It is not just spaghetti code, though. Right? It is code that does not constantly check that input fits an expected pattern. Which is most code. I'm reminded of this talk by Sussman.[1] Essentially, in software we are (understandably) adamant about not modifying the input of our application for "pure" functions. Nature goes right the opposite, often attempting to make the input work at all costs.

So, whereas a schema based application that models a user with an address is unable to cope with being given two addresses. A schema-less approach where it just stores blobs simply stores what it was given. If code that reads this is unable to make sense of multiple addresses, it will raise an error. Not necessarily unlike a user being told to send a package to an address, but given a list of addresses.

[1] http://www.infoq.com/presentations/We-Really-Dont-Know-How-T...