Turn messy datasets into analysis-ready data.
Recent revisions are kept. Restoring one creates a new revision.
