@inproceedings{wang-etal-2026-open,
title = "Open Rubric System: Scaling Reinforcement Learning with Pairwise Adaptive Rubric",
author = "Wang, Wen and
Jia, Ruipeng and
Yang, Yunyi and
Wu, Yuxin and
Gai, Yongbo and
Tao, Siyuan and
Zhou, Mengyu and
Lin, Jianhe and
Jiang, Xiaoxi and
Jiang, Guanjun",
editor = "Atanasova, Pepa and
Haffari, Gholamreza and
Wang, Minghan",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-industry.89/",
pages = "1501--1530",
ISBN = "979-8-89176-460-6",
abstract = "We argue that robust alignment for non-verifiable tasks is fundamentally a $\textit{principle generalization}$ problem: reward should be guided by explicit, auditable principles rather than compressed into a learned scalar function. To operationalize this view, we present the $\textbf{Open Rubric System (OpenRS)}$ , a modular, rubrics-based $\textit{LLM-as-a-Judge}$ framework built around $\textbf{Pairwise Adaptive Meta-Rubrics (PAMR)}$ and lightweight $\textit{Pointwise Verifiable Rubrics}$ (PVRs). OpenRS is governed by an explicit $\textit{meta-rubric}${---}not a rubric itself, but a higher-level guideline that specifies $\textit{how}$ rubrics should be generated, which criteria to prioritize, and which failure modes are non-negotiable{---}and instantiates concrete $\textit{adaptive rubrics}$ on the fly by conditioning on the semantic differences between two candidate responses. It then performs criterion-wise $\textit{pairwise}$ comparisons and aggregates criterion-level preferences externally, avoiding pointwise weighted scalarization while improving discriminability in open-ended settings. Finally, we instantiate OpenRS as reward supervision in pairwise RL training. Empirically, OpenRS achieves state-of-the-art results on four reward-modeling benchmarks and yields stable gains when replacing scalar reward models in end-to-end RL."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2026-open">
<titleInfo>
<title>Open Rubric System: Scaling Reinforcement Learning with Pairwise Adaptive Rubric</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wen</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruipeng</namePart>
<namePart type="family">Jia</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yunyi</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuxin</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yongbo</namePart>
<namePart type="family">Gai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Siyuan</namePart>
<namePart type="family">Tao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mengyu</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jianhe</namePart>
<namePart type="family">Lin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaoxi</namePart>
<namePart type="family">Jiang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Guanjun</namePart>
<namePart type="family">Jiang</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: Industry Track</title>
</titleInfo>
<name type="personal">
<namePart type="given">Pepa</namePart>
<namePart type="family">Atanasova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gholamreza</namePart>
<namePart type="family">Haffari</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Minghan</namePart>
<namePart type="family">Wang</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">979-8-89176-460-6</identifier>
</relatedItem>
<abstract>We argue that robust alignment for non-verifiable tasks is fundamentally a principle generalization problem: reward should be guided by explicit, auditable principles rather than compressed into a learned scalar function. To operationalize this view, we present the Open Rubric System (OpenRS) , a modular, rubrics-based LLM-as-a-Judge framework built around Pairwise Adaptive Meta-Rubrics (PAMR) and lightweight Pointwise Verifiable Rubrics (PVRs). OpenRS is governed by an explicit meta-rubric—not a rubric itself, but a higher-level guideline that specifies how rubrics should be generated, which criteria to prioritize, and which failure modes are non-negotiable—and instantiates concrete adaptive rubrics on the fly by conditioning on the semantic differences between two candidate responses. It then performs criterion-wise pairwise comparisons and aggregates criterion-level preferences externally, avoiding pointwise weighted scalarization while improving discriminability in open-ended settings. Finally, we instantiate OpenRS as reward supervision in pairwise RL training. Empirically, OpenRS achieves state-of-the-art results on four reward-modeling benchmarks and yields stable gains when replacing scalar reward models in end-to-end RL.</abstract>
<identifier type="citekey">wang-etal-2026-open</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-industry.89/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>1501</start>
<end>1530</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Open Rubric System: Scaling Reinforcement Learning with Pairwise Adaptive Rubric
%A Wang, Wen
%A Jia, Ruipeng
%A Yang, Yunyi
%A Wu, Yuxin
%A Gai, Yongbo
%A Tao, Siyuan
%A Zhou, Mengyu
%A Lin, Jianhe
%A Jiang, Xiaoxi
%A Jiang, Guanjun
%Y Atanasova, Pepa
%Y Haffari, Gholamreza
%Y Wang, Minghan
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-460-6
%F wang-etal-2026-open
%X We argue that robust alignment for non-verifiable tasks is fundamentally a principle generalization problem: reward should be guided by explicit, auditable principles rather than compressed into a learned scalar function. To operationalize this view, we present the Open Rubric System (OpenRS) , a modular, rubrics-based LLM-as-a-Judge framework built around Pairwise Adaptive Meta-Rubrics (PAMR) and lightweight Pointwise Verifiable Rubrics (PVRs). OpenRS is governed by an explicit meta-rubric—not a rubric itself, but a higher-level guideline that specifies how rubrics should be generated, which criteria to prioritize, and which failure modes are non-negotiable—and instantiates concrete adaptive rubrics on the fly by conditioning on the semantic differences between two candidate responses. It then performs criterion-wise pairwise comparisons and aggregates criterion-level preferences externally, avoiding pointwise weighted scalarization while improving discriminability in open-ended settings. Finally, we instantiate OpenRS as reward supervision in pairwise RL training. Empirically, OpenRS achieves state-of-the-art results on four reward-modeling benchmarks and yields stable gains when replacing scalar reward models in end-to-end RL.
%U https://aclanthology.org/2026.emnlp-industry.89/
%P 1501-1530
Markdown (Informal)
[Open Rubric System: Scaling Reinforcement Learning with Pairwise Adaptive Rubric](https://aclanthology.org/2026.emnlp-industry.89/) (Wang et al., EMNLP 2026)
ACL
- Wen Wang, Ruipeng Jia, Yunyi Yang, Yuxin Wu, Yongbo Gai, Siyuan Tao, Mengyu Zhou, Jianhe Lin, Xiaoxi Jiang, and Guanjun Jiang. 2026. Open Rubric System: Scaling Reinforcement Learning with Pairwise Adaptive Rubric. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track, pages 1501–1530, Budapest, Hungary. Association for Computational Linguistics.