The Fielder’s Fingerprint: Two Gloves, Same OAA, Opposite Skills

Two of the best defensive seasons in baseball last year belonged to a shortstop and a center fielder you’d never confuse for each other — and the way they were good is the whole story. Masyn Winn saved +21 outs above average at short. Pete Crow-Armstrong saved +24 in center. Nearly identical totals, two of the best gloves alive. But +17 of Winn’s outs came on balls he charged; almost none came going sideways. Crow-Armstrong is the mirror image — +22 of his outs came ranging laterally, and just +3 coming in. Same elite number. Opposite glove.

Outs Above Average is the best fielding stat the public has, and it does one thing: it adds up. It tells you how many outs a fielder saved, never which kind. So we went looking for the kind. We pulled Baseball Savant’s directional OAA — outs earned coming in, going back, and ranging to either side — for every qualified fielder over 2021–25, and we ran it the way we ran The Jump Tax and The Adjustable Swing: two analytical agents with opposite instincts, one interpretability-first and one a machine-learning engine, each working the data blind to the other, then forced to referee each other’s work. They converged on something the single number can’t see.

One honesty note up front, because it governs the whole piece: these are season totals, not play-by-play. We can show you that the direction in a glove is real, repeatable, and independent of speed. We can’t see the positioning card, the shift, or the route on any single ball — so when we call the tilt a “skill,” we mean a stable, measurable trait, not a coaching claim. Here’s the case.

1. One number, two gloves

Start with the cleanest version of the puzzle. Francisco Lindor and Javier Báez played the same position in the same league last year and finished a single out apart in total OAA — +5 and +6. By the scoreboard, the same shortstop. But Lindor’s glove tilts hard toward the ball hit in front of him (a +10 tilt), and Báez’s tilts the other way (-3): Báez does his damage with range and on balls hit deep, and gives a little back coming in. Two five-run shortstops, opposite crafts.

Each row is two fielders at the same position whose total OAA is within one out of each other — the left tag — but whose tilt points the opposite way. Lindor charges; Báez covers deep. Alex Bregman and Nolan Arenado were both +3 at third, pointing opposite directions. Josh Jung (+5) and Ryan McMahon (+6) at third are nearly a full mirror. The OAA column would call each pair a tie.

This is what we mean by a fingerprint. The tilt is simply the outs a fielder earns on balls in front of him minus the outs he earns going back — positive means he’s a charger, negative means he’s a deep-cover defender. It’s a small, legible cut of a number that usually arrives pre-summed, and the moment you take it the league stops looking uniform. The obvious next question is what the tilt is made of. The intuitive answer — it’s just athleticism, fast guys get to everything — turns out to be wrong.

2. It’s a read, not a wheel

If the tilt were just a byproduct of speed, it would be a curiosity, not a skill — you’d already see it on the sprint-speed leaderboard. So we plotted every qualified fielder’s tilt against his sprint speed. There is no relationship at all.

The cloud is flat: tilt and sprint speed correlate at r = 0.04 — statistically nothing. Fast fielders are as likely to be deep-cover defenders as chargers, and slow ones too. The thing that decides which direction a fielder is good in is his first move and his read, not his wheels. For contrast we ran the same test on lateral range, and there speed shows up loud and clear (r = 0.32, the same as total OAA’s correlation with speed) — ranging side to side really is mostly legs. Charging and retreating is something else.

This is why Winn and Crow-Armstrong can both be elite and look nothing alike. Crow-Armstrong is a burner who turns top-end speed into lateral outs; Winn is a shortstop whose value is almost entirely in how decisively he attacks the ball in front of him. Neither is “more athletic” in a way that explains the difference. The tilt is a separate axis of fielding — and a separate axis is only interesting if it lasts.

3. A real skill — but only up the middle

A trait that doesn’t repeat is just noise wearing a name. So we tested whether a fielder’s tilt carries over from one season to the next — and, to be strict about it, whether it carries over after we remove his total OAA, so we’re measuring the direction itself and not just “good fielders stay good.” Overall, it holds: residual tilt repeats year to year at r = 0.28. But the average hides the most honest part of the finding.

