I'm not necessarily opposed to something along these lines, except that it's basically already been tried: XSLT was essentially the inverse of this, where data endpoints were annotated with some static rendering descriptors. To me the XSLT approach seems cleaner, in that the URLs the browser is visiting are for the raw data itself, the URL scheme between machine and human is identical, and so on. XSLT also encouraged…
Yep. I used to love XSLT -- working in it finally forced me to understand recursion at, like, a deep cellular level -- but the whole time I was using it I knew it was a doomed language (because most people don't want to have to understand recursion at a deep cellular level just to be able to lay out a damn web page.)
But if you like it, that's cool.