That's a good way of putting it!
Heh, it's funny you mention transform matrices... that's exactly the chunk of code I'm working on right now. Did I even mention that in my post? Freaky!
So it's a bit tough to explain since I'm still working on it - but essentially what I have is kindof like this:
State -> Worker -> Tree -> Render
Where State itself contains a reference to both the last State and the last Tree.
The reason being that Tree generation walks it according to the specific hierarchy and sets the transforms. State needs to know about this to accurately detect touch and collision.
The part I'm working on right now is dealing with a camera... I had the above basically working but with a hardcoded camera. Now I'd like the camera to itself just be a "special" node in the tree... this would theoretically allow some cool things - like multiple cameras in the same scene, just by declaratively specifying the State/tree nodes.
It's a work-in-progress still, sorry there's no code to really reference, though I'm glad I'm holding off - I completely rewrote everything since my previous post on the details