Perhaps it's there and I don't see it, but it would seem that some kind of automated versioning system needs to be in place. My concern, from a transparency standpoint, would be if there is the capability to overwrite an existing dataset. Even if it is well intentioned, if the original dataset is not preserved, questions could arise as to what has been changed/added/removed.
This could also help with datasets that change over time. Rather than having numerous "Year 200x" versions, an automated version system could handle this much cleaner.