your decisions, only for them to work out great, in the years to come.

a chart showing Jackson Holliday's squared up rate by launch angle, with a red shadow showing the best angles for home runs. Holliday squares the ball up quite frequently in that shaded area, but the size of the frequency dots show that he very rarely hits the ball there.

When Jackson Holliday hits the ball at ideal home run angles, he squares the ball up a ton. The only problem? He rarely hits the ball at ideal home run angles. Your groundball percentage should NOT be higher than league average with this swing. Something to keep an eye on after his hand heals.

a table showing the daily wRC+ of the top six Phillies regulars in September, when the Phillies went 17-8. Brandon Marsh had the best overall numbers; Harrison Bader did the best in wins.

Boy, you will never guess who Philadelphia's phenomenal 17-8 September was driven by (the strong arm means "highest wRC+ in wins")