Rewriting it with a guarantee of not introducing any errors is still beyond current LLM capabilities, and there might be a certain correlation between that capability and the capability of writing it from scratch.