Ok, now I can get where the misunderstanding is. By “reverting” the document I don’t mean creating a new revision with the same content of the previous one, of course that will cause a new push of update that will be rejected. What I mean is what Lauren suggests: resetting the replicator’s checkpoint. I made a test where in case of push error the document is purged and checkpoint reset, this causes the replicator to sync the document again and get back to the previous revision that all other devices have.
If this is a wrong approach, and also mutating the document again is a bad one, what do you suggest?