I agree with your argument, but can you predict performance of an algorithm from big O complexity? (I assume you mean) Perhaps a few runs of data points could

I agree with your argument, but can you predict performance of an algorithm from big O complexity? (I assume you mean) Perhaps a few runs of data points could enable curve fitting as it converges. I feel like the "physics" of software is immature, closer to math than to practice, than for engineering - or something like that. So I'm also agreeing with the gist of the other replier, that software isn't yet mature enough for precise estimates. We still flounder on what's needed, and stumble around what kind of a thing to build. Precise specification is the exception.