That's a good point, you could indeed run the tests in your head. I think I can still save this though: this would prove that you only test for what you think about, and thus the existence or the absence of a test doesn't affect the development at all. Which could be countered again by property-based testing, but this one won't work in your head or one paper.
