evidence/current/claim5/EVAL.md

Claim 5 evaluation

Run uv run python repro/src/verify.py. It re-reads all four 100-run raw bundles, prints every endpoint result, and exits nonzero for an incomplete target set, malformed shape, non-finite value, missing baseline, or failed CBO-versus-BW comparison.