Saved in:
Bibliographic Details
Main Authors: Mahdavi, Hamed, Mahdavinia, Pouria, Malek, Samira, Mohammadipour, Pegah, Hashemi, Alireza, Daliri, Majid, Farhadi, Alireza, Khasahmadi, Amir, Mireshghallah, Niloofar, Honavar, Vasant
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.09021
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912640461176832
author Mahdavi, Hamed
Mahdavinia, Pouria
Malek, Samira
Mohammadipour, Pegah
Hashemi, Alireza
Daliri, Majid
Farhadi, Alireza
Khasahmadi, Amir
Mireshghallah, Niloofar
Honavar, Vasant
author_facet Mahdavi, Hamed
Mahdavinia, Pouria
Malek, Samira
Mohammadipour, Pegah
Hashemi, Alireza
Daliri, Majid
Farhadi, Alireza
Khasahmadi, Amir
Mireshghallah, Niloofar
Honavar, Vasant
contents State-of-the-art (SOTA) LLMs have progressed from struggling on proof-based Olympiad problems to solving most of the IMO 2025 problems, with leading systems reportedly handling 5 of 6 problems. Given this progress, we assess how well these models can grade proofs: detecting errors, judging their severity, and assigning fair scores beyond binary correctness. We study proof-analysis capabilities using a corpus of 90 Gemini 2.5 Pro-generated solutions that we grade on a 1-4 scale with detailed error annotations, and on MathArena solution sets for IMO/USAMO 2025 scored on a 0-7 scale. Our analysis shows that models can reliably flag incorrect (including subtly incorrect) solutions but exhibit calibration gaps in how partial credit is assigned. To address this, we introduce agentic workflows that extract and analyze reference solutions and automatically derive problem-specific rubrics for a multi-step grading process. We instantiate and compare different design choices for the grading workflows, and evaluate their trade-offs. Across our annotated corpus and MathArena, our proposed workflows achieve higher agreement with human grades and more consistent handling of partial credit across metrics. We release all code, data, and prompts/logs to facilitate future research.
format Preprint
id arxiv_https___arxiv_org_abs_2510_09021
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle RefGrader: Automated Grading of Mathematical Competition Proofs using Agentic Workflows
Mahdavi, Hamed
Mahdavinia, Pouria
Malek, Samira
Mohammadipour, Pegah
Hashemi, Alireza
Daliri, Majid
Farhadi, Alireza
Khasahmadi, Amir
Mireshghallah, Niloofar
Honavar, Vasant
Artificial Intelligence
Machine Learning
State-of-the-art (SOTA) LLMs have progressed from struggling on proof-based Olympiad problems to solving most of the IMO 2025 problems, with leading systems reportedly handling 5 of 6 problems. Given this progress, we assess how well these models can grade proofs: detecting errors, judging their severity, and assigning fair scores beyond binary correctness. We study proof-analysis capabilities using a corpus of 90 Gemini 2.5 Pro-generated solutions that we grade on a 1-4 scale with detailed error annotations, and on MathArena solution sets for IMO/USAMO 2025 scored on a 0-7 scale. Our analysis shows that models can reliably flag incorrect (including subtly incorrect) solutions but exhibit calibration gaps in how partial credit is assigned. To address this, we introduce agentic workflows that extract and analyze reference solutions and automatically derive problem-specific rubrics for a multi-step grading process. We instantiate and compare different design choices for the grading workflows, and evaluate their trade-offs. Across our annotated corpus and MathArena, our proposed workflows achieve higher agreement with human grades and more consistent handling of partial credit across metrics. We release all code, data, and prompts/logs to facilitate future research.
title RefGrader: Automated Grading of Mathematical Competition Proofs using Agentic Workflows
topic Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2510.09021