Catch Spark training pitfalls before a long run.
Recent revisions are kept. Restoring one creates a new revision.