Up the middle — shortstop, center, left — the tilt is a genuine skill, repeating at about 0.29 even after stripping out total OAA. At third, second, first, and in right field, it doesn’t carry over at all. The bootstrap intervals are wide (these are season totals on a few hundred fielders), and we’d rather show you that than hide it. The spread we saw in the twin map exists everywhere; the repeatable skill lives up the middle, where reads off the bat matter most and a fielder takes the most chances in both directions.

That split is the difference between “a pattern in one season’s data” and “a thing about the player.” A corner infielder’s tilt in a given year is mostly which balls happened to find him; a shortstop’s tilt is a property of the shortstop. It also told us where we’re allowed to name names — so we did, at the position where it means the most.

4. The shortstop board

Here is 2025’s shortstop tilt leaderboard — the chargers who live on the ball hit in front of them, and the deep-cover shortstops who give a little back coming in to make it up with range.

Masyn Winn is the purest charger in baseball and an elite defender outright — +17 of his outs came in, the highest in the game. Lindor, Dansby Swanson, and Geraldo Perdomo are chargers too. At the other end, Báez, Ezequiel Tovar, and CJ Abrams earn (or lose) their value the other way. Read the OAA column: many of these shortstops sit at nearly the same total while pointing opposite directions — the leaderboard the single number can’t print.

None of this means a charger is better than a deep-cover defender, or vice versa — the whole point is that they bank the same total. It’s a description of how a glove gets there, the kind of thing a hitter’s spray chart or a positioning coach would actually use. Which is exactly where we have to be careful about what we’re claiming.

5. A fingerprint, not a forecast

It would be easy to oversell this, so we’ll undersell it on purpose. The tilt is real, it’s speed-independent, and up the middle it repeats — but it is modest. Year-to-year it carries over at about 0.28, not the 0.5-plus you see for total OAA. And when we asked the hardest question — does knowing a fielder’s tilt help you predict his value next year, on top of what his position and speed already tell you? — the answer was basically no. The extra predictive lift was around 0.03 in R-squared, with a confidence interval that runs through zero. The direction is a stable trait, not a market inefficiency.

We also checked whether “chargers” and “deep-cover” fielders are real, separate types or just two ends of one dial. They’re a dial. When the machine-learning agent tried to find natural clusters, it couldn’t — the fielders fill a continuum, and the cleanest split it found was really just good-glove versus bad-glove, not charger versus retreater. So we use “charger” and “deep-cover” the way you’d use “pull hitter” — a useful label on a spectrum, not a box. The finding isn’t that there are two kinds of fielder. It’s that the single number erases a real, repeatable axis along which gloves differ.

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

A single model that found “a hidden directional skill” would be easy to wave off as a model artifact. So we didn’t run one. An interpretability-first agent (partial correlations, within-position residualization, bootstrap reliability) and a machine-learning agent (gradient-boosted and ridge models with grouped, player-purged cross-validation, permutation importance, clustering) each worked the data independently, then reviewed each other as hostile peers.

The cross-examination earned its keep. The machine-learning agent’s first headline — that direction adds real predictive value — was measured against the wrong baseline; the interpretability agent showed that once you control for position, the edge nearly vanishes, and both agreed to report the smaller, honest number. The interpretability agent, in turn, had to walk back a too-clean claim that one cell “passed” a threshold it only cleared on a coin-flip of method. When two engines that share no code agree on the numbers and on where each other oversold, what’s left is what’s real.

Five questions, two independent machines, five agreements — including the two that keep the finding modest.

What this means for tonight’s game

When the broadcast flashes a fielder’s Outs Above Average, you’re getting a true number with a fact filed off. Two shortstops at +5 can be opposite players: one who lives on the slow roller and the ball hit in front of him, one who earns it ranging deep into the hole. The next time you watch a defender come in and barehand a chopper, or drift back and run one down at the warning track, you’re seeing his fingerprint — the part of his game the leaderboard rounds away.

And it’s a read, not a radar gun. The fastest man on the field isn’t more likely to be the one who attacks the ball in front of him; that’s a separate instinct, it shows up most at the positions where reads matter most, and it’s stable enough to call a skill. OAA tells you how good a glove is. It just won’t tell you which way it points.