The Leadoff — Conventional Wisdom, Tested

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

Lindor and Baez were both ~5-run shortstops in 2025 -- but one charges in and one covers deep. OAA gives a glove a number and hides the direction. And the direction is a real, speed-independent skill (r=0.04) the scoreboard throws away.

Fielding · 9 min read

Read the analysis →

Fig.1 — Called-strike rate by location
2025 baseline · inside → away

The rulebook box never moves. The shadow zone around it does — and it tightens when the batter has two strikes.

By the Numbers

Pitches analyzed

Seasons of Statcast

Umpires graded

Investigations published

The Investigations

All analysis →

Teams

A Bad Start Is Forever. A Good One Isn't.

Every July, fans ask which first-half teams are real. We pooled five full seasons (2021-2025, 150 team-seasons) and tested it out of sample. A team's quality is real (random-split r=0.74), but the first half predicts only about a THIRD of the second -- and almost all of that third is in the basement. A bad first half predicts the second at R2=0.33; a good one at just 0.07; the .500 muddle is noise. 82% of clearly-bad starts stayed buried vs 67% of good ones staying up -- gravity pulls down harder. The signal locks in by ~game 40 (Memorial Day); run differential and the plain record predict it best (fancy stats add nothing); bat speed is noise. Plus: the biggest frauds (2023 Angels 44-37 -> 29-52) and what the rule says about 2026.

June 21, 2026 · 8 min read

Baserunning

The Jump Tax: The Steal Is a Decision, Not a Physique

Josh Naylor is among the slowest runners in baseball and stole 22 bases in 23 tries in 2025. Not with his legs, and not with a big lead -- his resting lead is 3rd-percentile -- but with an ambush: on the pitches he picks, his lead jumps to the 97th percentile, the biggest standing-to-go swing in the game. We decomposed the stolen base across 70 qualified runners with two independent methods over three rounds. The jump and the legs are real, separable, remarkably repeatable skills (the jump persists year over year at 0.79) -- but neither predicts who actually banks steal value. That is conversion, which tracks run value at r=0.87 and barely repeats. Single-season steal value is ~44% small-sample noise. There are two ways to be elite -- pure speed (Buxton) or pure ambush (Soto, Naylor) -- and they bank the same runs.

June 21, 2026 · 11 min read

Bat Tracking

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

Statcast started measuring the shape of a swing in 2025 — the attack angle — and the instruction industry turned it into a product. So we tested the premise: do hitters adjust their swing shape to the pitch, and does it help? The first half is true and genuinely cool — hitters have a 'dial,' steepening ~10° on low pitches and flattening on high ones, and how much they turn it is one of the most reliable, year-over-year-stable traits in hitting (split-half 0.94; two independent methods agree on who at r=0.99). The second half isn't: the dial adds no power (its effect can't be told from zero once you know a hitter's bat speed and average swing shape), and it doesn't save contact either — the more adjustable hitters whiff slightly more, worst at the zone extremes where matching your plane is supposed to help. 'Ideal attack angle rate,' the marketed metric, is ~88% bat speed. The least-adjustable qualified hitter in baseball is Juan Soto. Two independent methods, two adversarial review rounds; the cross-review caught and killed a contamination bug before it could ship.

June 16, 2026 · 11 min read

The Long Read

The Arm-Angle Gambit: We Went Looking for the Cheat Code. We Found a Tax.

We priced the arm-slot drop within-pitcher, two independent ways: slot credit ≈ 0 five ways, the bundled new pitch pays nothing once a leakage artifact is removed, and droppers quietly lose 0.46 in of four-seam ride (Wilcoxon p=0.011). The gambit is a tax with a brochure.

The findingPitchers

≈ 0

the run-value credit of dropping your arm slot — five constructions, two independent methods, all CIs spanning zero. The only effect that passes a significance test: −0.46 in of four-seam ride (10/15 droppers, Wilcoxon p=0.011).

Read the analysis →

Explore the Data

Game Reports Umpire Profiles ABS Challenges Pitcher Profiles Hitter Profiles Zone Explorer Count Calculator Open Explore →

One investigation a week. No hot takes, just the receipts.

CalledThird Weekly lands in your inbox during the season. Under 500 words — we respect your time.

Or follow on Twitter/X · Bluesky