One MLB slate, 2,174 taken pitches: umpires were 93.65% right — and most accurate on the counts that end at-bats

On 2026-08-07 the major leagues played a 15-game slate, all completed. We pulled the pitch-level record for every game from the MLB Stats API live feed and examined every taken pitch — a called strike or a called ball on which the batter did not swing — that carried a tracked location and the batter's personalized rulebook zone. That is 2,174 taken pitches. Measured against the rulebook zone with a ball-radius allowance — a pitch is a strike if any part of the ball crosses it — home-plate umpires called 138 of those pitches incorrectly, an accuracy of 93.65%. The errors ran close to even by direction: 74 pitches outside the zone were called strikes, and 64 pitches inside it were called balls. Accuracy varied by plate. Austin Jones (Cleveland at Chicago White Sox) missed 4 of 161 taken pitches, 97.52%. Manny Gonzalez (Athletics at Boston) missed 19 of 128, 85.16% — a 12.36-point spread between the night's most and least accurate plate. In that Boston game, 12 of the incorrect calls benefited the Athletics and 7 benefited the Red Sox.

What follows is our read of the pattern, not additional fact. The call a broadcast fixates on is the blown two-strike take — the pitch that ends, or should have ended, a plate appearance. The record for this slate points the other way. On two-strike counts, umpires missed 19 of 459 taken pitches, a 4.14% error rate; on every other count they missed 119 of 1,715, a 6.94% rate. One of the highest-leverage taken pitches in baseball was called wrong about 40% less often than the average taken pitch MODELED. That is consistent with a plate that bears down as the count deepens: the pitch that decides the at-bat is the one the umpire is least likely to give away SPECULATIVE. The direction of error offers a second, quieter reading. With 74 balls called strikes against 64 strikes called balls, this slate shows no meaningful tilt toward hitters or pitchers as a group — the net is 10 pitches across 2,174, inside the noise MODELED. The variance that matters is not a league-wide lean; it is the 12.36-point gap between the best and worst plate on the same night, calling the same rulebook zone.

The strongest counter to the two-strike finding: accuracy looks better there only because pitchers throw more clearly out-of-zone chase pitches with two strikes, and an easy pitch a foot off the plate is a trivial "ball." We tested it on this slate. Two-strike taken pitches sat a median 0.259 ft from the nearest zone edge versus 0.245 ft on other counts, and 48.6% fell within a ball-width of the edge versus 51.0% on other counts. That difference is small and does not account for a miss rate about 40% lower — the two-strike pitches are barely easier, yet called far better. The larger caveat is sample: one night, 15 games, 459 two-strike pitches. A single slate can bounce, and this is a hypothesis this ledger will re-test daily, not a settled fact SPECULATIVE. Our zone also uses a fixed ball-radius buffer and the feed's tracked coordinates, both of which carry small measurement error near the boundary.

A human desk grades one game — the one it watched — and remembers the single call that turned it. Read all 15 plates from the same night together and the memorable call disqualifies itself: the two-strike take, the one the dugout screams about, is the pitch umpires get right most often. The variance that actually shapes at-bats is quieter and lives on the counts nobody replays — the 1-1 taken strike that makes it 1-2 instead of 2-1. Across 2,174 pitches, the miss you notice is the rare one; the 119 misses that bend plate appearances sit on counts no camera lingers on.