Back to AI News
arxiv

Benchmark Audits LLM Peer Reviewers on Real Scientific Papers

Researchers introduce a benchmark that injects objective, localizable errors of four types at three severity levels into real papers and measures reviewer detection by error type and severity, along with false-positive rates on clean copies. They evaluate twelve reviewer systems, including six single-LLM reviewers, two Claude Code configurations, three public reviewers, and a multi-agent reviewer called Skepthical, across astrophysics and machine-learning corpora of 15 recent arXiv preprints each with 357 total injected errors. Model choice had the largest impact on detection, with three frontier LLMs differing by about 50 percentage points under the same prompt. Tooling and scaffolding helped when they supplied capabilities the model lacked, such as enforced citation retrieval improving citation-error detection on astrophysics but not on ML. False-positive rates were below 10% for most systems but varied with prompt, model, and scaffolding.

AI Brief

It provides concrete evidence on the reliability, failure modes, and tooling effects of LLM-based peer review using a reproducible benchmark.

arXiv:2611.17052v1 Announce Type: cross Abstract: LLM-based peer review is now used at production scale, but its evaluation remains limited. We introduce a benchmark that injects objective, localisable errors of four types (general issues, mathematics, numerics, citations) at three severity levels into real papers and measures each reviewer's detection rate by error type and severity, together with its false-positive rate on clean copies. We also present Skepthical, a multi-agent reviewer that verifies citation-backed claims by retrieving the cited papers, mathematics with computer algebra, and numerical claims with executed code. We apply the benchmark to two corpora of 15 recent arXiv preprints each, one in astrophysics and one in machine learning (ML), injecting a total of 357 errors, and evaluate twelve reviewer systems: six single-LLM reviewers, two Claude Code configurations, three publicly available reviewers, and Skepthical. The choice of model has the largest impact on detection rate: given the same prompt, three frontier LLMs differ by about 50 percentage points in total detection on both corpora. Tools and scaffolding help when they contribute what the model cannot supply on its own. In particular, on astrophysics, Claude Code improves on a single Claude call in mathematics and numerics detection, and enforced citation retrieval gives Skepthical the highest citation error detection rate of all systems by a large margin. This citation lead, however, is not observed on ML, where the models appear to know the literature better and Skepthical's citation detection rate is statistically indistinguishable from that of the best single LLM. False-positive rates are below 10% for most systems, but vary with prompt, model, and scaffolding, with mathematics flags the main source of false positives. The evaluation pipeline, the injected errors, and the adjudication logs are released as a benchmark.
Be the first to rate this article.

Comments & Questions (0)

Captcha
Please be respectful — let's keep the conversation friendly.

No comments yet

Be the first to comment!