Still not sure I buy it. I can extract a document representation from my relational data that matches your document store at read time. I can effectively choose my schema when I read, there’s no need to rewrite it at the storage layer.