One thing that is not clear with HoloJS: how do you aquire the head position and orientation on the JS side to be able to do gaze-centered selection (or, more accurately, nose-pointer selection, but people tend to call that gaze)? It seems to pass the view and projection matrices directly to the shader code. I haven't seen where it is available in a scriptable API.

It would be better to include an extension to the WebVR API to provide the environment geometry. There is a lot of work going into WebVR applications and frameworks. It'd be a shame if that work couldn't be leveraged for AR.

Also, I feel pretty strongly that automatically applying the stereo rendering is a mistake. This prevents being able to do any sort of changes between eyes, as is necessary for displaying stereo photos, photoshperes, or cubemaps. They can be a huge visual fidelity cheat on limited systems.