Elamur Research⌕
☰
home / papers / 2026 / SAGE: Steerable Agentic Data Generation for Deep Search with Execution Feedback

SAGE: Steerable Agentic Data Generation for Deep Search with Execution Feedback

Paper · 2026 · Fangyuan Xu, Rujun Han, Yanfei Chen et al.

Summary

Training search agents that chain many retrieval and reasoning steps requires hard question–answer pairs. Such data is costly for humans to write. Existing multi-hop datasets also rarely go beyond a few hops and depend on annotators or on Wikipedia link structure. SAGE addresses this with a two-agent loop. A data generator starts from a randomly sampled corpus document and a target number of search steps. It searches the corpus and drafts a QA pair grounded in the retrieved evidence. A separate search agent then tries to answer the question several times. Its traces show whether the pair is correct and whether it truly needs the intended number of searches. These traces go back to the generator, which revises the pair over several rounds. The authors used Wikipedia as the corpus and trained 3B and 7B agents with RL in the Search-R1 framework. The resulting data was more correct, harder and called for more varied reasoning. It also improved downstream results on both in-domain and out-of-domain benchmarks. The pipeline is cheap and reproducible because it relies on a fixed corpus rather than paid web-search APIs.

Key points

  • Error-correction mechanism: Generating questions backward from evidence was not enough on its own. Generators often misjudged how many searches a question would need, and only execution feedback from the search agent exposed this gap.
  • Downstream gains: Agents trained on SAGE data improved by about 27% (relative) on in-domain evaluation. On out-of-domain sets (Musique, FRAMES), the relative gain reached up to 23%, across 3B and 7B models.
  • Difficulty of the data: SAGE questions average roughly 4.9 searches. For comparison, NQ, HotpotQA and Musique need about 1.3–2.7, and FRAMES needs 3.2.
  • Transfer to web search: Agents trained only on fixed-corpus retrieval also worked with Google Search at inference time, with no extra training.
  • Error analysis and limitations:
    • About 20% of discarded incorrect pairs came from generator hallucinations, and about 7% came from ambiguous questions.
    • A pair counts as correct if the search agent solves it in at least one of K attempts (pass@K=1). This check can still let some wrong pairs through.
    • Experiments cover only Wikipedia and models of up to 7B parameters.

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

🏢GoogleMountain View · United States492 works · 440,462 citations🎓New York UniversityNew York · United States219 works · 103,487 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksInformation Systems · 1,730 works

Cite

Cite as Fangyuan Xu, Rujun Han, Yanfei Chen, Zifeng Wang, I-Hung Hsu, Jun Yan, Vishy Tirumalashetty, Eunsol Choi, Tomas Pfister and Chen-Yu Lee. 2026. SAGE: Steerable Agentic Data Generation for Deep Search with Execution Feedback. Findings of the Association for Computational Linguistics: EACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-eacl.174

@inproceedings{xu2026sage,
  title = {{SAGE: Steerable Agentic Data Generation for Deep Search with Execution Feedback}},
  author = {Fangyuan Xu, Rujun Han, Yanfei Chen, Zifeng Wang, I-Hung Hsu, Jun Yan, Vishy Tirumalashetty, Eunsol Choi, Tomas Pfister and Chen-Yu Lee},
  booktitle = {Findings of the Association for Computational Linguistics: EACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-eacl.174},
  url = {https://doi.org/10.18653/v1/2026.findings-eacl.174}
}

