Elamur Research⌕
☰
home / papers / 2026 / Inference-Time Scaling of Verification: Self-Evolving Deep Research Agents via Test-Time Rubric-Guided Verification

Inference-Time Scaling of Verification: Self-Evolving Deep Research Agents via Test-Time Rubric-Guided Verification

Paper · 2026 · Yuxuan Wan, Tianqing Fang, Zaitang LI et al.

Summary

Deep Research Agents (DRAs) often give unreliable answers because of wrong actions, failed tool calls or hallucinations. Long tasks make human oversight impractical. Most prior work improves the policy model through post-training. This paper instead improves agents at inference time with an iterative verify-and-refine loop that needs no extra training. The authors studied failed Cognitive Kernel-Pro (CK-Pro) trajectories on WebAggregatorQA and built a DRA Failure Taxonomy from them. Its rubrics drive DeepVerifier, a plug-in verifier that relies on the asymmetry of verification: it splits checking an answer into a few small retrieval sub-questions instead of solving the task again. It then sends structured feedback to the agent for another attempt. DeepVerifier beats vanilla agent-as-judge and LLM-judge baselines in meta-evaluation F1. Used at test time, it raises accuracy on GAIA, XBench-DeepSearch and BrowseComp. The authors also release an SFT dataset for open models. The work shows that verification itself can be scaled at inference time, and the taxonomy could supply reward signals for RL training of agents.

Key points

  • Taxonomy data:
    • Built from 2,997 agent steps over 90 WebAggregatorQA tasks, run with CK-Pro on a Claude-3.7-Sonnet backbone.
    • Two annotators marked 555 error points, with about 63% overlap between them.
    • Errors are grouped into 5 major categories and 13 sub-categories.
  • Verifier quality: DeepVerifier improves meta-evaluation F1 by 12–48% over agent-as-judge and LLM-judge baselines.
  • GAIA results with Claude-3.7-Sonnet:
    • Full set: accuracy rises from about 52% to a peak of 60.1%.
    • Web subset: accuracy rises from 51.1% to a peak of 63.3%.
    • Gains top out after roughly 3–4 feedback rounds.
    • Each check is limited to at most 3 follow-up questions.
  • Other benchmarks: best gains are +6 points on XBench-DeepSearch and +5 points on BrowseComp.
  • Open resources:
    • DeepVerifier-4K contains 4,646 verification prompt-response pairs, filtered from 400 trajectories.
    • DeepVerifier-8B is Qwen3-8B fine-tuned on this dataset plus CK-Pro data. After 10 feedback rounds it reaches 32.2% on GAIA, a gain of 5.5 points from reflection.
    • For comparison, CK-Pro-8B gains 2.6 points from reflection and Qwen3-8B barely improves.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🏢TencentShenzhen · China134 works · 19,743 citations🎓Renmin University of ChinaBeijing · China114 works · 16,118 citations🎓Chinese University of Hong KongHong Kong · China126 works · 14,073 citations🎓Singapore Management UniversitySingapore · Singapore27 works · 2,382 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

Cite

Cite as Yuxuan Wan, Tianqing Fang, Zaitang LI, Yintong Huo, Wenxuan Wang, Haitao Mi, Dong Yu and Michael R. Lyu. 2026. Inference-Time Scaling of Verification: Self-Evolving Deep Research Agents via Test-Time Rubric-Guided Verification. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1243

@inproceedings{wan2026inference,
  title = {{Inference-Time Scaling of Verification: Self-Evolving Deep Research Agents via Test-Time Rubric-Guided Verification}},
  author = {Yuxuan Wan, Tianqing Fang, Zaitang LI, Yintong Huo, Wenxuan Wang, Haitao Mi, Dong Yu and Michael R. Lyu},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1243},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1243}
}

