# Timestamped per-run report copies are ephemeral; the canonical latest run is
# committed as report.md. Re-run `run_cap_eval.py` to regenerate.
results/
