metadata
task_categories:
- text-classification
language:
- en
tags:
- mathematics
- proof-verification
- arxiv
pretty_name: ArxivMathGradingBench
size_categories:
- n<1K
ArxivMathGradingBench
A benchmark of 35 arXiv mathematics papers that contained an error later corrected by the authors in a subsequent version, each annotated with the location of the error. Introduced in Pseudo-Formalization for Automatic Proof Verification to evaluate LLM proof verifiers on real research-level mathematics.
Downloading the papers
This dataset provides the paper metadata and error annotations. To download the paper PDFs, follow the instructions in the code repository:
https://github.com/Slim205/pseudo-formalization
Columns
| column | description |
|---|---|
arxiv_id |
arXiv identifier (e.g. 2501.04482) |
version |
paper version used by the benchmark (e.g. v1 / v2) |
url |
arXiv abstract URL |
v1_date |
date of the first arXiv version |
category |
arXiv subject class (e.g. math.FA) |
category_name_readable |
human-readable subject name |
comments |
the author's revision comment (often describing the fix) |
Location of Error |
ground-truth error location(s), e.g. Theorem 19 |
Liscence |
the paper's license as recorded on arXiv |
title_extracted_from_tex |
paper title |
Citation
@article{barkallah2026pseudo,
title={Pseudo-Formalization for Automatic Proof Verification},
author={Barkallah, Slim and Bailey, Luke and Wen, Kaiyue and Abouzaid, Mohammed and Ma, Tengyu},
journal={arXiv preprint arXiv:2605.20531},
year={2026}
}