Dual-Process Scaffold Reasoning for Enhancing LLM Code Debugging

Po-Chung Hsieh, Chin-Po Chen, Jeng-Lin Li, Ming-Ching Chang · 2025

Read Dual-Process Scaffold Reasoning for Enhancing LLM Code Debugging by Po-Chung Hsieh, Chin-Po Chen, Jeng-Lin Li, Ming-Ching Chang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-08052.txt.

Description

Po-Chung Hsieh and co-authors frame code debugging by language models through the psychology of dual process thinking. Fast intuition and slow analysis play different roles in human problem solving, and model outputs versus intermediate steps can be read the same way. Prior work has not examined the slow reasoning stream closely enough for debugging work.

Their Scaffold Reasoning splits work into scaffold, analytic, and integration streams. Reference code construction merges with analysis of the buggy code before a final answer forms. On DebugBench the method reaches a high pass rate with fast average inference, beating other reasoning styles, and analysis shows how pathways vary by difficulty.

Catalog record

CREATORPo-Chung Hsieh, Chin-Po Chen, Jeng-Lin Li, Ming-Ching Chang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE22 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-08052

Browse the full library · About the Tower of Babel