# Dataset Safety & Provenance Checklist (RD1–RD10)

Version: 1.0.1 · Date: 2025-12-03

- [x] PII/secret scrub: no tokens/URLs; build/test logs redacted. Attested by DSSE when signing manifest.
- [x] License compatibility: all cases authored in-repo under Apache-2.0; third-party snippets none. NOTICE up to date.
- [x] Feed/tool lockfile: manifest.sample.json pins hashes for schemas, scorer, builder, and baseline submissions (when present).
- [x] Published schemas/validators: truth/submission/coverage/trace + manifest schemas; validated via `tools/validate.py` and `tools/verify_manifest.py`.
- [x] Evidence bundles: coverage + traces + attestation + sbom recorded per case (sample manifest).
- [x] Binary case recipe: `cases/**/build/build.sh` pinned `SOURCE_DATE_EPOCH` and env templates under `benchmark/templates/determinism/`.
- [x] Determinism CI: `ci/run-ci.sh` + `tools/verify_manifest.py` run twice to compare hashes; Java track uses vendored Temurin 21 via `tools/java/ensure_jdk.sh`.
- [x] Signed baselines: baseline submissions may include DSSE path in manifest (not required for sample kit); rulepack hashes recorded separately.
- [x] Submission policy: CLA/DSSE optional in sample; production kits require DSSE envelope recorded in `signatures`.
- [x] Semantic versioning & changelog: see `benchmark/CHANGELOG.md`; manifest `version` mirrors dataset release.
- [x] Offline kit packaging: `tools/package_offline_kit.sh` produces deterministic tarball with manifest + schemas + tools.