Dude: A Dual-Detection Multi-Agent System for Paper-Code Discrepancy Detection

Forum topic · 小凯 · 2026-09-06

Summary

Dude is the first dual-detection multi-agent system designed to detect discrepancies between research papers and their associated code, addressing the limits of single-agent LLM approaches. The authors—Weijie Liu, Running Zhao, and Wenhao Yuan—identify a key problem: the granularity asymmetry between paper language and code language causes over-interpretation and over-reporting in multi-agent designs, inflating false positives. To mitigate this, Dude introduces a granularity-aligned negotiation mechanism and a two-stage salience-filtering process that prevents agents from erroneously reporting discrepancies. Experiments on a real-world paper-code discrepancy dataset show that Dude improves recall and accuracy by up to 22.8% and achieves F1 scores up to 18.7% higher than baseline methods. The work is motivated by the growing volume of research submissions exceeding manual review capacity, making automated reproducibility checking increasingly important. Details are available in arXiv paper 2509.00007 (published 2026-09-06).

Overview

Field: AI/ML Authors: Weijie Liu, Running Zhao, Wenhao Yuan Published: 2026-09-06 arXiv: 2509.00007

Abstract (translated)

LLM-empowered paper-code discrepancy detection has received growing attention since the scaling of research submissions exceeds manual review capability. However, the limited context capacity and one-sided discrepancy detection of existing single-agent LLM paradigms lead to inferior recall performance in detecting discrepancies. In this paper, the authors propose Dude, the first Dual-Detection Multi-Agent System for paper-code discrepancy detection.

They discover that the granularity asymmetry of paper-language and code-language introduces over-interpretation and over-reporting challenges in a multi-agent system design for discrepancy detection, resulting in increased false positives. To address this, Dude proposes:

Granularity-aligned negotiation: a mechanism that reconciles the different granularity of paper text and code to prevent misreporting.
Two-stage salience-filtering: a filtering process that suppresses insignificant or spurious discrepancy reports.

Results

Experiments on a real-world paper-code discrepancy dataset show that Dude significantly improves recall and accuracy by up to 22.8%, with F1 scores up to 18.7% higher than baseline methods.

---

*Auto-collected on 2026-09-06.*