Your argument doesn't seem to apply to most code in natural sciences. The difference being that their theoretical models are, in most cases, only approximated b

Your argument doesn't seem to apply to most code in natural sciences. The difference being that their theoretical models are, in most cases, only approximated by code. Yet, in papers, they claim to prove or test their theoretical model by experiments on the approximation. Note that this is nothing terribly new. Sometimes experiments testing a hypothesis give false positives because something went wrong in the experiment. In that sense there is no real difference between a faulty thermometer and a bug in your code. Having written this, I think your argument in fact does apply after all. In natural sciences one would argue that if your buggy code confirms an invalid hypothesis, someone redoing your argument with the same code would not uncover the problem. Publishing your code invites people to use your faulty thermomenter. Of course I'm assuming that the published paper contains all necessary details to reproduce the experiment. In case of e.g. climate models or modelling the formation of galaxies that might be a problem because the code _is_ the experiment. Describing what code does is very hard, and it would be easier to just publish it in stead.