Hugging Face Dataset Clone-Cache Snapshot for Message-Visible Data-Refactoring Candidates (2026-07-31)

Full git histories of 270 of the 273 Hugging Face dataset repositories named in data/message_refactoring_candidates.csv of the Hugging-Face-Dataset-Mining project, captured 2026-07-31. Upstream repositories drift: within seven weeks of mining, three were deleted or made private and four had their histories rewritten so the mined commit no longer exists anywhere. This snapshot freezes what still existed at capture time, because an archive can only ever contain what is held when it is made. A bundled MANIFEST.csv lists every mined dataset/commit pair and whether the snapshot holds it: 367 of 374 resolve here; the seven that do not were already unrecoverable at capture. The clones were made under GIT_LFS_SKIP_SMUDGE=1, so git history, commit messages, file trees, and non-LFS file contents are complete, while files stored in Git LFS are present as pointer text only. Head samples of many such payloads are committed in the project repository under data/evidence/. Each of the 270 datasets remains governed by its own upstream license; this snapshot is archived for preservation and verification of the project's labels.