"No. The "unless" clause is still incorrect. We can know a great deal about the fraction of people who think B, and it still cannot serve even as meta-evidence for or against B."

This can't be right. I have a hundred measuring devices. Ninety are broken and give a random answer with an unknown distribution, while ten give an answer that strongly correlates with the truth. Ninety say A and ten say B. If I examine a random meter that says B and find that it is broken, then surely that has to count as strong evidence against B.

This is probably an unnecessarily subtle point, of course; the overall thrust of the argument is of course correct.