I will also think again about having input output nodes on the canvas as it does solve some issues. I however prefer the visual metaphor of having them floating at the edge of the screen.
I am considering making it zoomable / pan able but in someways I like the restriction on complexity. I want to encourage people to create "sub-graphs" instead of add more nodes.
I know its not far from perfect yet but EtcH is designed from the ground up to be a "language" rather than a way to wire together pre-defined nodes. Im glad its those innovative parts that you like :)