The Adjustable Swing: Hitters Have a Dial. It Isn’t Wired to Anything.

In 2025, Statcast started measuring something it never could before: not just how fast the bat moves, but the shape of its path — the attack angle, the up-or-down tilt of the barrel at contact. The swing-instruction world pounced. “Get your attack angle into the ideal window.” “Match your swing plane to the pitch.” A new leaderboard, a new product. So we asked the question those products are quietly built on: do hitters actually adjust the shape of their swing to the pitch — and if they do, does it make them better?

The first half is true, and it’s genuinely cool. Hitters do have a dial: across the league, a swing’s attack angle falls about 10° as the pitch climbs from the bottom of the zone to the top — steep and uppercut on the low ball, flat and level up high. And how much a given hitter turns that dial is one of the most stable things about him. We ran the whole study the way we ran the Arm-Angle Gambit and the Two-Strike Brake: two analytical agents with opposite instincts — one interpretability-first, one machine-learning — each working ~476,000 tracked swings blind to the other, then forced to referee each other’s work. They agreed on which hitters have the biggest dial to a correlation of r = 0.99.

The second half is where the brochure falls apart.

One honesty note up front, because it governs everything below: this is all associational, measured on the swings hitters chose to take. We can show you that the dial doesn’t travel with better outcomes. We can’t prove that surgically changing a hitter’s dial would do nothing — only that, across the league, the hitters who have the dial aren’t cashing it in. Here’s the case.

1. The dial is real — and beautifully consistent

Start with the part that holds up. For every tracked swing we have the attack angle and the height of the pitch; line them up within each hitter and a clean pattern falls out. On a pitch at the knees, the average swing tilts up around 16°; on a pitch at the letters, it flattens toward 6°. That’s the dial — an adjustment of roughly 10° across the vertical zone, and it bends at the ends, because there’s only so flat or so steep a human swing goes.

Every hitter sits somewhere on a spectrum of how hard he turns the dial. Jake Cronenworth swings almost 20° steeper on the low pitch than the high one. Juan Soto barely moves — his line is nearly flat, the same swing shape from the shins to the letters. Both are real, repeatable signatures.

And “repeatable” is not a throwaway word here. Split a hitter’s 2025 swings in half at random and the two halves agree on his dial at a Spearman-Brown reliability of 0.94. Better still, his 2025 dial predicts his 2026 dial at r = 0.88 — a full season apart. For comparison, plenty of stats that hitters get praised or benched over don’t clear 0.5 on the same test. If reliability alone made something valuable, the dial would be a headline skill. It is exactly that reliable. It is also, as far as we can measure, worth nothing.

2. The payoff that isn’t: power

The first thing a swing-shape evangelist will tell you is that matching your plane lets you square the ball up — more flush contact, more damage. So we built the per-hitter model the honest way, adding one variable at a time and watching what each one buys, with the outcome being expected production on contact (xwOBAcon, the quality of a hitter’s batted balls stripped of luck and defense).

Each dot is a hitter: how adjustable his dial is (left to right) against the quality of his contact (bottom to top), shaded by bat speed. There is a faint upward drift (r = 0.21) — but the color gives it away. The dots that drift up are the darker, faster-bat ones; the dial just keeps company with bat speed. Soto sits top-left: almost no dial, the best contact in the pool. The next two sections show what happens when you account for the company it keeps.

In numbers: a hitter’s bat speed alone explains 47% of the variation in contact quality across the league. Layer in his average attack angle and his “ideal” rate and you climb to 57%. Now add the dial — how much he adjusts — and the model improves by 0.2 percentage points. Its standardized effect is +0.003 with a confidence interval that runs from negative to positive: statistically, you cannot distinguish it from doing nothing. Both agents, one interpretability-first and one a gradient-boosted machine that was free to find any nonlinear interaction it liked, landed in the same place. The dial does not buy power.

3. The payoff that isn’t: contact

Here’s the steelman, and it’s a good one. Maybe the dial was never about quality of contact — maybe it’s about making contact at all. The whole point of matching your swing plane to a pitch at the top or bottom of the zone is to not miss it. And whiffs really are concentrated at the edges: hitters miss on roughly 79% of competitive swings down at the shins and 47% up at the letters, versus just 14% on a pitch in the heart. If the dial earns its keep anywhere, it’s there.

So Round 2 of the study locked onto whiffs, and specifically on the extremes and on two-strike counts where contact matters most. The answer was not “no effect.” It was slightly worse than that.

