Guess you could intercept all the traffic and that way know which parts are used. I guess you could build it up the same way as the actual schema and then compare it agains the used one. That's for schema coverage.

We log (a bit indirectly) what fields are used in queries into our system; we then have a tool that can identify which fields are used and which are not. We'll eventually use that to deprecate and finally remove unused fields. That's a nice ability to have, and not something you can effectively do in a REST world where you have no idea what data is actually needed by clients because one-size-ill-fits-all.