References

58 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Dzmitry Bahdanau, Nicolas Gontier, Gabriel Huang, Ehsan Kamalloo, Rafael Pardinas, Alex Piché, Torsten Scholak, Oleh Shliazhko, Jordan Prince Tremblay, Karam Ghanem, Soham Parikh, Mitul Tiwari, and Quaizar Vohra. 2024. Tapeagents: a holistic framework for agent development and optimization. arXiv pr… find
  2. Kaiyuan Chen, Yixin Ren, Yang Liu, Xiaobo Hu, Haotong Tian, Tianbao Xie, Fangfu Liu, Haoye Zhang, Hongzhang Liu, Yuan Gong, and 1 others. 2025. xbench: Tracking agents productivity scaling with profession-aligned real-world evaluations. arXiv preprint arXiv:2506.13651. — [arXiv:2506.13651](https://arxiv.org/abs/2506.13651) find
  3. Sehyun Choi, Tianqing Fang, Zhaowei Wang, and Yangqiu Song. 2023. KCTS: knowledge-constrained tree search decoding with token-level hallucination detection. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, EMNLP 2023, Singapore, December 6-10, 2023, pages 14… find
  4. Tianyu Fan, Xinyao Niu, Yuxiang Zheng, Fengji Zhang, Chengen Huang, Bei Chen, Junyang Lin, and Chao Huang. 2025. Understanding deepresearch via reports. arXiv preprint arXiv:2510.07861. — [arXiv:2510.07861](https://arxiv.org/abs/2510.07861) find
  5. Tianqing Fang, Hongming Zhang, Zhisong Zhang, Kaixin Ma, Wenhao Yu, Haitao Mi, and Dong Yu. 2025a. Webevolver: Enhancing web agent self-improvement with coevolving world model. arXiv preprint arXiv:2504.21024. — [arXiv:2504.21024](https://arxiv.org/abs/2504.21024) find
  6. Tianqing Fang, Zhisong Zhang, Xiaoyang Wang, Rui Wang, Can Qin, Yuxuan Wan, Jun-Yu Ma, Ce Zhang, Jiaqi Chen, Xiyun Li, Hongming Zhang, Haitao Mi, and Dong Yu. 2025b. Cognitive kernel-pro: A framework for deep research agents and agent foundation models training. Preprint, arXiv:2508.00414. — [arXiv:2508.00414](https://arxiv.org/abs/2508.00414) find
  7. Yangyi Fang, Jiaye Lin, Xiaoliang Fu, Cong Qin, Haolin Shi, Chaowen Hu, Lu Pan, Ke Zeng, and Xunliang Cai. 2026a. How to allocate, how to learn? dynamic rollout allocation and advantage modulation for policy optimization. arXiv preprint arXiv:2602.19208. find
  8. Yangyi Fang, Jiaye Lin, Xiaoliang Fu, Cong Qin, Haolin Shi, Chang Liu, and Peilin Zhao. 2026b. Proximity-based multi-turn optimization: Practical credit assignment for llm agent training. arXiv preprint arXiv:2602.19225. find
  9. Kaituo Feng, Manyuan Zhang, Hongyu Li, Kaixuan Fan, Shuang Chen, Yilei Jiang, Dian Zheng, Peiwen Sun, Yiyuan Zhang, Haoze Sun, and 1 others. 2025. Onethinker: All-in-one reasoning model for image and video. arXiv preprint arXiv:2512.03043. — [arXiv:2512.03043](https://arxiv.org/abs/2512.03043) find
  10. Gonzalo Gonzalez-Pumariega, Vincent Tu, Chih-Lun Lee, Jiachen Yang, Ang Li, and Xin Eric Wang. 2025. The unreasonable effectiveness of scaling agents for computer use. find
  11. Google DeepMind. 2025. Gemini deep research — your personal research assistant. find
  12. Anisha Gunjal, Anthony Wang, Elaine Lau, Vaskar Nath, Yunzhong He, Bing Liu, and Sean Hendryx. 2025. Rubrics as rewards: Reinforcement learning beyond verifiable domains. arXiv preprint arXiv:2507.17746. find
  13. Hongliang He, Wenlin Yao, Kaixin Ma, Wenhao Yu, Yong Dai, Hongming Zhang, Zhenzhong Lan, and Dong Yu. 2024. Webvoyager: Building an end-to-end web agent with large multimodal models. arXiv preprint arXiv:2401.13919.
