@inproceedings{hong-etal-2026-assay,
title = "{ASSAY}: Reconciling Stability and Validity in Reference-Free Scientific {LLM} Judges",
author = "Hong, Soyeon and
Kim, Heegyu and
Hong, Ji Hoon and
Hwang, Jindong and
Ha, Juhyeon and
Cho, Sung Beom and
Cho, Hyunsouk",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1625/",
pages = "34580--34600",
ISBN = "979-8-89176-462-0",
abstract = "LLM judges have emerged as a key approach to evaluating open-ended scientific outputs that resist automated verification, yet their faithfulness in reference-free settings remains underexplored, particularly for fields such as novel materials synthesis where no ground-truth reference exists. Such judges produce false positives that accept scientifically invalid recipes and false negatives that penalize semantically equivalent ones in unfamiliar notation. We introduce ASSAY, a probe suite built on domain-grounded augmentations of ground-truth synthesis recipes, and reveal that these two failure modes form a systematic Stability{--}Validity trade-off across thirteen judge models. To resolve this trade-off, we train SynthJudge-8B via DPO on contrastive pairs derived from the same augmentations, improving judgment accuracy from 51.4{\%} to 85.1{\%} while maintaining agreement with domain experts. We release our probe suite and trained model to support faithful automated evaluation in scientific fields beyond materials synthesis at https://github.com/SoyeonHH/ASSAY."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="hong-etal-2026-assay">
<titleInfo>
<title>ASSAY: Reconciling Stability and Validity in Reference-Free Scientific LLM Judges</title>
</titleInfo>
<name type="personal">
<namePart type="given">Soyeon</namePart>
<namePart type="family">Hong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Heegyu</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ji</namePart>
<namePart type="given">Hoon</namePart>
<namePart type="family">Hong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jindong</namePart>
<namePart type="family">Hwang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Juhyeon</namePart>
<namePart type="family">Ha</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sung</namePart>
<namePart type="given">Beom</namePart>
<namePart type="family">Cho</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hyunsouk</namePart>
<namePart type="family">Cho</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>LLM judges have emerged as a key approach to evaluating open-ended scientific outputs that resist automated verification, yet their faithfulness in reference-free settings remains underexplored, particularly for fields such as novel materials synthesis where no ground-truth reference exists. Such judges produce false positives that accept scientifically invalid recipes and false negatives that penalize semantically equivalent ones in unfamiliar notation. We introduce ASSAY, a probe suite built on domain-grounded augmentations of ground-truth synthesis recipes, and reveal that these two failure modes form a systematic Stability–Validity trade-off across thirteen judge models. To resolve this trade-off, we train SynthJudge-8B via DPO on contrastive pairs derived from the same augmentations, improving judgment accuracy from 51.4% to 85.1% while maintaining agreement with domain experts. We release our probe suite and trained model to support faithful automated evaluation in scientific fields beyond materials synthesis at https://github.com/SoyeonHH/ASSAY.</abstract>
<identifier type="citekey">hong-etal-2026-assay</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1625/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>34580</start>
<end>34600</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T ASSAY: Reconciling Stability and Validity in Reference-Free Scientific LLM Judges
%A Hong, Soyeon
%A Kim, Heegyu
%A Hong, Ji Hoon
%A Hwang, Jindong
%A Ha, Juhyeon
%A Cho, Sung Beom
%A Cho, Hyunsouk
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F hong-etal-2026-assay
%X LLM judges have emerged as a key approach to evaluating open-ended scientific outputs that resist automated verification, yet their faithfulness in reference-free settings remains underexplored, particularly for fields such as novel materials synthesis where no ground-truth reference exists. Such judges produce false positives that accept scientifically invalid recipes and false negatives that penalize semantically equivalent ones in unfamiliar notation. We introduce ASSAY, a probe suite built on domain-grounded augmentations of ground-truth synthesis recipes, and reveal that these two failure modes form a systematic Stability–Validity trade-off across thirteen judge models. To resolve this trade-off, we train SynthJudge-8B via DPO on contrastive pairs derived from the same augmentations, improving judgment accuracy from 51.4% to 85.1% while maintaining agreement with domain experts. We release our probe suite and trained model to support faithful automated evaluation in scientific fields beyond materials synthesis at https://github.com/SoyeonHH/ASSAY.
%U https://aclanthology.org/2026.findings-emnlp.1625/
%P 34580-34600
Markdown (Informal)
[ASSAY: Reconciling Stability and Validity in Reference-Free Scientific LLM Judges](https://aclanthology.org/2026.findings-emnlp.1625/) (Hong et al., Findings 2026)
ACL