Don’t we already have multiple declarative formats that can be built over WebGL? What is the advantage of baking yet another one into WebGL itself? Because it maps more directly to the native WebGL data structures? Or because it has the WebGL object model baked into the declarative nodes?
In VRML 2.0, the debate over object/markup impedance mismatching was long. However, it made the point that interoperability via data transfer (bits on the wire, as Bray says) is very limited particularly where rendering and behavioral fidelity have to be very strong across platforms.
If I understand you, the main advantage of your proposal is the DOM scripting is simplified for those authors who want to instance and navigate directly in the WebGL objects. Is that right?
Interstingly, the response to every one of your points
below is the same. Why discuss this here? Why this scene graph rather than any
other? How can you use this to produce other scene rendering techniques? I'm
discussing a declarative form of WebGL. It has all the benefits and
characteristics of WebGL with the additional benefits of native traversal, DOM
and CSS integration and events.