# Contributing data snapshots

1. Create a personal fine-grained Hugging Face token with write access to this Dataset.
2. Authenticate locally with `hf auth login`.
3. Add a new immutable directory under `snapshots/<snapshot-id>/`.
4. Include a manifest with refs, model revisions, validation status, limitations, exclusions, and
   a content digest.
5. Add or update compact JSONL views under `tables/`.
6. Update `index.json` only after the candidate Validator accepts the snapshot.
7. Upload with `hf upload graycatHCO3/agent-env-scaling-data . . --repo-type dataset`.

Never share or commit a token. Historical failures may be retained only when clearly marked and
must not replace the current accepted snapshot.