  14. Chen Hu, Haikuo Du, Heng Wang, Lin Lin, Mingrui Chen, Peng Liu, Ruihang Miao, Tianchi Yue, Wang You, Wei Ji, Wei Yuan, Wenjin Deng, Xiaojian Yuan, Xiaoyun Zhang, Xiangyu Liu, Xikai Liu, Yanming Xu, Yicheng Cao, Yifei Zhang, and 48 others. 2025a. Step-deepresearch technical report. Preprint, arXiv:25… — [arXiv:2512.20491](https://arxiv.org/abs/2512.20491) find
  15. Mengkang Hu, Yuhang Zhou, Wendong Fan, Yuzhou Nie, Bowei Xia, Tao Sun, Ziyu Ye, Zhaoxuan Jin, Yingru Li, Qiguang Chen, Zeyu Zhang, Yifeng Wang, Qianshuo Ye, Bernard Ghanem, Ping Luo, and Guohao Li. 2025b. Owl: Optimized workforce learning for general multi-agent assistance in real-world task automat… find
  16. Minda Hu, Tianqing Fang, Jianshu Zhang, Junyu Ma, Zhisong Zhang, Jingyan Zhou, Hongming Zhang, Haitao Mi, Dong Yu, and Irwin King. 2025c. Webcot: Enhancing web agent reasoning by reconstructing chain-of-thought in reflection, branching, and rollback. arXiv preprint arXiv:2505.20013. find
  17. Zenan Huang, Yihong Zhuang, Guoshan Lu, Zeyu Qin, Haokai Xu, Tianyu Zhao, Ru Peng, Jiaqi Hu, Zhanming Shen, Xiaomeng Hu, and 1 others. 2025. Reinforcement learning with rubric anchors. arXiv preprint arXiv:2508.12790. — [arXiv:2508.12790](https://arxiv.org/abs/2508.12790) find
  18. Jonas Hübotter, Frederike Lübeck, Lejs Behric, Anton Baumann, Marco Bagatella, Daniel Marta, Ido Hakimi, Idan Shenfeld, Thomas Kleine Buening, Carlos Guestrin, and Andreas Krause. 2026. Reinforcement learning via self-distillation. arXiv preprint arXiv:2601.20802. find
  19. Eric Li and Jim Waldo. 2024. Websuite: Systematically evaluating why web agents fail. arXiv preprint arXiv:2406.01623. — [arXiv:2406.01623](https://arxiv.org/abs/2406.01623) find
  20. Kuan Li, Zhongwang Zhang, Huifeng Yin, Liwen Zhang, Litu Ou, Jialong Wu, Wenbiao Yin, Zhengwei Tao, Xinyu Wang, Weizhou Shen, Junkai Zhang, Dingchu Zhang, Xixi Wu, Yong Jiang, Ming Yan, Pengjun Xie, Fei Huang, and Jingren Zhou. 2025. Websailor: Navigating super-human reasoning for web agent. find
  21. Mukai Li, Qingcheng Zeng, Tianqing Fang, Zhenwen Liang, Linfeng Song, Qi Liu, Haitao Mi, and Dong Yu. 2026. Verified critical step optimization for LLM agents. CoRR, abs/2602.03412. find
  22. Zihan Liu, Shun Zheng, Xumeng Wen, Yang Wang, Jiang Bian, and Mao Yang. 2025. Deep self-evolving reasoning. arXiv preprint arXiv:2510.17498. — [arXiv:2510.17498](https://arxiv.org/abs/2510.17498) find
  23. Xing Han Lù, Amirhossein Kazemnejad, Nicholas Meade, Arkil Patel, Dongchan Shin, Alejandra Zambrano, Karolina Stánczak, Peter Shaw, Christopher J. Pal, and Siva Reddy. 2025. Agentrewardbench: Evaluating automatic evaluations of web agent trajectories. arXiv preprint arXiv:2504.08942. — [arXiv:2504.08942](https://arxiv.org/abs/2504.08942) find
  24. Grégoire Mialon, Clémentine Fourrier, Craig Swift, Thomas Wolf, Yann LeCun, and Thomas Scialom. 2023. Gaia: a benchmark for general ai assistants. ArXiv, abs/2311.12983.
  25. Moonshot AI. 2025a. Kimi-k2. find
  26. Moonshot AI. 2025b. Kimi-researcher: End-to-end rl training for emerging agentic capabilities. find
  27. OpenAI. 2025. Introducing deep research. Technical report, OpenAI. find
  28. Jiayi Pan, Yichi Zhang, Nicholas Tomlin, Yifei Zhou, Sergey Levine, and Alane Suhr. 2024. Autonomous evaluation and refinement of digital agents. arXiv preprint arXiv:2404.06474. find
  29. Perplexity AI. 2025. Introducing perplexity deep research. find
  30. Long Phan, Alice Gatti, Ziwen Han, Nathaniel Li, Josephina Hu, Hugh Zhang, Chen Bo Calvin Zhang, Mohamed Shaaban, John Ling, Sean Shi, and 1 others. 2025. Humanity’s last exam. arXiv preprint arXiv:2501.14249.
