Source record

Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2609.22111.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

ReAgent is an automated auditing framework that checks whether research documents produced by LLM agents are consistent with their code repositories. It combines static analysis of claimed methodologies with dynamic execution of experiments, producing structured audit reports. Evaluated on a benchmark of agent-generated paper-repository pairs, it outperforms static and reproduction-based baselines.

As autonomous research agents become common, this offers a concrete method to catch mismatches between claimed results and actual code execution — filling a gap left by text-only peer review.

Cited by

No citations on record.