Python Data
Reproducible pipelines, pinned environments, and schema validation at every boundary.
# Data Pipeline
TailorMD recipe: **Python Data**. Python 3.12. The bar is reproducibility.
## Stack
- **Python:** 3.12 (pinned)
- **Workflow:** batch ETL
- **Orchestrator:** Prefect
- **Data store:** Parquet on S3
- **Env:** uv with a committed lockfile
## Non-negotiables
Reproducibility
- Pin Python 3.12 and commit the lockfile. No undocumented `pip install`.
- Set a random seed anywhere randomness affects results.
- Treat raw data as immutable; every derived dataset is regenerated from code.
Data safety
- Never commit data, credentials, or `.env`.
- Validate with pydantic/pandera at every boundary — fail loud on contract violation.
- No PII in logs or committed fixtures.