(Didn't realize this site doesn't email reply notifications, thus the delayed response.)

What I'm saying is that someone who answers "algorithms" is clearly not taking that view of substrate-independence, but they could hypothesize that only some side-effects matter. A MOSFET-brain-simulation and a desert-rocks-brain-simulation could share computational properties beyond input-output, even though the side-effects are clearly not identical.

(Not saying that I endorse that hypothesis, just that it's not the same as the "side effects don't matter" version.)