But this doesn't address the explicit, simple, standardisation of app/interaction structure. If it is defined in a full procedural language, it is obscure and un-manipulable.

The application structure should be moved as much as possible into that JSON, XML, or similar, and standardised. So the browser has a fairly simple and very general app-engine, and everything specific is in a simple, open format. In a sense: turn as much as possible into 'content', including interaction and app intelligence. Then, like other 'content' like movies or music, it is dumb, but infinitely accessible and manipulable.