> Boltzmann's fluctuations are pretty easily defeated by Feynman's argument: small fluctuations are exponentially more likely than large fluctuations, so if we make a new observation (e.g. opening a door) the fluctuation hypothesis predicts that we'll see random noise; yet we don't, we see more ordered structure. The hypothesis is refuted.

You've missed the point of Boltzmann brains. No one who studies this stuff thinks we're a Boltzmann brain. The question is whether we can rule out theories that predict Boltzmann brains as the numerically dominant observer with our observations. Indeed, many very popular theories of cosmology (e.g., eternal inflation) are in apparent conflict here.