docs/dataset_report.md

# Dataset Report

## Summary

- Normalized drugs: 10
- Drug classes: 10
- Interactions: 2
- Graph edges: 18
- Synthetic patients: 20
- Scenarios (easy/medium/hard): 100/200/200
- Retrieval corpus documents: 8

## Source Policy

- Core vocabulary/interactions are treated as core sources.
- Secondary sources are allowed fallback with explicit provenance.
- Weak/noisy safety signals are labeled as such in provenance metadata.

## Artifacts

Artifacts are stored under `data/processed`, `data/scenarios`, and `data/artifacts`.