Research

Paper

TESTING March 02, 2026

Tool Verification for Test-Time Reinforcement Learning

Authors

Ruotong Liao, Nikolai Röhrich, Xiaohan Wang, Yuhui Zhang, Yasaman Samadzadeh, Volker Tresp, Serena Yeung-Levy

Abstract

Test-time reinforcement learning (TTRL) has emerged as a promising paradigm for self-evolving large reasoning models (LRMs), enabling online adaptation on unlabeled test inputs via self-induced rewards through majority voting. However, a spurious yet high-frequency unverified consensus can become a biased and reinforced reward signal, leading to incorrect mode collapse. We address this failure mode with T^3RL (Tool-Verification for Test-Time Reinforcement Learning), which introduces test-time tool verification into reward estimation. Concretely, a verifier uses an external tool as evidence (e.g., from code execution) to upweight verified rollouts in a verification-aware voting, producing more reliable pseudo-labels for training. Across various math difficulties (MATH-500, AMC, and AIME 2024) and diverse backbone types, T^3RL significantly improves over TTRL, with larger gains on harder problems. More broadly, T^3RL can be viewed as verified online data synthesis, highlighting test-time tool verification as a key mechanism for stabilizing self-evolution.

Metadata

arXiv ID: 2603.02203
Provider: ARXIV
Primary Category: cs.AI
Published: 2026-03-02
Fetched: 2026-03-03 04:34

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.02203v1</id>\n    <title>Tool Verification for Test-Time Reinforcement Learning</title>\n    <updated>2026-03-02T18:57:52Z</updated>\n    <link href='https://arxiv.org/abs/2603.02203v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.02203v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Test-time reinforcement learning (TTRL) has emerged as a promising paradigm for self-evolving large reasoning models (LRMs), enabling online adaptation on unlabeled test inputs via self-induced rewards through majority voting. However, a spurious yet high-frequency unverified consensus can become a biased and reinforced reward signal, leading to incorrect mode collapse. We address this failure mode with T^3RL (Tool-Verification for Test-Time Reinforcement Learning), which introduces test-time tool verification into reward estimation. Concretely, a verifier uses an external tool as evidence (e.g., from code execution) to upweight verified rollouts in a verification-aware voting, producing more reliable pseudo-labels for training. Across various math difficulties (MATH-500, AMC, and AIME 2024) and diverse backbone types, T^3RL significantly improves over TTRL, with larger gains on harder problems. More broadly, T^3RL can be viewed as verified online data synthesis, highlighting test-time tool verification as a key mechanism for stabilizing self-evolution.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n    <published>2026-03-02T18:57:52Z</published>\n    <arxiv:comment>12 pages, 11 figures</arxiv:comment>\n    <arxiv:primary_category term='cs.AI'/>\n    <author>\n      <name>Ruotong Liao</name>\n    </author>\n    <author>\n      <name>Nikolai Röhrich</name>\n    </author>\n    <author>\n      <name>Xiaohan Wang</name>\n    </author>\n    <author>\n      <name>Yuhui Zhang</name>\n    </author>\n    <author>\n      <name>Yasaman Samadzadeh</name>\n    </author>\n    <author>\n      <name>Volker Tresp</name>\n    </author>\n    <author>\n      <name>Serena Yeung-Levy</name>\n    </author>\n  </entry>"
}