Document IDs Provided by Metalearning Database and References

Should document IDs be provided by the metalearning database? Is there a case for some, but not all, of the collections? Should references use user-generated IDs or DB generated IDs?

Users can provide document IDs. Validation can prevent duplicate IDs from being inserted, placing the onus of deduplication efforts on the user. When there is an ID collision (extremely rare for UUID, unless the user forgets to update the ID), the user has to decide whether the existing and new documents are the same. If not, the user can generate a new ID.

This would also allow teams maintaining private databases to keep their IDs without having to maintain a separate mapping between databases.

Any other arguments in favor of or against this?