Not the best description doesn't cut it, it's just false description. I can easily write a system with 100% test coverage, bugged as hell and containing so many weird abstraction tricks that it takes a significant effort to figure out even what is happening in the simplest scenario. And yes, part of that hundred percent coverage is going to be llm style: test that something does what you already know it does