<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="shihab-etal-2026-est">
    <titleInfo>
        <title>EST-PRM: Stress-Testing Process Reward Models Before They Become Load-Bearing</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Ibne</namePart>
        <namePart type="given">Farabi</namePart>
        <namePart type="family">Shihab</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Fariya</namePart>
        <namePart type="family">Afrin</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Sanjeda</namePart>
        <namePart type="family">Akter</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Anuj</namePart>
        <namePart type="family">Sharma</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-10</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Hung-Yi</namePart>
            <namePart type="family">Lee</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Sunipa</namePart>
            <namePart type="family">Dev</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jessy</namePart>
            <namePart type="family">Li</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Desmond</namePart>
            <namePart type="family">Elliott</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Budapest, Hungary</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-459-0</identifier>
    </relatedItem>
    <abstract>Process reward models (PRMs) are widely used in language-model training with dense, step-level supervision. They assume that PRM scores remain stable proxies for step correctness under label-preserving transformations that alter reasoning structure while preserving final answers. However, this assumption remains insufficiently validated. Such transformations may alter the relationship between PRM scores and correctness, producing different failure modes across models. To address this gap, we introduce EST-PRM, a stress-testing framework for dense process rewards. It evaluates three transformations: (1) step inflation, (2) dependency-aware step reordering, and (3) confidence markers. We further define a vulnerability decomposition that separates reward inflation from loss of correctness sensitivity. We evaluate five PRM-style models on 4,687 reasoning chains from MATH-500, GSM8K, and PRMBench. Results reveal substantial differences in vulnerability patterns across models. Math-Shepherd exhibits the strongest sensitivity to position-based perturbations, with a Pearson correlation drop of 0.139 \pm 0.031 and a 29.7 \pm 4.2% score inflation rate, while Qwen2.5-Math-PRM is most affected by step inflation, reaching a 47.6 \pm 4.3% inflation rate. Confidence-based perturbations further distort reward calibration, revealing inconsistencies in correctness estimation. Finally, we evaluate three mitigation strategies and identify trade-offs between robustness coverage and false-positive rates.</abstract>
    <identifier type="citekey">shihab-etal-2026-est</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.778/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>18085</start>
            <end>18108</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T EST-PRM: Stress-Testing Process Reward Models Before They Become Load-Bearing
%A Shihab, Ibne Farabi
%A Afrin, Fariya
%A Akter, Sanjeda
%A Sharma, Anuj
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F shihab-etal-2026-est
%X Process reward models (PRMs) are widely used in language-model training with dense, step-level supervision. They assume that PRM scores remain stable proxies for step correctness under label-preserving transformations that alter reasoning structure while preserving final answers. However, this assumption remains insufficiently validated. Such transformations may alter the relationship between PRM scores and correctness, producing different failure modes across models. To address this gap, we introduce EST-PRM, a stress-testing framework for dense process rewards. It evaluates three transformations: (1) step inflation, (2) dependency-aware step reordering, and (3) confidence markers. We further define a vulnerability decomposition that separates reward inflation from loss of correctness sensitivity. We evaluate five PRM-style models on 4,687 reasoning chains from MATH-500, GSM8K, and PRMBench. Results reveal substantial differences in vulnerability patterns across models. Math-Shepherd exhibits the strongest sensitivity to position-based perturbations, with a Pearson correlation drop of 0.139 \pm 0.031 and a 29.7 \pm 4.2% score inflation rate, while Qwen2.5-Math-PRM is most affected by step inflation, reaching a 47.6 \pm 4.3% inflation rate. Confidence-based perturbations further distort reward calibration, revealing inconsistencies in correctness estimation. Finally, we evaluate three mitigation strategies and identify trade-offs between robustness coverage and false-positive rates.
%U https://aclanthology.org/2026.emnlp-main.778/
%P 18085-18108

Markdown (Informal)

[EST-PRM: Stress-Testing Process Reward Models Before They Become Load-Bearing](https://aclanthology.org/2026.emnlp-main.778/) (Shihab et al., EMNLP 2026)

• EST-PRM: Stress-Testing Process Reward Models Before They Become Load-Bearing (Shihab et al., EMNLP 2026)

ACL

• Ibne Farabi Shihab, Fariya Afrin, Sanjeda Akter, and Anuj Sharma. 2026. EST-PRM: Stress-Testing Process Reward Models Before They Become Load-Bearing. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 18085–18108, Budapest, Hungary. Association for Computational Linguistics.