MLB Research¶
A free, commercially-usable (AGPL-3.0), honest MLB research database: a clean grain ladder of standard and advanced statistics, event-derived back to 1910, every formula cited to its source, every accuracy and leakage limitation documented.
Get the data¶
pip install mlb-research
import mlb_research
batting = mlb_research.load("batting_season", season=2023)
The published tables are Parquet files on Hugging Face; the loader downloads and caches them locally. No database, no cloned repository.
- Dataset: huggingface.co/datasets/cbwinslow/mlb-research
- Run SQL in your browser: Run SQL — DuckDB-WASM against the published Parquet, nothing server-side.
- Source: github.com/cbwinslow/mlb-baseball
What's documented here¶
| Page | What it covers |
|---|---|
| Data dictionary | Every published table, its grain, its columns, its source, its null policy |
| Grain ladder | How game, season, team, and career figures relate — and why rates are recomputed, never averaged |
| Formulas & citations | Every published metric with its formula and the source it is cited to |
| Honest limitations | Coverage boundaries, tie-out tolerances, regular-season-only scope, "missing is not zero" |
What this is not¶
This site documents the public research database. Tuned prediction models, novel metrics still in validation, and betting research are a separate internal product and are not published here. See the project's phased ladder.