| pretty_name: FastPLMs research and validation artifacts | |
| license: other | |
| license_name: source-specific-terms | |
| license_link: https://github.com/Synthyra/FastPLMs/blob/main/docs/licensing.md | |
| # FastPLMs artifacts | |
| This dataset holds the measured reports and golden regression tensors maintained by [FastPLMs](https://github.com/Synthyra/FastPLMs). It is not a training dataset or a set of model checkpoints. | |
| The source repository keeps `evidence.toml`, which pins an immutable dataset revision and each payload's SHA-256 digest and size. Fetch explicitly with `python -m tools.artifacts.evidence_store fetch` before offline documentation, release, or parity checks. | |
| Paths preserve the source workspace layout: `docs/evidence`, `docs/validation`, and `tests/goldens`. Runtime configuration, model manifests, dependency locks, and small synthetic fixtures remain in Git. | |
| ## Evidence status | |
| The v2 confidence reports preserve historical numeric results. Their `metrics_review` fields mark correlations, bootstrap intervals, and acceptance gates as requiring recomputation after the tied-rank correction. Raw per-target predictions were not recovered from the closed training workstation or W&B. These archived values must not be presented as corrected evaluation results. No v2 confidence weights are included. | |
| Other artifacts retain their recorded source revisions, environments, and evidence boundaries. Presence in this dataset is not an assertion that all release gates pass. | |
| ## Terms and provenance | |
| Payloads originated in the public FastPLMs repository at PR #49 (`1ae17f2b3955db28fe9c076b5c7e2394c6fc83ac`), with review-status annotations added to the v2 confidence reports. Golden metadata identifies originating checkpoints and source revisions. Applicable model and source terms remain in effect; migration does not relicense them. Consult the source repository's `LICENSES`, model manifest, and licensing documentation. No blanket permissive license is asserted for the combined collection. | |