So I guess, when it comes down to the wire, what wins out, the needs of real users, or the documented description of a protocol?

The fact that said protocol is incapable of addressing real needs is a failure state, and historically this has been addressed by building better tooling on top of the protocol, or replacing it wholesale.