That is clearly a demonstration of Turing completeness, but it's a demonstration of Turing completeness of HTML5 + CSS3 + copying from output back to input (the same as eg string rewriting systems) and that could stand to be stated more clearly.

Technically yes, but only when it's paired with the understanding that arithmetic plus copying the output back to input is Turing complete. HTML5+CSS is only supplying the arithmetic half of the equation.

I agree, I also think there is some weirdness going on with the 'we can presume it has infinite memory'. There seems to be many assumptions of different strength all waved with that same phrase, depending on the system in question.

Without that presumption how could anything ever be considered Turing complete? If you're simulating something with infinite memory you get to pretend you have infinite memory. Feasibility has already been suspended.

Edit: ok I get it. Still, GP observation is hardly at that level from my (limited) perspective. Not being able to advance by itself means it's not a standalone system - otoh external actor is obviously implicit for many other examples as well. I suppose they (and me) are taking the word machine too literally.

But "pens are Turing-complete" seems a bit, well, diluted.