The ABS challenge system runs out of challenges before it runs out of bad calls

Baseball spent 150 years arguing about balls and strikes. In 2026 it started settling those arguments in seconds, on the video board, with a camera. Tap your helmet, and Hawk-Eye tells 40,000 people whether the umpire was right.

The system is popular, it's fast, and — this is the part worth sitting with — it mostly proves the umpires are good. Across 212,221 called pitches in the first half, big-league umpires got the call right 92.6% of the time. That's the honest headline nobody clicks on.

Here's the one they should.

The ninth-inning double-whammy

I pulled every ABS challenge of the first half — 6,053 of them — from MLB's public game feeds and tracked two things by inning: how often a challenge succeeded, and how often a clearly-blown call went unchallenged because the team had no challenges left to spend. Both curves bend the wrong way, at the same time, in the same inning.

In the first three innings, challenges succeed ~59% of the time and almost no blown call goes uncorrectable. By the ninth, challenge success has fallen to 40% — and more than a third (36%) of obvious missed calls can't be challenged at all, because the challenging team is tapped out.

A system built to correct the calls that matter most is least able to correct them in the moments that matter most.

The obvious objection is that the ninth is full of blowouts where nobody's trying. It isn't the explanation. Split by score: in tie-or-one-run ninth innings — the ones that decide games — challenges still succeed only ~47%, and the availability collapse is identical to blowouts (~26% of clear misses uncorrectable either way). The moot rate doesn't care whether the game is close; nine innings simply drains the bank. That failure is structural, not situational.

And it doesn't stay in the regular season. The challenge system is used in October. A team that has spent both challenges has no recourse on a blown call in a win-or-go-home ninth — the exact situation the system was sold to fix.

First, why you should believe any of this

Before the rest of the findings, the part that makes them trustworthy — and the part that turned out to be a genuine surprise. MLB's own public feed is hiding about a quarter of all challenges. When you pull challenges from the StatsAPI game feed the way any analyst naturally would — off the pitch record — you silently miss every challenge that ends an at-bat (a created or confirmed third strike, a fourth ball). Those live in a different, play-level field.

Skip them and you drop ~25% of all challenges, every one of them terminal, and report an inflated overturn rate (~55%) instead of the true one (53.0%). Recovering those hidden challenges is what makes this dataset complete — and the proof it's right is that the corrected total reconciles to Baseball Savant's official count within 1% (6,053 vs 6,040), role splits nearly identical, pitcher count exact. If your challenge numbers don't match Savant's, this is probably why.

The strike zone itself is reconstructed from pitch coordinates and calibrated against actual ABS verdicts — it agrees ~87% of the time on borderline pitches, near the ceiling given tracking noise. That's precise enough to trust the shapes and rankings here; treat any single percentage within about an inch of the edge as an estimate. And it cross-validates: the league overturn rate matches Savant, the accuracy figure matches Umpire Scorecards, and the umpires who grade worst here are the ones the challenge data independently flags. Two unrelated methods, same answer.

Why it fails late: everyone panics at the edge of their nightmare

The late-game collapse isn't the pitches getting harder. On matched pitch difficulty, late challenges are on the same borderline pitches as early ones — they just succeed less. It's the humans. And the pattern is beautifully specific: each role falls apart as it approaches the outcome it fears.

Pitch difficulty can't produce a mirror like that, because the pitch doesn't know who's challenging or which way the count is running. The symmetry is the tell: this is emotion, not optics. Nowhere is it more human than the hitter's last strike.

Hitters are terrible at challenging strike three

A third of all hitter challenges come on a called third strike — and they're wrong 61% of the time. Down to their last strike, hitters throw a Hail Mary at a pitch that was usually a clear strike, lose, and get rung up anyway. 572 times this half, a hitter challenged his own strikeout, lost, and had it confirmed to the whole ballpark on the video board. Same pitches, worse decisions. It is the purest desperation in the dataset, and it is very, very human.

What it costs: the calls that get away

Because I computed the ABS verdict on every taken pitch, not just the challenged ones, I can count the calls the system left on the table: clear umpire misses — at least an inch and a half past any measurement noise — that went unchallenged. There were about 2,000 with a challenge still available, roughly 1.7 per game a team simply didn't catch. The best teams convert only about half of the catchable misses in their favor; the worst, under a quarter.

And then the cruelest bucket: the clear blown call in a spot where the team had no challenge left — not a strategic pass, a structural dead end. A hitter rung up on a pitch inches off the plate in the ninth, empty challenge bank, nothing anyone can do. That's not a team failure. That's the system reaching its limit exactly where the stakes are highest.

The umpires are fine. The mechanism isn't.

None of this is an umpire story. Grade every taken pitch against the ABS zone and the accuracy curve is reassuring: a coin flip right at the edge, near-perfect a couple of inches out, 92.6% overall.

The finding isn't that umpires are bad. It's that the challenge system, layered on top of good umpires, has a hole shaped exactly like the moments it was built for. As MLB weighs whether to move from challenges to a full automated zone, that's the sharpest empirical input available:

The safety net develops its biggest holes where the fall is longest — including October. The challenge system runs out of challenges before it runs out of bad calls.