Earlier quoted context omitted.
Just a few remarks, notwithstanding your/my opinions. Physics has not "one of the better reproducibility rates" but 100%; per the Scientific Method any valid theory must: - be reproducible (verifiable empirically) - have predictive power - not negate existing theories This constitutes 'acceptable proof'. It yields a 'compound effect' to theoretical research: for instance, Newton's Laws were never wrong, just narrower…
I think you might have misunderstood? Academic physics publications do not have perfect reproduction rates. Especially theoretical physics. The criteria for a valid theory doesn’t prevent human mistakes. Everything you said applies to social sciences too. Per the scientific method any valid theory must be reproducible and predictive. It’s an unfortunate fact of life that the academic research and publication process…
It is my understanding that margins of error in reproducible rates are on-par with our limits in engineering capabilities to 'test' experimentally. The validation becomes statistic because the real world is too chaotic/complex/whatever external cause, but the sum of evidence must tend to show (mean, SD, etc) that the theory applies to the real world, and predicts "a range of possible within some margin of error", which in practice results from our imprecisions. E.g. how we continuously prove Einstein's GR.
Quantum uncertainty is a whole other paradigm which I don't think can be discussed in the same terms; a lot of our 'logic' does not apply and thus I'd rather refrain from using classical frameworks of thinking to even characterize it.
But the fact stands that you should always be able to verify experimentally a valid theory; our failure in practice, sometimes, has nothing to do with theory.
Social sciences look at category of problems considered "hard" by any scientific metric because of their complexity. It's like meteorology and indeed astrophysics, we lack the processing power (both our brains and computers, for now) to go all-in on theory. I was however surprised, in my studies, that the more down-to-earth approach of anthropology (naive, dare I say seeking first principles) seemed to yield observation 'closer' to theory than its modern sister sociology, whose subtopic —the modern world, as opposed to very specific cultures for a quite narrow spacetime— is vastly too multi-factorial to even begin to modelize, let alone theorize.