nature-reviewer-skills: 450 Recurring Critique Patterns from 23,000 Nature Review Comments

nature-reviewer-skills is a multidisciplinary pre-submission review system spanning materials science, engineering, chemistry, and Earth-system science.

The project has so far analyzed 319 publicly available peer-review files from Nature journals, extracted more than 23,000 individual reviewer-comment units, and distilled over 450 recurring review patterns. Its purpose is practical: before submission, it helps authors identify the points most likely to draw scrutiny from reviewers at highly selective journals, so that a manuscript can be revised with those weaknesses in view.

Yizhe Yuan of Shanghai Jiao Tong University previously released the widely used nature-skills project, which includes a nature-reviewer skill. His approach is criterion-based review: it builds a structured general review framework around Nature's published editorial criteria, including originality, scientific importance, and relevance to an interdisciplinary readership.

I took a different route: corpus-distilled review. Rather than beginning with editorial criteria, I extract reviewer-comment units directly from public peer-review files, identify the concrete questions reviewers actually raise, and then abstract those observations into discipline-specific reviewer memory and evidence-chain stress tests. The point is not to imitate the tone of a reviewer. It is to approximate the recurring logic of real review: where claims tend to be challenged, what forms of evidence are considered weak, and which assumptions reviewers repeatedly ask authors to make explicit.

The project is available on GitHub: GeoGeekLab/nature-reviewer-skills. Testing, issues, and contributions are welcome.