More-adjustable hitters whiff more, not less — and the gap is widest in the bottom third of the zone, exactly where a matched swing plane is supposed to rescue contact. The effect is small (a couple percentage points of whiff rate per standard deviation of dial), but both methods agree on its sign and size, and it points the wrong way for the swing-plane story.

This is the cleanest convergence in the study. Claude’s mixed-effects model and Codex’s gradient-boosted classifier — built on different filters, different estimators, validated by hitter so no one can memorize a name — put the extreme-zone whiff penalty at +0.037 and +0.036 respectively. The same number, twice, from two machines that share no code. The most adjustable hitters tend to be aggressive, faster-bat, lower-contact swingers; the dial rides along with a whiff-prone profile rather than curing one. (Two-strike counts: somewhere between no effect and mildly negative — never the contact insurance the theory predicts.)

4. “Ideal attack angle” is mostly bat speed

Which brings us to the metric actually being sold. “Ideal attack angle rate” — the share of a hitter’s swings that land in the 5–20° window — is now printed on leaderboards and pitched as a swing target. And on its face it correlates with production. The question is whether it’s telling you anything that a radar gun pointed at the bat wouldn’t.

On its own, ideal-AA rate has a small positive association with contact quality (left). Account for bat speed — just bat speed — and about 88% of that association evaporates (right), leaving a sliver you can’t distinguish from zero. Both agents, to the third decimal.

To be careful about what this does and doesn’t say: ideal-AA rate is not random noise, and we’re not claiming a hitter should ignore his swing. We’re claiming something narrower and, for anyone selling the metric, more damning — that once you know how hard a hitter swings, his ideal-AA rate carries no independent signal about how well he hits. It is, to a first approximation, bat speed in different clothing.

5. Why we trust this: two methods, made to fight

A single model that found “the dial does nothing” would be easy to wave off as a modeling choice. So we didn’t run one model; we ran two, with opposite biases, and made them cross-examine each other’s work across two rounds. The cross-examination earned its keep immediately.

In Round 1 the two agents appeared to disagree on one secondary test — whether the most-adjustable hitters out-hit the least on raw contact quality. One said yes, one said no. Rather than split the difference, the review traced the gap to its cause: the machine-learning agent had accidentally let ~31,000 swing-and-miss results (worth zero) leak into an outcome that was supposed to be balls-in-play only. Rebuilt clean, on the agreed definition, both agents produced the same positive-but-trivial raw gap — a gap that then vanishes the moment you account for bat speed. The disagreement was a bug, not a finding, and a single-pipeline study would have shipped it.

When two methods that share no assumptions and no code agree five times out of five — including on the size of an effect to the third decimal — the finding isn’t hiding in a modeling choice. It’s in the data.

6. The leaderboard, and the Soto problem

Here are the hitters who turn the dial the hardest, and the ones who barely turn it at all. Read the right-hand columns: the most-adjustable hitters are good, average, and bad in roughly equal measure, and so are the least. The dial sorts hitters cleanly — it just doesn’t sort them by how well they hit.

Juan Soto is the cleanest illustration in the dataset. He has almost no dial — very nearly the same swing shape whether the pitch is at his shins or his shoulders — and the highest “ideal attack angle” rate in the qualified pool, and the best contact quality in it. He doesn’t adjust the shape of his swing to the pitch. He just has a great one, and trusts it. (Fair caveat: he’s the least-adjustable qualified hitter in 2025 and fourth-least in 2026 — the title is “among tracked, qualified hitters,” not a law of physics. The point survives the caveat.)

What this means for tonight’s game

When you hear that a hitter has “fixed his swing” or “gotten his attack angle into the ideal range,” the data says: be skeptical that it’s the thing helping him. The ability to bend your swing to the pitch is real, measurable, and remarkably consistent — it’s just not what separates good hitters from bad ones. That’s bat speed, and the quality of the one swing they own. The most adjustable swing in the league belongs to a solid regular; the least belongs to Juan Soto.

None of this closes the book on swing shape. We measured the vertical dial; the horizontal one — staying through the ball versus pulling off it — is a separate question we’ll take up next. And everything here is measured on the swings hitters chose to take, so we’re describing what the dial travels with, not what changing it would cause. But the specific, sellable promise — turn your dial, find the ideal window, hit better — doesn’t survive contact with two independent looks at half a million swings.