A method for extracting reasoning logic from LLM traces and converting it into reusable, CPU-executable programs that solve new problem instances efficiently.