Consider the previously mentioned thesis: the author's brother was not interested in a goal like 'increasing, over the population of all females, the success of an approach' or 'learning how to pick up any girl', but rather something like 'how to get a reasonably attractive girl, period'.

Exactly, which is why talking about statistical models in this context is "academic", in the sense of "interesting to academics, but not particularly relevant to practitioners". Statistical models from experimental research can certainly inform practical approaches, but sometimes, one has to be "sorry for the Good Lord" in reverse: the theory may be utterly, totally, wrong, and yet still work.

If you want your rationality to protect something, let it protect results rather than "truth".

If the seduction techniques worked on only one girl in an entire bar (but infallibly), that'd be fine by them.

Well, as long as it was a girl they were interested in! ;-)

But by the same token, the reader of a self-help book is only interested in whether a technique fixes their problem, not a problem or all problems. The bigger picture of truth and generalizability is -- rightly and rationally -- not their concern.