MLB Models — Inputs Overview
High-level view of what each model family feeds on · ✓ uses · ~ partial/proxy · — doesn't
Two complementary engines. The production models are top-down (team & season-form aggregates + park + weather + bullpen). The per-pitch sim is bottom-up (pitch-level data + per-batter profiles + actual lineups). They share very few inputs, so they look at the game differently.
No model uses the betting line as an input. By design — the goal is calibrated accuracy on real outcomes; the market line is only the benchmark for edge/value, never a feature.
Model × input family
PROD-ML = production moneyline (full-game) · PROD-TOT = production totals · SIM = per-pitch sim (first-5-innings markets) · SIDE = edge/bet-selection layer.
| Input family | PROD-ML | PROD-TOT | SIM | SIDE |
|---|
| PITCHING |
| Season rate stats (ERA / WHIP / K / BB / HR) | ✓ | ✓ | — | — |
| 30-day rolling form (xFIP, K-rate, contact, velo) | ✓ | ✓ | — | — |
| Recent workload | ✓ | ✓ | — | — |
| Handedness | ✓ | ✓ | ✓ | — |
| Pitch-level detail (velo / movement / location / mix) | — | — | ✓ | — |
| HITTING / LINEUP |
| Team offense rolling form | ✓ | ✓ | — | — |
| Offense vs opposing handedness | ✓ | ✓ | ~ | — |
| Lineup-aggregated platoon | ✓ | ✓ | — | — |
| Actual lineup order | — | — | ✓ | — |
| Per-batter contact/quality profiles | — | — | ✓ | — |
| Baserunner advancement | — | — | ✓ | — |
| CONTEXT |
| Park factor | ✓ | ✓ | ✓ | — |
| Weather | ✓ | ✓ | — | — |
| Bullpen workload | ✓ | ✓ | — | — |
| MARKET |
| Betting line — as a model feature | — | — | — | — |
| Betting line — as edge benchmark | ~ | ~ | ~ | ✓ |
Overview only. Production models refit continuously on completed games; the sim runs point-in-time per slate.