Hmm .. I think you might have that the wrong way around. URLs do exactly identify a document in the sense that the document that is returned is the document in question. That you might get a different document each time is a different matter and relates, again, to value, not identity. (Then there's idempotency which is a useful concept that only appears to causes eyeballs to roll. Even the top in in Google is snarky.) And identity and value are both arbitrary -- John today is John tomorrow, but even that will also cause philosophers to twitch. I guess it not that surprising that programming languages and the like encode the same confusion!
The confusion only goes to show that there is no one obvious definition. That's the beauty of the relational model. John is John, but what it can identify is context-dependent (a string, a human in a group of humans, a human in a group of humans in time?).
