When you can run the test but not design one that proves it works

I can run the test procedure step by step and record what happens. But designing a test that actually proves the design works, that would catch it if it were flawed, is beyond me right now. I go through the motions and still can't say for sure whether we really validated anything.