ABS Recognition Study Methodology

This appendix documents the fixed research population, reconstruction, validation, terminology, and temporal tests behind “One in Five: Inside the ABS Recognition Gap.”

Scope and frozen snapshot

The analysis covers 2,195 MLB regular-season games from March 25 through September 9, 2026. The accepted snapshot contains 645,793 physical pitches. Every published Article 1 number was frozen after a separate publication-validation review; the date range was not expanded afterward.

Reconstructed ABS geometry

Pascal reconstructed the 2026 ABS decision environment from public pitch location and strike-zone measurements. A pitch was classified by whether the modeled ball intersected the reconstructed ABS zone. The source measurements were retained separately from derived classifications, and boundary distance was stored as a signed value before conversion to inches.

Official-decision validation

The reconstruction was compared with all 9,485 official ABS challenge decisions in the snapshot. It agreed on 9,482 decisions, or 99.968 percent. All three disagreements were preserved. Two were within 0.005 inches of the reconstructed boundary; the third involved a substantial difference in the source zone-top value.

Recognition population

The reconstructed zone classified 11,704 called strikes as estimated incorrect. Of these, 10,755 occurred while the batter had a legal challenge available. The exclusions were 789 opportunities after the affected team had exhausted its challenges and 160 pitches subject to the position-player-pitching restriction. There were no unknown or other exclusions.

Recognition is an observable action: a batter challenged a legal estimated incorrect called strike. Challenge outcome is not part of this definition. The recognition rate is therefore 2,112 divided by 10,755, or 19.64 percent.

Boundary sensitivity

The publication review repeated the analysis after excluding every pitch whose absolute reconstructed boundary distance was at or below 0.05, 0.10, 0.25, and 0.50 inches. The recognition rate rose from 19.64 percent with no exclusion to 22.51 percent under the 0.50-inch exclusion. The positive association between miss distance and challenge action remained under every threshold.

Temporal predictive testing

Fixed logistic-regression specifications were evaluated with expanding monthly holdouts: each test month followed all training observations. The model blocks added geometry first, then measured pitch characteristics, then situation. Situation included count, outs, runners, inning, score differential, and remaining challenge inventory. A later batter analysis estimated player information from earlier games and evaluated it on later observations.

Situation produced a substantial held-out improvement. The pitch-characteristic block did not add validated held-out predictive value beyond geometry in the accepted specification. These are predictive findings, not causal claims.

Publication validation

Before drafting, Pascal separately checked upstream artifact integrity, all three geometry disagreements, boundary sensitivity, funnel arithmetic, metric definitions, representative examples, prior work, claim language, and frozen publication numbers. The review also distinguished recognition rate from overall challenge rate, challenge success rate, and correction rate.

This appendix describes the accepted Article 1 analysis. Future articles may examine batter differences, feature contributions, and challenge-resource management, each under its own validation scope.