Back to projects August 2026 - Present

A Moving Baseline: Measuring MLB's Strike Zone

  • R
  • Python
  • Statcast
  • Pre-registration
  • GAMs
  • DuckDB

Introduction

For clubs evaluating catchers and pitchers, framing and pitch-location comparisons across seasons must account for changes in what umpires call strikes. Public estimates of the 2026 shrink circulated within weeks (Clemens, FanGraphs, 28 April 2026). In 2025 the umpire grading buffer narrowed from two inches outside the edge to three-quarters of an inch either side (Andrews, FanGraphs, 5 May 2025); in 2026 regular-season ABS challenges began. The question is how much the zone changed at each step, against three old-rule seasons on one measurement plane; no published estimate I found gives both with intervals.

Methods

I re-projected 2022-2025 pitches to Statcast’s 2026 plate midpoint. The main height rule used roster height plus a fixed offset throughout. Sample, from Statcast and the MLB game feed: 1,830,231 called pitches in 12,060 regular-season games, 2022 through 21 September 2026. Near the edge, I fitted a binomial GAM to the umpire’s original, pre-challenge call by location and season, with umpire effects partially pooled. Each season’s zone is its 50 percent contour; model-conditional intervals (95% unless marked 90%) use 1,000 coefficient draws and omit season-to-season variation. With ABS-measured heights, the rebuilt geometry, not the fitted model, matched 10,164 of 10,168 ABS challenge verdicts. I measured both steps net of the 2022-2024 trend. The placebo pair, 2023 and 2024, had no change to the zone or umpire grading.

Results

The pre-registered placebo failed: the zone grew 10.8 (90% CI 7.7 to 13.9) square inches, beyond its ±3 margin, so every step is descriptive. Net of the trend, the area fell 23.4 (95% CI 18.6 to 28.3) square inches in 2025 and 33.1 (95% CI 28.7 to 37.5) in 2026 (Figure 1). The zone lost width mostly in 2025 and height mostly in 2026 (Table 1). Within 3 inches of the rule-book zone’s edge, the called-strike rate fell 10.69 (95% CI 10.04 to 11.28) percentage points from 2024 to 2026. Clemens’s comparison reads 2025 at the plate front and 2026 at mid-plate. That convention gives a loss of 20.0 (95% CI 13.0 to 28.2) square inches; with both at mid-plate it is 32.8 (95% CI 29.3 to 36.6). The binned estimator failed the ±3 agreement tolerance, giving contractions of 29.5 and 40.7 square inches. Top-edge intervals covered 91 of 100 recovery runs, below the 93 bound.

Dot chart of strike-zone area by season, 2022 to 2026, in square inches with 95 percent bars. The points are 2022 494.6, 2023 484.3, 2024 495.1, 2025 471.9, 2026 439.1. Pale period bands mark 2022–2024, 2025 and 2026 (through 21 September). Downward arrows show the steps net of the 2022–2024 trend. Each runs from the prior season's level plus the trend down to the fitted level. They are 2025 −23.4 (95 percent: −28.3 to −18.6) and 2026 −33.1 (−37.5 to −28.7). A small rust bracket between 2023 and 2024 marks that change, +10.8 sq in (90%: 7.7 to 13.9), labelled no zone rule change. Both arrows are longer than that bracket.
Figure 1. Fitted 50 percent contour areas (72-inch batter, 2024 pitch mix), with 95% model-conditional intervals. Arrows are the 2025 and 2026 steps net of the 2022–2024 trend of +0.3 sq in per season (95%: −1.6 to 2.2), each from the prior season plus that trend. The bracket is the 2023-to-2024 change, 10.8 sq in (90%: 7.7 to 13.9), with zone and grading rules unchanged; that placebo failed. Grading changed in 2025, regular-season challenges began in 2026; neither step is attributed. A confirmatory sealed-set calibration check is pending.
Table with six rows and five numeric columns: top edge and bottom edge and half-width in inches, area in square inches, and shadow-band strike rate in percentage points, each cell a point estimate with a 95 percent interval. Rows: the 2022-2024 trend per season (area 0.3 [-1.6, 2.2] sq in); the 2023-to-2024 change with the zone rules unchanged, the pre-registered placebo with a 90% interval, area 10.8 [7.7, 13.9] sq in against a margin of 3 and shadow-band rate 1.54 [1.03, 2.04] pp against 0.5, marked FAILED; the 2025 step net of trend, area -23.4 [-28.3, -18.6] sq in; the 2026 step net of trend, area -33.1 [-37.5, -28.7] sq in; the 2024-to-2026 total, area -56.0 [-59.6, -52.0] sq in; and, below a rule, the measurement-plane adjustment, not a period change, 2025 read at mid-plate minus at the front, area 12.9 [6.3, 19.2] sq in. The placebo row also carries edge cells added after the tag and not pre-registered, each with a 95 percent interval: top 0.47 [0.37, 0.57], bottom 0.16 [0.08, 0.24] and half-width 0.12 [0.06, 0.18] inches. A footnote states that no step is attributed to its rule, that top-edge intervals undercovered in the recovery test, and that the shadow band is the pitches within 3 inches either side of where the ball just touches the rule-book zone's edge. Its last lines, exploratory and not pre-registered, give the 2022-to-2023 area change, -10.3 [-13.4, -7.3] sq in at 90 percent, and the 2025 area step against the 2022-2024 mean plus one season of trend, -19.7 [-23.3, -16.1] sq in, and against 2023 alone, -12.4 [-15.8, -8.5] sq in.
Table 1. Changes in the fitted 50 percent strike contour (72-inch batter, 2024 pitch mix, mid-plate), by period: the trend, the failed placebo pair, the two trend-adjusted steps, the unadjusted 2024-to-2026 change, and the measurement-plane adjustment. The note under the table defines the cells. A confirmatory sealed-set calibration check is pending after the postseason.

