> But because what is codified is codified without systematic knowledge of what works or how it works, the preservation of benefit is essentially random with weak selective pressure acting in the aggregate of beliefs

But you agree this must be much better than random? Evolutionary pressure on species is also rather weak: unfit specimen survive and fit specimen die due to chance all the time. But look where it got us when averaged over long periods of time.

I don't buy the "systematic knowledge of what works or how it works" part. That's what NLP scientists used to say about neural nets while building monstrous systems based on "systematic knowledge of grammar". You definitely don't have to understand "how it works" to be able to make good predictions.

> But you agree this must be much better than random?

Well, no, without a definition of what domain it is supposed to be better in, and what the actual alternative it is being compared to more concretely than "random" (irreligious humans don't behave randomly, and, in fact, even without religion preserve traditions, some of which are useful), and probably some argument to make the case, no, I'm not going to agree with that.

> You definitely don't have to understand "how it works" to be able to make good predictions.

You have to actually make predictions to make predictions, certainly. And religion is manifestly very bad at making predictions where it does make them, and the things you are talking about are very much not predictions, they are memes in the original sense.