Miniscule progress on an alternative notebook renderer that hacks the observable runtime to capture all the values. Behold, "foo" is rendered in two places! One of those places is a giant cells containing every other cell in the notebook.... that is what visualizer will be. A cell that renders the whole. So then the outer HTML frame will not actually provide any DOM nodes for the runtime to render into externally, the engine will start and visualizer will dynamically create DOM nodes to render the rest from the inside. An inversion, much more hackable that way.