Conclusion

The 2025 zone was already smaller than any 2022-2024 zone, before regular-season challenges began, so framing and location baselines that start from 2025 miss that step. Table 1’s shifts serve three uses: season-adjusted framing and pitch-location baselines, edge targets for pitchers, and the plane correction for 2025-to-2026 Statcast comparisons. Main limit: the failed placebo; the design cannot assign either step to its rule. The sealed-set definition and acceptance criteria were committed publicly before the sealed set was opened. Games from 22 September 2026 on remain sealed for a confirmatory calibration check after the postseason.

Explore the data

The zone, the planes, the challenges

Drag, step and hover. Every number is also in a table.

Status

  • Pre-registration: frozen at tag prereg-v1.
  • Data through 2026-09-21.
  • Sealed: regular-season games from 2026-09-22 and the postseason.
  • Panel A: study data. Panel B: study data. Panel C: study data.

Strike zone explorer

STUDY DATA2022 to 2026, binned shares

Look at the gold line. It is where a taken pitch is called a strike half the time. The pale rectangle is the rule-book zone.

STUDY DATA2022 to 2026, binned shares

Surface: League

Per-umpire surfaces are not published. The design cannot rank individual umpires reliably, and the pre-registration rules them out.

Surface

Keys 1 to 4 choose a view. Key 2 looks straight on from behind the plate, which is the view to read values from. Tab reaches the probe: arrow keys move it one cell, Enter pins it, Esc clears the pin, C switches the cut.

Drag to turn. Arrow keys turn the view when it has focus. T shows the table. R resets.

The share of called pitches that were called strikes, in cells of 1 in, at the middle of the plate, seen from behind the plate. Heights are scaled to a batter 72 in tall. Depth and colour show the same value, and 1.00 stands 8.5 in toward you. The gold line is the 50% contour. The pale rectangle is the rule-book zone. The line 1.45 in outside it is where the ball just touches the zone. Cells with fewer than 30 pitches are not drawn. These are binned shares, not the fitted surfaces of the study. Stepping between periods shows what changed and when, not why. The study’s pre-registered placebo failed: from 2023 to 2024, with no change to the zone or umpire grading, the zone still grew beyond its margin, so every step is descriptive.

A surface stands upright over home plate and is seen from behind the plate. Left to right is position across the plate in inches. Up is height, scaled to a batter 72 in tall. The surface comes toward the viewer where a called strike is more likely, by 8.5 in where the probability is 1.00. Colour repeats the same value, from dark blue at 0.00 to pale blue at 1.00. A gold line marks the 50% contour. A pale rectangle marks the rule-book zone, 17 in wide, from 27% to 53.5% of batter height. A control with three stops switches between 2022 to 2024, 2025 and 2026. The surface is the league surface of that period. A list chooses a second contour to draw beside it. Moving a probe over the surface reads the probability at that point and draws the profile of the surface through it.

