Still not sure I follow well. The way you present it, it seems mutation testing is used to perfect your unit tests? Kind of shrink them to the minimum viable test that will still reliably pass or fail when presented with the right / wrong input?
I’m also seeing a lot of intersection with property tests, or maybe I’m imagining it. When you do those you directly attack the expected invariants of your function and then you are presented with the exact offending input – so you can change your actual code (not the test).
Haven’t thought about the aspect of getting a slightly reduced test code that still passes and that might mean that the original test is more heavyweight – that one sounds pretty interesting. It’s just that I’m skeptical; how likely is it that you will get that randomly improved test in reality? Did you have such occurrences while using mutation testing?