@inproceedings{gan-etal-2026-scaling,
title = "Scaling Unverifiable Rewards: A Case Study on Visual Insights",
author = "Gan, Shuyu and
Mooney, James and
Hao, Pan and
Wang, Renxiang and
Hong, Mingyi and
Wang, Qianwen and
Kang, Dongyeop",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-acl.1724/",
doi = "10.18653/v1/2026.findings-acl.1724",
pages = "34537--34569",
ISBN = "979-8-89176-395-1",
abstract = "Large Language Model (LLM) agents can increasingly automate complex reasoning through Test-Time Scaling (TTS), an iterative refinement process guided by reward signals.However, many real-world tasks involve multi-stage pipelines whose final outcomes lack verifiable rewards or sufficient data to train robust reward models, making judge-based refinement prone to error accumulation across stages.We propose $\textbf{Selective TTS}$, a $\textit{process-based refinement}$ framework that scales inference across stages of a multi-agent pipeline, instead of repeatedly refining a single output over time as in prior work.By distributing compute across stages and pruning low-quality branches early using process-specific judgers, Selective TTS mitigates the judge drift and stabilizes refinement.Grounded in a data science workflow, we build an end-to-end multi-agent pipeline for generating visually insightful reports from a given dataset, and design a reliable LLM-based judge model that aligns with human experts (Kendall{'}s $\tau$=0.55) to evaluate them.Our proposed selective TTS then improves insight quality under a fixed compute budget, increasing mean scores from $\textbf{61.64}$ (baseline) to $\textbf{65.86}$ while reducing variance.We hope our findings serve as the first step toward scaling complex, open-ended tasks with unverifiable rewards like scientific discovery. Our code and generated reports are publicly available at https://minnesotanlp.github.io/insight-scaling-webpage."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="gan-etal-2026-scaling">
<titleInfo>
<title>Scaling Unverifiable Rewards: A Case Study on Visual Insights</title>
</titleInfo>
<name type="personal">
<namePart type="given">Shuyu</namePart>
<namePart type="family">Gan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">James</namePart>
<namePart type="family">Mooney</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pan</namePart>
<namePart type="family">Hao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Renxiang</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mingyi</namePart>
<namePart type="family">Hong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qianwen</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dongyeop</namePart>
<namePart type="family">Kang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Liakata</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Viviane</namePart>
<namePart type="given">P</namePart>
<namePart type="family">Moreira</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiajun</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Jurgens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-395-1</identifier>
</relatedItem>
<abstract>Large Language Model (LLM) agents can increasingly automate complex reasoning through Test-Time Scaling (TTS), an iterative refinement process guided by reward signals.However, many real-world tasks involve multi-stage pipelines whose final outcomes lack verifiable rewards or sufficient data to train robust reward models, making judge-based refinement prone to error accumulation across stages.We propose Selective TTS, a process-based refinement framework that scales inference across stages of a multi-agent pipeline, instead of repeatedly refining a single output over time as in prior work.By distributing compute across stages and pruning low-quality branches early using process-specific judgers, Selective TTS mitigates the judge drift and stabilizes refinement.Grounded in a data science workflow, we build an end-to-end multi-agent pipeline for generating visually insightful reports from a given dataset, and design a reliable LLM-based judge model that aligns with human experts (Kendall’s τ=0.55) to evaluate them.Our proposed selective TTS then improves insight quality under a fixed compute budget, increasing mean scores from 61.64 (baseline) to 65.86 while reducing variance.We hope our findings serve as the first step toward scaling complex, open-ended tasks with unverifiable rewards like scientific discovery. Our code and generated reports are publicly available at https://minnesotanlp.github.io/insight-scaling-webpage.</abstract>
<identifier type="citekey">gan-etal-2026-scaling</identifier>
<identifier type="doi">10.18653/v1/2026.findings-acl.1724</identifier>
<location>
<url>https://aclanthology.org/2026.findings-acl.1724/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>34537</start>
<end>34569</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Scaling Unverifiable Rewards: A Case Study on Visual Insights
%A Gan, Shuyu
%A Mooney, James
%A Hao, Pan
%A Wang, Renxiang
%A Hong, Mingyi
%A Wang, Qianwen
%A Kang, Dongyeop
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%S Findings of the Association for Computational Linguistics: ACL 2026
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-395-1
%F gan-etal-2026-scaling
%X Large Language Model (LLM) agents can increasingly automate complex reasoning through Test-Time Scaling (TTS), an iterative refinement process guided by reward signals.However, many real-world tasks involve multi-stage pipelines whose final outcomes lack verifiable rewards or sufficient data to train robust reward models, making judge-based refinement prone to error accumulation across stages.We propose Selective TTS, a process-based refinement framework that scales inference across stages of a multi-agent pipeline, instead of repeatedly refining a single output over time as in prior work.By distributing compute across stages and pruning low-quality branches early using process-specific judgers, Selective TTS mitigates the judge drift and stabilizes refinement.Grounded in a data science workflow, we build an end-to-end multi-agent pipeline for generating visually insightful reports from a given dataset, and design a reliable LLM-based judge model that aligns with human experts (Kendall’s τ=0.55) to evaluate them.Our proposed selective TTS then improves insight quality under a fixed compute budget, increasing mean scores from 61.64 (baseline) to 65.86 while reducing variance.We hope our findings serve as the first step toward scaling complex, open-ended tasks with unverifiable rewards like scientific discovery. Our code and generated reports are publicly available at https://minnesotanlp.github.io/insight-scaling-webpage.
%R 10.18653/v1/2026.findings-acl.1724
%U https://aclanthology.org/2026.findings-acl.1724/
%U https://doi.org/10.18653/v1/2026.findings-acl.1724
%P 34537-34569
Markdown (Informal)
[Scaling Unverifiable Rewards: A Case Study on Visual Insights](https://aclanthology.org/2026.findings-acl.1724/) (Gan et al., Findings 2026)
ACL
- Shuyu Gan, James Mooney, Pan Hao, Renxiang Wang, Mingyi Hong, Qianwen Wang, and Dongyeop Kang. 2026. Scaling Unverifiable Rewards: A Case Study on Visual Insights. In Findings of the Association for Computational Linguistics: ACL 2026, pages 34537–34569, San Diego, California, United States. Association for Computational Linguistics.