I think there's a flaw in the model. "Assume that the amount of HTML currently loaded is finite but sufficient for all of the state to be properly rendered."

I think there's a flaw in the model. "Assume that the amount of HTML currently loaded is finite but sufficient for all of the state to be properly rendered." So consider <html> <div class="myContainer"> <div>1</div> <div>2</div> <div>3</div> . . . I've never met an HTML streaming solution that won't just stream this sequentially, which means no </div> for myContainer will ever be emitted (and therefore the HTML will never be well-formed, and the CSS will never have sufficient information to lay out). If the HTML were streamed such that <html><div class="myContainer"></div></html> were received, and then the interior of myContainer were streamed, that'd be a different story, but that doesn't exist. So I don't think this blog post's argument that CSS3 is Turing complete works for any real implementation.