That looks like the massively old design we used for MongoDB initially. Was the CouchDB one started that early? Or was is simply a Tuple serialization scheme the contributor used?
If anything we should be almost like MongoDB (except for the hideous _id).