Your example of experiment design in a chemical process plant reminded me of a similar approach involving early Macintosh operating systems late last century. There were dozens of extensions one could add to the system to process video, change window appearance, support new devices and so on. Sometimes, they didn’t play well together. The system would crash or hang during startup. So, one company wrote an extension to help people figure out which extension was causing the problem. It did a logarithmic search, first choosing a subset of roughly half the extensions. If the startup succeeded, it tried the other bunch. Then it would try smaller subsets. Eventually, it would find the extension, or it would try a sophisticated search to find which pair of extensions were in conflict.

It was slow since it involved repeatedly booting one’s Mac, and since failure to startup might require intervention, it could be tedious. Still, the program managed a comprehensive search for the problem without checklists and the like. Unlike thrashing around manually enabling and disabling extensions, this extension had an algorithm which could find the problem in a minimal number or startups.