Non-consensus and right is the best framework for making investments, since you need non-consensus in order to get a cheap price. But for merely estimating the probability that something will succeed, consensus (whether positive or negative) is probably slightly predictive in that regard, as you point out w.r.t. survivor bias.

I do put a lot of stock into Paul's essay, though. If someone credible and highly intelligent has some non-consensus opinion, it's a good idea to suspend judgement and really listen to their reasoning.