eval-freezelisted

# Eval Freeze Run ID: `$0` Arguments: `$ARGUMENTS` If no run ID was provided, stop and show: ```text /eval-freeze <run-id> [--suite <path>] ``` A run created with `--suite` must be frozen with the same `--suite <path>`: the evaluator reference is copied from there. Freeze refuses the mismatch either way. Set `<run-dir>` to `evals/runs/$0`. ## Procedure 1. Confirm `<run-dir>/RESULT.md` and `<run-dir>/venture/venture.yaml` exist. 2. Do not read `evals/reference/`, the live rubric, or any evaluator-only content. 3. Run: ```bash npm run evidence:check -- evals/runs/$0/venture ``` Stop if it fails. Repair state integrity before freezing. 4. Run: ```bash npm run eval:freeze -- $ARGUMENTS ``` The deterministic freeze primitive independently revalidates the case, canonical venture state, complete decision, evidence/experiment integrity, effective framework hash and evaluator-source hashes. Only after those checks pass does it create `<run-dir>/evaluator/` with the frozen reference/rubric/scoring contract. 5. Run: ```bash npm run eval:verify -- $0 ``` 6. Require successful integrity and evaluator-provenance verification. Report the frozen artifact SHA-256 and framework SHA-256. 7. Do not modify any frozen artifact or evaluator input after this point. `scores/` — and a legacy root `SCORE.md` — is intentionally the only scoring output outside the frozen digest. ## Completion Tell the user to open a **fresh Claude Code session** and run