ABS Challenge System: Umpires Overturned 44% of the Time

We pulled 49,510 pitches from the first 12 days of 2026 spring training and found 185 ABS challenges. The overturn rate, the strike vs ball breakdown, and what it tells us about the new system.

185 challenges · 49,510 pitches · 12 days

Start-to-Start Autocorrelation Is Basically Zero

Pitcher earned run outcomes from one start to the next have a correlation of 0.0008. What that means for "bounceback" narratives and recent form as a predictive feature.

13,604 starts · 12 features · 48 tests · 0 passed

Why Unders Win More Often Than They Should

Baseball scoring distributions are right-skewed. Games can blow out to 15 runs but can't go below zero. This creates a structural frequency advantage for under bettors that books can't fully eliminate.

We Scanned 16 Markets and Killed Most of Them

Stolen bases, batter strikeouts, pitcher hits allowed, team totals, game totals, moneylines. We tested them all with walk-forward backtesting on real closing odds. Here's what didn't survive.

Features That Don't Predict Pitcher Performance

Umpire tendencies (YoY r = -0.01). Catcher framing (YoY r = 0.108). Rolling box-score features (0 of 12 passed validation). Rest days (near-zero correlation). The graveyard of things we tested.