References

38 references, 22 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. 2024. Self-RAG: Learning to retrieve, generate, and critique through self-reflection. In The Twelfth International Conference on Learning Representations.
  2. Zijian Chen, Xueguang Ma, Shengyao Zhuang, Ping Nie, Kai Zou, Andrew Liu, Joshua Green, Kshama Patel, Ruoxi Meng, Mingyi Su, Sahel Sharifymoghaddam, Yanxi Li, Haoran Hong, Xinyu Shi, Xuye Liu, Nandan Thakur, Crystina Zhang, Luyu Gao, Wenhu Chen, and Jimmy Lin. 2025. Browsecomp-plus: A more fair and… — [arXiv:2508.06600](https://arxiv.org/abs/2508.06600) find
  3. Yihong Dong, Xue Jiang, Jiaru Qian, Tian Wang, Kechi Zhang, Zhi Jin, and Ge Li. 2025. A survey on code generation with llm-based agents. ArXiv, abs/2508.00083. — [arXiv:2508.00083](https://arxiv.org/abs/2508.00083) find
  4. Matthew Dunn, Levent Sagun, Mike Higgins, V. Ugur Guney, Volkan Cirik, and Kyunghyun Cho. 2017. Searchqa: A new qa dataset augmented with context from a search engine. arXiv preprint arXiv:1704.05179.
  5. Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yi Dai, Jiawei Sun, Qianyu Guo, Meng Wang, and Haofen Wang. 2023. Retrieval-augmented generation for large language models: A survey. ArXiv, abs/2312.10997.
  6. Rujun Han, Yanfei Chen, Zoey CuiZhu, Lesly Miculicich, Guan Sun, Yuanjun Bi, Weiming Wen, Hui Wan, Chunfeng Wen, Solène Maître, George Lee, Vishy Tirumalashetty, Emily Xue, Zizhao Zhang, Salem Haykal, Burak Gokturk, Tomas Pfister, and Chen-Yu Lee. 2025. Deep researcher with test-time diffusion. Prep… — [arXiv:2507.16075](https://arxiv.org/abs/2507.16075) find
  7. Rujun Han, Yuhao Zhang, Peng Qi, Yumo Xu, Jenyuan Wang, Lan Liu, William Yang Wang, Bonan Min, and Vittorio Castelli. 2024. Rag-qa arena: Evaluating domain robustness for long-form retrieval augmented question answering. arXiv preprint arXiv:2407.13998. find
  8. Carlos E Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik R Narasimhan. 2024. SWE-bench: Can language models resolve real-world github issues? In The Twelfth International Conference on Learning Representations.
  9. Bowen Jin, Hansi Zeng, Zhenrui Yue, Jinsung Yoon, Sercan O Arik, Dong Wang, Hamed Zamani, and Jiawei Han. 2025. Search-r1: Training LLMs to reason and leverage search engines with reinforcement learning. In Second Conference on Language Modeling.
  10. Mandar Joshi, Eunsol Choi, Daniel S. Weld, and Luke Zettlemoyer. 2017. Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, Vancouver, Canada. Association for Computationa…
  11. Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. 2020. Dense passage retrieval for open-domain question answering. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 6769–6781, Onli…
  12. Satyapriya Krishna, Kalpesh Krishna, Anhad Mohananey, Steven Schwarcz, Adam Stambler, Shyam Upadhyay, and Manaal Faruqui. 2025. Fact, fetch, and reason: A unified evaluation of retrieval-augmented generation. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Associa… find
  13. Chess, and John Schulman. 2021. Webgpt: Browser-assisted question-answering with human feedback. ArXiv, abs/2112.09332.
  14. Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Jacob Devlin, Kenton Lee, Kristina Toutanova, Llion Jones, Matthew Kelcey, Ming-Wei Chang, Andrew M. Dai, Jakob Uszkoreit, Quoc Le, and Slav Petrov. 2019. Natural…
  15. Ralph Peeters, Aaron Steiner, Luca Schwarz, Julian Yuya Caspary, and Christian Bizer. 2025. Webmall – a multi-shop benchmark for evaluating web agents. ArXiv, abs/2508.13024. — [arXiv:2508.13024](https://arxiv.org/abs/2508.13024) find
  16. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. In Advances in Neural Information Proc…
  17. Kuan Li, Zhongwang Zhang, Huifeng Yin, Liwen Zhang, Litu Ou, Jialong Wu, Wenbiao Yin, Baixuan Li, Zhengwei Tao, Xinyu Wang, Weizhou Shen, Junkai Zhang, Dingchu Zhang, Xixi Wu, Yong Jiang, Ming Yan, Pengjun Xie, Fei Huang, and Jingren Zhou. 2025a. Websailor: Navigating super-human reasoning for web a… find
  18. Xiaoxi Li, Guanting Dong, Jiajie Jin, Yuyao Zhang, Yujia Zhou, Yutao Zhu, Peitian Zhang, and Zhicheng Dou. 2025b. Search-o1: Agentic search-enhanced large reasoning models. ArXiv, abs/2501.05366.
  19. Xiaoxi Li, Guanting Dong, Jiajie Jin, Yuyao Zhang, Yujia Zhou, Yutao Zhu, Peitian Zhang, and Zhicheng Dou. 2025c. Search-o1: Agentic search-enhanced large reasoning models. CoRR, abs/2501.05366.
