validation/RELEASE_VALIDATION.md

# Release validation — v1.0.1

The release is prepared for a public Hugging Face research repository. Publishing is performed by the user-run BAT/Python publisher with a local write token.

| Check | Result | Scope |
|---|---|---|
| Original model tests | 9 passed | Reference model and included full-run fixtures; no new experiment |
| Publication tests | 13 passed | Mock APIs: account boundary, public/private handling, file allowlist, checksums, conflict protection, retry and anonymous-byte verification failures |
| Export schema/provenance checks | Passed | 1,132 records across six configurations, exact source mappings and codeword distance |
| Scientific source preservation | Passed | 27 scientific files byte-identical; supplement Markdown has exactly one recorded notation-formatting correction |
| Citation metadata | Passed | CFF validated against the official citation-file-format schema; author and release identity consistent |
| Hub README | Parsed by huggingface_hub 1.32.0 | Six explicit configs, one default; no server-side viewer ingestion claimed |
| Publisher dependency availability | Confirmed on PyPI | huggingface_hub 1.32.0; Python requirement >=3.10 |
| Windows launcher | Inspected, CRLF format checked | Not executed on Windows in this Linux environment |
| Publication itself | Requires user-run uploader | No write token was provided here; no authenticated upload was attempted |

The scientific PDFs retain the original release's reviewed layout. No new physical evidence, experimental result or physical-threshold validation is introduced by publishing metadata or passing software tests.

Machine-readable results: [release_checks.json](release_checks.json). Test output: [publisher_tests.txt](publisher_tests.txt) and [model_recheck.txt](model_recheck.txt). Source-preservation record: [SCIENTIFIC_BASELINE_SHA256.json](SCIENTIFIC_BASELINE_SHA256.json).

Recheck offline with `python tools/validate_release.py`, `python publisher/test_publish.py` and `python publisher/publish.py --dry-run`. The first needs requirements-validation.txt; publisher tests and dry-run need only Python's standard library.

The successful local publishing run verifies every checksum-listed file by anonymous download at the actual uploaded commit and writes publication_receipt.json. Neither the prepared ZIP nor these offline tests asserts that the repository is already live, searchable, endorsed or indexed in the Dataset Viewer.