@inproceedings{li-etal-2026-vict,
title = "{VICT}: Verifier-Instrumented Credit Tracing for Reinforcement Learning of Long-Horizon {LLM} Agents",
author = "Li, Pengcheng and
Zhang, Zhengyang and
Zhang, Dongxu and
Huang, Sui and
Ma, Shaohua",
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.2395/",
pages = "51468--51486",
ISBN = "979-8-89176-462-0",
abstract = "Fine-grained credit assignment is a central challenge in reinforcement learning for long-horizon LLM agents. Standard objectives often train from programmatically verifiable terminal rewards by broadcasting each sparse outcome to every action in a trajectory. Existing methods typically seek finer credit from the rollout side, constructing auxiliary trajectory signals or additional comparisons to estimate action importance. Although useful, these approaches still treat the verifier that judged success as a scalar reward, discarding its internal task structure. Our key insight is that many verifiable tasks already encode the relevant checks inside their terminal verifier. We propose VICT (Verifier-Instrumented Credit Tracing), a training-time interface that exposes executable or evidence-backed atoms and traces them back to actions through dependency-valid proof edges. VICT redistributes group-relative advantage only along those edges, shifting credit assignment from rollout-side inference to verifier-side tracing. It preserves the original terminal reward, abstains when evidence is incomplete or ambiguous, and changes only the training-time advantage tensor, requiring no learned critic, process labels, branch rollouts, or inference-time verifier access. Under matched rollout budgets on ALFWorld and WebShop, VICT improves over outcome-only and rollout-side credit baselines; ablations rule out dense atom rewards, final-commit credit, temporal proximity, and sparsity as sufficient explanations."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="li-etal-2026-vict">
<titleInfo>
<title>VICT: Verifier-Instrumented Credit Tracing for Reinforcement Learning of Long-Horizon LLM Agents</title>
</titleInfo>
<name type="personal">
<namePart type="given">Pengcheng</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhengyang</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dongxu</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sui</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shaohua</namePart>
<namePart type="family">Ma</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>Fine-grained credit assignment is a central challenge in reinforcement learning for long-horizon LLM agents. Standard objectives often train from programmatically verifiable terminal rewards by broadcasting each sparse outcome to every action in a trajectory. Existing methods typically seek finer credit from the rollout side, constructing auxiliary trajectory signals or additional comparisons to estimate action importance. Although useful, these approaches still treat the verifier that judged success as a scalar reward, discarding its internal task structure. Our key insight is that many verifiable tasks already encode the relevant checks inside their terminal verifier. We propose VICT (Verifier-Instrumented Credit Tracing), a training-time interface that exposes executable or evidence-backed atoms and traces them back to actions through dependency-valid proof edges. VICT redistributes group-relative advantage only along those edges, shifting credit assignment from rollout-side inference to verifier-side tracing. It preserves the original terminal reward, abstains when evidence is incomplete or ambiguous, and changes only the training-time advantage tensor, requiring no learned critic, process labels, branch rollouts, or inference-time verifier access. Under matched rollout budgets on ALFWorld and WebShop, VICT improves over outcome-only and rollout-side credit baselines; ablations rule out dense atom rewards, final-commit credit, temporal proximity, and sparsity as sufficient explanations.</abstract>
<identifier type="citekey">li-etal-2026-vict</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2395/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>51468</start>
<end>51486</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T VICT: Verifier-Instrumented Credit Tracing for Reinforcement Learning of Long-Horizon LLM Agents
%A Li, Pengcheng
%A Zhang, Zhengyang
%A Zhang, Dongxu
%A Huang, Sui
%A Ma, Shaohua
%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 li-etal-2026-vict
%X Fine-grained credit assignment is a central challenge in reinforcement learning for long-horizon LLM agents. Standard objectives often train from programmatically verifiable terminal rewards by broadcasting each sparse outcome to every action in a trajectory. Existing methods typically seek finer credit from the rollout side, constructing auxiliary trajectory signals or additional comparisons to estimate action importance. Although useful, these approaches still treat the verifier that judged success as a scalar reward, discarding its internal task structure. Our key insight is that many verifiable tasks already encode the relevant checks inside their terminal verifier. We propose VICT (Verifier-Instrumented Credit Tracing), a training-time interface that exposes executable or evidence-backed atoms and traces them back to actions through dependency-valid proof edges. VICT redistributes group-relative advantage only along those edges, shifting credit assignment from rollout-side inference to verifier-side tracing. It preserves the original terminal reward, abstains when evidence is incomplete or ambiguous, and changes only the training-time advantage tensor, requiring no learned critic, process labels, branch rollouts, or inference-time verifier access. Under matched rollout budgets on ALFWorld and WebShop, VICT improves over outcome-only and rollout-side credit baselines; ablations rule out dense atom rewards, final-commit credit, temporal proximity, and sparsity as sufficient explanations.
%U https://aclanthology.org/2026.findings-emnlp.2395/
%P 51468-51486
Markdown (Informal)
[VICT: Verifier-Instrumented Credit Tracing for Reinforcement Learning of Long-Horizon LLM Agents](https://aclanthology.org/2026.findings-emnlp.2395/) (Li et al., Findings 2026)
ACL