NextPitchAI — hitter's read

What this tool actually does

On 35% of pitches it gives you a read. It is right 72.5% of the time there. A count-split scouting report would be right 68.3% on those same pitches.

Edge: +4.2 points · it says nothing on the other 65%.

Measured on 729,688 pitches from a season the model never trained on (temporal, validate on 2025). This is an overlay, not a replacement for a scouting report. Across all situations a fixed per-count rule is as good or better (+0.7 points). Its value is knowing when to deviate.

Situation

Player names aren't loaded, so the pickers show MLB ids. Run python site/player_names.py --from-serving on a machine with internet, then rebuild the serving bundle.

Pitcher source He throws

Pitcher

Batter (optional — leave blank if it's you) You bat

Balls Strikes Outs

Inning Pitch # in AB Pitch count

1B 2B 3B

Last few pitches (most recent last — this is where most of the edge comes from)

Set the situation and ask.