PN — Prepare datasets before the first run with puma prepare-datasets, which…

Prepare datasets before the first run with puma prepare-datasets, which downloads sources, applies stratified sampling (seed=42, 50 items/class for the balanced 200-issue Jira set), splits TAWOS 80/20, and verifies SHA-256 checksums.

Source: PUMA project documentation · Traceability: corpus unit ANXHN-003 · Confidence: verified-at-primary-source