Source: MLB Statcast via Baseball Savant and the MLB Stats API. Aggregates only.

Pitch flight and the two planes

STUDY DATA2022 to 2024, mean path per pitch type

Look at the two marks each pitch leaves, one on each plane. The mark on the gold middle plane is lower, and the gap is larger for the curveball than for the fastball.

STUDY DATA2022 to 2024, mean path per pitch type

At the two planes

Height difference by pitch type

Drag to turn. Arrow keys turn the view when it has focus. T shows the table. R resets.

Each line is the mean path of one pitch type from release to the plate, at true scale. The paths are means over called pitches from 2022 to 2024. Left-handed pitchers are mirrored, so every path is a right-hander’s. Through 2025, Statcast reported where a pitch crossed the front of the plate. In 2026 it reports the middle of the plate, 8.5 in further on. A pitch is still falling there, so it crosses the middle plane lower, and a steeper pitch drops more. The study moves every pitch from 2022 to 2025 to the middle plane along its own path.

A scene shows home plate, 17 in wide, and the paths of two pitches from the release point to the plate at true scale. Two upright frames stand over the plate: the front plane at the front edge of the plate, and the middle plane 8.5 in behind it. Each path leaves a mark where it crosses each frame. Two small side views show the last 20 in of each path, with the ball drawn at its real radius of 1.45 in, and measure how much lower the pitch is at the middle plane than at the front plane. A chart gives the same difference for every pitch type in the file.

Source: MLB Statcast via Baseball Savant and the MLB Stats API. Aggregates only.

Challenges along the zone edge

STUDY DATA2026 regular season, through 2026-09-21, locations only

Look at the gold line and at how the strips change as they move away from it. The gold line is the ABS boundary, where the ball just touches the zone.

STUDY DATA2026 regular season, through 2026-09-21, locations only

Zone check

10,164 of 10,168

99.96% of all challenged pitches

The zone rebuilt from each batter’s measured height gave the same in-or-out verdict as ABS on these challenged pitches.

Traced to abs-umpires: out/ch1/tab/T2_zone_gate.csv, receipt W3.8

Pitches comparedSame verdict as ABSPercent
Outside the half-inch band7,573 of 7,57499.99%
Inside the half-inch band2,591 of 2,59499.88%
Top edge1,707 of 1,707100.00%
Side edges4,979 of 4,98099.98%
Bottom edge3,478 of 3,48199.91%

Arrow keys move between strips when the view has focus. T shows the table. R resets.

Each strip counts the regular-season challenged pitches of 2026 at one distance from the ABS boundary, in half-inch steps, for one edge of the zone. The data runs through 2026-09-21. The distance has the ball radius taken off, so 0.00 in means the ball just touches the zone and a negative distance is a strike by ABS. Unrolled, the four edges share one distance axis. Strips with fewer than 30 pitches are not drawn. The strips show where challenged pitches were. They do not show the umpire’s call or whether a challenge was overturned.

A drawing of the strike zone from the catcher’s view, with bands around its four edges. Each band covers half an inch of distance from the ABS boundary, the line where the ball just touches the zone. A band is brighter where more challenged pitches fell. A switch straightens the four edges into four columns that share one distance axis, with distance outside the zone upward. Two handles set a range of distance, and the panel counts the challenged pitches inside it.

Source: MLB Statcast via Baseball Savant and the MLB Stats API. Aggregates only.

  • surfaces.json. MLB Advanced Media and Baseball Savant, read from the abs-umpires warehouse (open analysis set, MLB regular season, data through 2026-09-21). Written by scripts/abslab/export_real.py. Generated . sha256 bdf288c3f189.
  • flight.json. MLB Advanced Media and Baseball Savant, read from the abs-umpires warehouse (open analysis set, MLB regular season, data through 2026-09-21). Written by scripts/abslab/export_real.py. Generated . sha256 d3bdb5917985.
  • challenges.json. MLB Advanced Media and Baseball Savant, read from the abs-umpires warehouse (open analysis set, MLB regular season, data through 2026-09-21). Written by scripts/abslab/export_real.py. Generated . sha256 da9784260c91.

Data: MLB Advanced Media and Baseball Savant. Aggregates only; raw feeds are not redistributed.

Copyright 2026 MLB Advanced Media, L.P. Use of any content on this page acknowledges agreement to the terms posted here http://gdx.mlb.com/components/copyright.txt

https://github.com/hpagni/abs-umpires

Final paper coming December 2026.