I don't care if the code works because it was formed because of temporal understanding or if it works because an LLM predicted enough tokens correctly, I care that it works.