This post actually mostly uses the subset of Python where nullability is checked. The point is not to introduce new LLM capabilities, but to understand more about how existing LLMs are reasoning about code.