  31. Aymeric Roucher, Albert Villanova del Moral, Thomas Wolf, Leandro von Werra, and Erik Kaunismäki. 2025. Smolagents: a smol library to build great agentic systems. find
  32. Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. Reflexion: Language agents with verbal reinforcement learning. Advances in Neural Information Processing Systems, 36:8634–8652.
  33. Charlie Snell, Jaehoon Lee, Kelvin Xu, and Aviral Kumar. 2024. Scaling llm test-time compute optimally can be more effective than scaling model parameters. arXiv preprint arXiv:2408.03314.
  34. Kevin Song, Anand Jayarajan, Yaoyao Ding, Qidong Su, Zhanda Zhu, Sihang Liu, and Gennady Pekhimenko. 2025. Aegis: Taxonomy and optimizations for overcoming agent-environment failures in llm agents. arXiv preprint arXiv:2508.19504. — [arXiv:2508.19504](https://arxiv.org/abs/2508.19504) find
  35. Jiabin Tang, Tianyu Fan, and Chao Huang. 2025. Autoagent: A fully-automated and zero-code framework for llm agents. arXiv preprint arXiv:2502.05957.
  36. Zhengwei Tao, Jialong Wu, Wenbiao Yin, Junkai Zhang, Baixuan Li, Haiyang Shen, Kuan Li, Liwen Zhang, Xinyu Wang, Yong Jiang, Pengjun Xie, Fei Huang, and Jingren Zhou. 2025. Webshaper: Agentically data synthesizing via information-seeking formalization. find
  37. Tongyi DeepResearch Team, Baixuan Li, Bo Zhang, Dingchu Zhang, Fei Huang, Guangyu Li, Guoxin Chen, Huifeng Yin, Jialong Wu, Jingren Zhou, and 1 others. 2025. Tongyi deepresearch technical report. arXiv preprint arXiv:2510.24701. — [arXiv:2510.24701](https://arxiv.org/abs/2510.24701) find
  38. Rui Wang, Ce Zhang, Junyu Ma, Jianshu Zhang, Hongru Wang, Yi Chen, Boyang Xue, Tianqing Fang, Zhisong Zhang, Hongming Zhang, Haitao Mi, Dong Yu, and Kam-Fai Wong. 2025. Explore to evolve: Scaling evolved aggregation logic via proactive online exploration for deep research agents. find
  39. Tianyi Wang, Long Li, Hongcan Guo, Yibiao Chen, Yixia Li, Yong Wang, Yun Chen, and Guanhua Chen. 2026. Anchored policy optimization: Mitigating exploration collapse via support-constrained rectification. CoRR, abs/2602.05717. — [arXiv:2602.05717](https://arxiv.org/abs/2602.05717) find
  40. Jason Wei. 2025. Asymmetry of verification and verifier’s law. Accessed: 2025-10-30. find
  41. Jason Wei, Zhiqing Sun, Spencer Papay, Scott McKinney, Jeffrey Han, Isa Fulford, Hyung Won Chung, Alex Tachard Passos, William Fedus, and Amelia Glaese. 2025. Browsecomp: A simple yet challenging benchmark for browsing agents. arXiv preprint arXiv:2504.12516. find
  42. Jialong Wu, Baixuan Li, Runnan Fang, Wenbiao Yin, Liwen Zhang, Zhengwei Tao, Dingchu Zhang, Zekun Xi, Gang Fu, Yong Jiang, Pengjun Xie, Fei Huang, and Jingren Zhou. 2025a. Webdancer: Towards autonomous information seeking agency. preprint arXiv:2505.22648. arXiv Jialong Wu, Wenbiao Yin, Yong Jiang,…
  43. Yiliu Yang, Yilei Jiang, Qunzhong Wang, Yingshui Tan, Xiaoyong Zhu, Sherman SM Chow, Bo Zheng, and Xiangyu Yue. 2025. Quadsentinel: Sequent safety for machine-checkable control in multi-agent systems. arXiv preprint arXiv:2512.16279. — [arXiv:2512.16279](https://arxiv.org/abs/2512.16279) find
  44. Wenhao Yu, Zhenwen Liang, Chengsong Huang, Kishan Panaganti, Tianqing Fang, Haitao Mi, and Dong Yu. 2025. Guided self-evolving llms with minimal human supervision. CoRR, abs/2512.02472. — [arXiv:2512.02472](https://arxiv.org/abs/2512.02472) find
  45. Mert Yuksekgonul, Federico Bianchi, Joseph Boen, Sheng Liu, Zhi Huang, Carlos Guestrin, and James Zou. 2024. Textgrad: Automatic" differentiation" via text. arXiv preprint arXiv:2406.07496.