  20. Junteng Liu, Yunji Li, Chi Zhang, Jingyang Li, Aili Chen, Ke Ji, Weiyu Cheng, Zijia Wu, Chengyu Du, Qidi Xu, Jiayuan Song, Zhengmao Zhu, Wenhu Chen, Pengyu Zhao, and Junxian He. 2025. Webexplorer: Explore and evolve for training long-horizon web agents. Preprint, arXiv:2509.06501. — [arXiv:2509.06501](https://arxiv.org/abs/2509.06501) find
  21. Rui Lu, Zhenyu Hou, Zihan Wang, Hanchen Zhang, Xiao Liu, Yujiang Li, Shi Feng, Jie Tang, and Yuxiao Dong. 2025. Deepdive: Advancing deep search agents with knowledge graphs and multi-turn rl. arXiv preprint arXiv:2509.10446. — [arXiv:2509.10446](https://arxiv.org/abs/2509.10446) find
  22. Grégoire Mialon, Clémentine Fourrier, Craig Swift, Thomas Wolf, Yann LeCun, and Thomas Scialom. 2023. Gaia: a benchmark for general ai assistants. Preprint, arXiv:2311.12983.
  23. Reiichiro Nakano, Jacob Hilton, Suchir Balaji, Jeff Wu, Ouyang Long, Christina Kim, Christopher Hesse, Shantanu Jain, Vineet Kosaraju, William Saunders, Xu Jiang, Karl Cobbe, Tyna Eloundou, Gretchen Krueger, Kevin Button, Matthew Knight, Benjamin Long Phan, Alice Gatti, Ziwen Han, Nathaniel Li, Jose… — Humanity's Last Exam
  24. John Schulman, Philipp Moritz, Sergey Levine, Michael I. Jordan, and P. Abbeel. 2015. High-dimensional continuous control using generalized advantage estimation. CoRR, abs/1506.02438. find
  25. John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. 2017. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347. find
  26. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Jun-Mei Song, Mingchuan Zhang, Y. K. Li, Yu Wu, and Daya Guo. 2024. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. ArXiv, abs/2402.03300.
  27. Wenxuan Shi, Haochen Tan, Chuqiao Kuang, Xiaoguang Li, Xiaozhe Ren, Chen Zhang, Hanting Chen, Yasheng Wang, Lifeng Shang, Fisher Yu, and Yunhe Wang. 2025a. Pangu deepdiver: Adaptive search intensity scaling via open-web reinforcement learning. ArXiv, abs/2505.24332. — [arXiv:2505.24332](https://arxiv.org/abs/2505.24332) find
  28. Wenxuan Shi, Haochen Tan, Chuqiao Kuang, Xiaoguang Li, Xiaozhe Ren, Chen Zhang, Hanting Chen, Yasheng Wang, Lifeng Shang, Fisher Yu, and Yunhe Wang. 2025b. Pangu deepdiver: Adaptive search intensity scaling via open-web reinforcement learning. arXiv preprint arXiv:2505.24332. — [arXiv:2505.24332](https://arxiv.org/abs/2505.24332) find
  29. 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
  30. Harsh Trivedi, Niranjan Balasubramanian, Tushar Khot, and Ashish Sabharwal. 2022. MuSiQue: Multi-hop questions via single-hop question composition. Transactions of the Association for Computational Linguistics, 10:539–554. find
  31. Harsh Trivedi, Niranjan Balasubramanian, Tushar Khot, and Ashish Sabharwal. 2023. Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), p…
  32. Liang Wang, Nan Yang, Xiaolong Huang, Binxing Jiao, Linjun Yang, Daxin Jiang, Rangan Majumder, and Furu Wei. 2022. Text embeddings by weakly-supervised contrastive pre-training. ArXiv, abs/2212.03533.
  33. Jason Wei, Zhiqing Sun, Spencer Papay, Scott McKinney, Jeffrey Han, Is abella Fulford, Hyung Won Chung, Alexandre Passos, William Fedus, and Amelia Glaese. 2025. Browsecomp: A simple yet challenging benchmark for browsing agents. ArXiv, abs/2504.12516.
  34. Jialong Wu, Baixuan Li, Runnan Fang, Wenbiao Yin, Liwen Zhang, Zhengwei Tao, Dingchu Zhang, Zekun Xi, Yong Jiang, Pengjun Xie, Fei Huang, and Jingren Zhou. 2025. Webdancer: Towards autonomous information seeking agency.
  35. Qwen An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Guanting Dong, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxin Yang, Jingren Zhou, Junyang Lin, and 25 others. 2024. Qwen2.5 technical report. ArXiv, a… find
  36. Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, and Christopher D. Manning. 2018. HotpotQA: A dataset for diverse, explainable multi-hop question answering. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 2369…
  37. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik R Narasimhan, and Yuan Cao. 2023. React: Synergizing reasoning and acting in language models. In The Eleventh International Conference on Learning Representations.
  38. Shuyan Zhou, Frank F. Xu, Hao Zhu, Xuhui Zhou, Robert Lo, Abishek Sridhar, Xianyi Cheng, Tianyue Ou, Yonatan Bisk, Daniel Fried, Uri Alon, and Graham Neubig. 2024. Webarena: A realistic web environment for building autonomous agents. In The Twelfth International Conference on Learning Representation…

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/xu-2026-sage-steerable-agentic-data-generation/