<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="jiang-etal-2026-code">
    <titleInfo>
        <title>Code-TVT: Synthesizing Test-Case-Driven Verifiable Tasks for Code Reinforcement Learning</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Jin</namePart>
        <namePart type="family">Jiang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Junlong</namePart>
        <namePart type="family">Zhang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yuchen</namePart>
        <namePart type="family">Yan</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name>
        <namePart>Zhaoze</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Changzhi</namePart>
        <namePart type="family">Zhou</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Suncong</namePart>
        <namePart type="family">Zheng</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Liangcai</namePart>
        <namePart type="family">Gao</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>Verifiers play a central role in large language model reasoning by providing reliable supervision for post-training methods such as reinforcement learning. In code generation, however, verification is still predominantly based on fixed test cases, offering a narrow but underexploited form of verifiable supervision. We explore how existing test cases can be reformulated into more diverse and challenging sources of verifiable training signals. We propose Code-TVT (Test-Case-Driven Verifiable Tasks), a data synthesis framework that transforms program test cases into intrinsically verifiable code reasoning tasks. Using test cases already available in programming problems, we design two complementary tasks: Edge Test-Case Generation, which constructs edge-case inputs that expose subtle correctness failures, and Input Reconstruction via Program Output Inversion, which reasons backward from outputs to plausible inputs under execution-based verification. Experiments show that training with Code-TVT consistently improves code generation performance on three time-windowed LiveCodeBench splits (LCB-2408-2411, LCB-V5, and LCB-V6) under rejection-sampling fine-tuning and Group Relative Policy Optimization.</abstract>
    <identifier type="citekey">jiang-etal-2026-code</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.1574/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>35842</start>
            <end>35865</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Code-TVT: Synthesizing Test-Case-Driven Verifiable Tasks for Code Reinforcement Learning
%A Jiang, Jin
%A Zhang, Junlong
%A Yan, Yuchen
%A Zhou, Changzhi
%A Zheng, Suncong
%A Gao, Liangcai
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%A Zhaoze
%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 jiang-etal-2026-code
%X Verifiers play a central role in large language model reasoning by providing reliable supervision for post-training methods such as reinforcement learning. In code generation, however, verification is still predominantly based on fixed test cases, offering a narrow but underexploited form of verifiable supervision. We explore how existing test cases can be reformulated into more diverse and challenging sources of verifiable training signals. We propose Code-TVT (Test-Case-Driven Verifiable Tasks), a data synthesis framework that transforms program test cases into intrinsically verifiable code reasoning tasks. Using test cases already available in programming problems, we design two complementary tasks: Edge Test-Case Generation, which constructs edge-case inputs that expose subtle correctness failures, and Input Reconstruction via Program Output Inversion, which reasons backward from outputs to plausible inputs under execution-based verification. Experiments show that training with Code-TVT consistently improves code generation performance on three time-windowed LiveCodeBench splits (LCB-2408-2411, LCB-V5, and LCB-V6) under rejection-sampling fine-tuning and Group Relative Policy Optimization.
%U https://aclanthology.org/2026.emnlp-main.1574/
%P 35842-35865

Markdown (Informal)

[Code-TVT: Synthesizing Test-Case-Driven Verifiable Tasks for Code Reinforcement Learning](https://aclanthology.org/2026.emnlp-main.1574/) (Jiang et al., EMNLP 2026)

• Code-TVT: Synthesizing Test-Case-Driven Verifiable Tasks for Code Reinforcement Learning (Jiang et al., EMNLP 2026)

ACL

• Jin Jiang, Junlong Zhang, Yuchen Yan, Zhaoze, Changzhi Zhou, Suncong Zheng, and Liangcai Gao. 2026. Code-TVT: Synthesizing Test-Case-Driven Verifiable Tasks for Code Reinforcement Learning. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 35842–35865, Budapest, Hungary. Association for Computational Linguistics.