Catcher Framing (Challenge Era)
Called strikes above expectation, valued by count, and re-measured for the ABS challenge system.
What it predicts
Runs saved by getting borderline pitches called strikes, per catcher-season.
Inputs
- Every taken pitch 2021–26 (Statcast): location, zone height, count, batter and pitcher hand, pitch family, catcher
- Each challenged pitch’s original umpire call (Savant ABS data, 2026)
- Savant blocking, throwing and failed-challenge costs for the Catching page’s total
Method
- Expected strike probability from a gradient-boosted classifier per season; framing = (called strike − expected) × the run value of a strike over a ball in that count
- Shrunk with empirical-Bayes ballast from the spread across catchers vs. the noise of each called pitch
- 2026 measured twice: on final calls (what counts) and on the umpires’ original calls, which splits the drop between umpires and overturned challenges; the split is reported on variances, with each season’s noise estimated from the ballast
- Catching total = final-call framing + blocking + throwing + the cost of failed challenges vs. average: a challenge the catcher won is already a called strike, so it is counted once
- Projections re-value pre-2026 framing at the challenge-era rate (the 2026 spread over the 2024–25 spread)
How well it works
Computed by the pipeline from the data on this site
- Agreement with the public measures (catchers with 2,000+ called pitches)
- Savant 2023: r = 0.97 · 2024: r = 0.97 · 2025: r = 0.98 · 2026: r = 0.92 FanGraphs 2021: 0.85 · 2022: 0.94 · 2023: 0.95 · 2024: 0.94 · 2025: 0.93 · 2026: 0.92
- Year-to-year repeatability (framing per pitch, same catcher)
- 2021→2022 r = 0.50, slope 0.502022→2023 r = 0.58, slope 0.632023→2024 r = 0.66, slope 0.632024→2025 r = 0.67, slope 0.652025→2026 r = 0.46, slope 0.34
- The challenge era (2026)
- Spread among regular catchers 8.2 → 4.7 runs (57%); on the umpires’ original calls 5.4 Measured on variances, so 83% of the drop is umpires calling a tighter zone, and the rest is challenges reversing calls. Projections re-value pre-2026 framing at 57%.
- 2026 catcher projections: framing as published, re-valued, and dropped
- RMSE 1.139 as published → 1.096 re-valued; 1.082 with pre-2026 framing dropped (WAR, 98 catchers) Dropping pre-2026 framing scored best on this in-sample test, by 0.014 WAR. We keep 57% because 2026 itself measured framing at 57% of its old spread, not at zero. The re-value rate comes from 2026 itself, so this checks consistency; 2027 is the first blind test.
Data: every taken pitch 2021–26 (Statcast); Savant and FanGraphs framing for the agreement check; 2026 ABS challenges (Baseball Savant).
What it can't see
- One season of the challenge era; the rate will be re-measured every year.
- The 2026 zone was redefined (fixed to each batter), so part of the drop credited to umpires may be the zone itself.
- Umpire identity isn’t in the public data, so zone differences by umpire are noise here.
- Game-calling and pitcher comfort aren’t measurable from public data.
- The Winter Plan values catchers through FanGraphs WAR, whose catcher defense uses FanGraphs framing, not this model.
How we'd measure impact
- Catcher projection error in 2027 (the first blind test of the re-valuation)