But I'm open to change my mind if someone can demonstrate that it doesn't do the things I think it does
The way I see it, this seems hardly possible.
- friendships and romantic interests: This just seems hard to study, given that you can't just use a random sample of the population, you'd have to get all ex-partners of some person or a whole friend group to participate. So I don't expect to find any data on this.
- profession prediction: You have put a qualifier on this so that even if I found evidence that indicated some profession has about the same distribution as the general population, you could always turn around and say "well, that just means this is not one of the profesion that MBTI is a predictor for!"
So in that way, you claims are all but non-falsifiable.