  46. Dingling Zhang, He Zhu, Jincheng Ren, Kangqi Song, Xinran Zhou, Boyu Feng, Shudong Liu, Jiabin Luo, Weihao Xie, Zhaohui Wang, and 1 others. 2025a. How far are we from genuinely useful deep research agents? arXiv preprint arXiv:2512.01948. — [arXiv:2512.01948](https://arxiv.org/abs/2512.01948) find
  47. He Zhu, Tianrui Qin, King Zhu, Heyuan Huang, Yeyi Guan, Jinxiang Xia, Yi Yao, Hanhao Li, Ningning Wang, Pai Liu, Tianhao Peng, Xin Gui, Xiaowan Li, Yuhui Liu, Yuchen Eleanor Jiang, Jun Wang, Changwang Zhang, Xiangru Tang, Ge Zhang, and 5 others. 2025a. Oagents: An empirical study of building effecti… find
  48. King Zhu, Hanhao Li, Siwei Wu, Tianshun Xing, Dehua Ma, Xiangru Tang, Minghao Liu, Jian Yang, Jiaheng Liu, Yuchen Eleanor Jiang, Changwang Zhang, Chenghua Lin, Jun Wang, Ge Zhang, and Wangchunshu Zhou. 2025b. Scaling test-time compute for llm agents. Preprint, arXiv:2506.12928. — [arXiv:2506.12928](https://arxiv.org/abs/2506.12928) find
  49. King Zhu, Hanhao Li, Siwei Wu, Tianshun Xing, Dehua Ma, Xiangru Tang, Minghao Liu, Jian Yang, Jiaheng Liu, Yuchen Eleanor Jiang, and 1 others. 2025c. Scaling test-time compute for llm agents. arXiv preprint arXiv:2506.12928. — [arXiv:2506.12928](https://arxiv.org/abs/2506.12928) find
  50. Mingchen Zhuge, Changsheng Zhao, Dylan Ashley, Wenyi Wang, Dmitrii Khizbullin, Yunyang Xiong, Zechun Liu, Ernie Chang, Raghuraman Krishnamoorthi, Yuandong Tian, Yangyang Shi, Vikas Chandra, and Jürgen Schmidhuber. 2024. Agent-as-a-judge: Evaluate agents with agents. arXiv preprint arXiv:2410.10934. find
  51. Hangfan Zhang, Siyuan Xu, Zhimeng Guo, Huaisheng Zhu, Shicheng Liu, Xinrun Wang, Qiaosheng Zhang, Yang Chen, Peng Ye, Lei Bai, and Shuyue Hu. 2025b. The path of self-evolving large language models: Achieving data-efficient learning via intrinsic feedback. arXiv preprint arXiv:2510.02752. — [arXiv:2510.02752](https://arxiv.org/abs/2510.02752) find
  52. Yuxin Zuo, Kaiyan Zhang, Li Sheng, Shang Qu, Ganqu Cui, Xuekai Zhu, Haozhan Li, Yuchen Zhang, Xinwei Long, Ermo Hua, Biqing Qi, Youbang Sun, Zhiyuan Ma, Lifan Yuan, Ning Ding, and Bowen Zhou. 2025. Ttrl: Test-time reinforcement learning. In Advances in Neural Information Processing Systems (NeurIPS… find
  53. Hongming Zhang, Xiaoman Pan, Hongwei Wang, Kaixin Ma, Wenhao Yu, and Yu Dong. 2024. Cognitive kernel: An open-source agent system towards generalist autopilots. CoRR, abs/2409.10277. find
  54. A Jianshu Zhang, Chengxuan Qian, Haosen Sun, Haoran Lu, Dingcheng Wang, Letian Xue, and Han Liu. 2026. Progresslm: Towards progress reasoning in vision-language models. arXiv preprint arXiv:2601.15224.
  55. Annotation Instructions This instruction is used for the human annotator for summarizing the error points in each erroneous trajectory. find
  56. Yifei Zhou, Sergey Levine, Jason Weston, Xian Li, and Sainbayar Sukhbaatar. 2025a. Self-challenging language model agents. arXiv preprint arXiv:2506.01716. — [arXiv:2506.01716](https://arxiv.org/abs/2506.01716) find
  57. Yifei Zhou, Sergey Levine, Jason Weston, Xian Li, and Sainbayar Sukhbaatar. 2025b. Self-challenging language model agents. In Advances in Neural Information Processing Systems (NeurIPS 2025). find
  58. Bin Zhu, Qianghuai Jia, Tian Lan, Junyang Ren, Feng Gu, Feihu Jiang, Longyue Wang, Zhao Xu, and Weihua Luo. 2026. Marco deepresearch: Unlocking efficient deep research agents via verification-centric design. arXiv preprint arXiv:2603.28376. — [arXiv:2603.28376](https://arxiv.org/abs/2603.28376) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wan-2026-inference-time-scaling-verification-self/