Tag
This paper introduces Code-Guided Reasoning (CGR), an evaluation protocol for measuring how executable reasoning scaffolds improve small language model performance on multiple-choice question answering tasks, showing a significant accuracy improvement over direct answering.