You still have to account for the non-deterministic behavior of an LLM, when do you know you have exhausted its possible outcomes for any given piece of code?