It is this: “When you really get stuck, and something isn’t working that once worked, try to find something similar to it that still works. Then gradually change the working system, one thing at a time, to be more like the broken system.”

Or, in short, compare the working system to the broken system.

This is the fundamental idea behind, for example, git-bisect, but the concept is more general, and extends to hardware, auto mechanics, anything. It’s really a part of science. That is, in science, people make hypotheses, do experiments to try to corroborate, and to disprove those hypotheses, and peers try to duplicate the experimental results. Just because the answers to questions aren’t straightforward doesn’t make something art and not science. The bleeding edge of science consists of pretty much nothing but questions to which the answers are anything but straightforward.