| # Code release checks | |
| | Item | Verification | | |
| | --- | --- | | |
| | Final paper title and 15 authors | `CITATION.cff` | | |
| | Team code and dataset URLs | `README.md` | | |
| | Real source attribution | `PROVENANCE.md`, `LICENSE` | | |
| | Runnable no-data example | `bash scripts/run_tiny.sh` | | |
| | Corrupt files and invalid inputs rejected | `python3 -m unittest discover -s tests -v` | | |
| | Code hashes and required metadata | `python3 scripts/check_release.py` | | |
| | Exact dataset metadata 1.0.2 | `dataset_metadata/croissant.json` | | |
| | Actual reproduction scope | `README_REPRODUCE.md`, `RELEASE_SCOPE.md` | | |
| | Full Blender rendering | Requires separate assets and Blender; not covered by no-data checks | | |
| | Paper-scale NVS and baseline sweep | Not included as complete reproduction entry points | | |
| After modifying tracked code, regenerate `SHA256SUMS`, run tests and checks, and inspect the file list before publication. Do not publish local assets, credentials, checkpoints or generated outputs. A passing smoke test does not establish experimental reproducibility or asset-license compliance. | |