Elamur Research⌕
☰
home / papers / 2026 / LEDGER: Scaling Agentic Document Editing with Dependency-aware Graph Retrieval

LEDGER: Scaling Agentic Document Editing with Dependency-aware Graph Retrieval

Paper · 2026 · Hang Wang, Utkarsh Garg, Reza Davari et al.

Summary

The paper tackles agentic document editing: an AI agent applies localized changes to long, structured documents from high-level instructions. Edits often break cross-references, terminology, or style elsewhere in the text. Feeding the whole document for every edit is costly and can blur the model's focus. Similarity-based retrieval, in turn, tends to overlook structural links. LEDGER, from Microsoft, addresses this by first parsing a document (via HTML into a DOM tree) into units such as sections, paragraphs, figures and equations. It then uses LLM-driven extraction protocols to build a typed dependency graph over those units. For each edit, the system traverses the graph to collect both what the target relies on and what depends on it. It ranks these by edge type and fits them into a token budget. After the edit, it checks consistency and updates the graph incrementally. On a new benchmark of about 1.9k cases, consistency rose from 56% to 76% across six models, with lower token cost. Explicit structure thus appears able to partly replace expensive model reasoning.

Key points

  • The graph uses four edge types: REFERENCES (explicit mentions like "Figure 2"), DEPENDS (implicit prerequisites, detected via a counterfactual LLM test), CONTAINS (hierarchy), and RELATED (semantic similarity, threshold-based).
  • The benchmark has roughly 1.9k test cases across varied document types, lengths, edit scenarios and complexity levels. Evaluated models were Claude 4.5 Haiku/Sonnet/Opus and GPT-4.1/5/5.2.
  • Context per edit stays near 1,500 tokens regardless of document size. Scalability tests show up to 85% fewer tokens than regenerating the full document.
  • LEDGER at low reasoning effort reached the baseline's quality at high reasoning effort while consuming fewer tokens.
  • Limitations include:
    • evaluation mostly on English documents, so other languages may need adapted extraction, especially for coreference-based dependencies;
    • reliance on well-structured, parseable documents;
    • graph quality bounded by the underlying LLM.

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

 Fields, topics and concepts

📚 Shelves 5The 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 worksInformation Systems · 1,437 works

🧭 Topics 1Research subjects and languages the work is about.Information retrieval & search · 473 works

Cite

Cite as Hang Wang, Utkarsh Garg, Reza Davari, Huitian Jiao, Hao Cheng, Baolin Peng, Si-Qing Chen and Tao Ge. 2026. LEDGER: Scaling Agentic Document Editing with Dependency-aware Graph Retrieval. 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.515

@inproceedings{wang2026ledger,
  title = {{LEDGER: Scaling Agentic Document Editing with Dependency-aware Graph Retrieval}},
  author = {Hang Wang, Utkarsh Garg, Reza Davari, Huitian Jiao, Hao Cheng, Baolin Peng, Si-Qing Chen and Tao Ge},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.515},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.515}
}

References

26 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Miltiadis Allamanis, Marc Brockschmidt, and Mahmoud Khademi. 2018. Learning to represent programs with graphs. In International Conference on Learning Representations. find
  2. Shengnan An, Zexiong Ma, Zeqi Lin, Nanning Zheng, Jian-Guang Lou, and Weizhu Chen. 2024. Make your llm fully utilize the context. Advances in Neural Information Processing Systems, 37:62160–62188. find
  3. Jiawen Chen, Yajun Yang, Chenyang Wang, Heng Zhang, Chao Qiu, and Xiaofei Wang. 2021. Multitask offloading strategy optimization based on directed acyclic graphs for edge computing. IEEE Internet of Things Journal, 9(12):9367–9378. find
  4. Xinyun Chen, Maxwell Lin, Nathanael Schärli, and Denny Zhou. 2023. Teaching large language models to self-debug. In International Conference on Learning Representations. find
  5. Daya Guo, Shuo Ren, Shuai Lu, Zhangyin Feng, Duyu Tang, Shujie Liu, Long Zhou, Nan Duan, Alexey Svyatkovskiy, Shengyu Fu, Michele Tufano, Shao Kun Deng, Colin Clement, Dawn Drain, Neel Sundaresan, Jian Yin, Daxin Jiang, and Ming Zhou. 2021. Graph-CodeBERT: Pre-training code representations with data… find
  6. William L Hamilton, Rex Ying, and Jure Leskovec. 2017. Inductive representation learning on large graphs. In Advances in Neural Information Processing Systems, volume 30.
  7. Coleman Richard Charles Hooper, Sehoon Kim, Hiva Mohammadzadeh, Monishwaran Maheswaran, Sebastian Zhao, June Paik, Michael W Mahoney, Kurt Keutzer, and Amir Gholami. 2025. Squeezed attention: Accelerating long context length llm inference. In Proceedings of the 63rd Annual Meeting of the Association… find
  8. Wenlong Huang, Pieter Abbeel, Deepak Pathak, and Igor Mordatch. 2022. Language models as zero-shot planners: Extracting actionable knowledge for embodied agents. In International Conference on Machine Learning.
  9. Hongye Jin, Xiaotian Han, Jingfeng Yang, Zhimeng Jiang, Zirui Liu, Chia-Yuan Chang, Huiyuan Chen, and Xia Hu. 2024. LLM maybe LongLM: Selfextend LLM context window without tuning. In International Conference on Machine Learning. find
  10. Jaz Kandola, Nello Cristianini, and John Shawe-taylor. 2002. Learning semantic similarity. Advances in neural information processing systems, 15. find
  11. Thomas N Kipf and Max Welling. 2017. Semi-supervised classification with graph convolutional networks. In International Conference on Learning Representations.
  12. Philippe Laban, Alexander Richard Fabbri, Caiming Xiong, and Chien-Sheng Wu. 2024. Summary of a haystack: A challenge to long-context llms and rag systems. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 9885–9903.
  13. Xunhao Lai, Jianqiao Lu, Yao Luo, Yiyuan Ma, and Xun Zhou. 2025. Flexprefill: A context-aware sparse attention mechanism for efficient long-sequence inference. arXiv preprint arXiv:2502.20766. — [arXiv:2502.20766](https://arxiv.org/abs/2502.20766) find
  14. 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…
  15. Di Liu, Meng Chen, Baotong Lu, Huiqiang Jiang, Zhenhua Han, Qianxi Zhang, Qi Chen, Chengruidong Zhang, Bailu Ding, Kai Zhang, and 1 others. 2024a. Retrievalattention: Accelerating long-context llm inference via vector retrieval. arXiv preprint arXiv:2409.10516. find
  16. Nelson F Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. 2024b. Lost in the middle: How language models use long contexts. Transactions of the Association for Computational Linguistics, 12:157–173.
  17. Adyasha Maharana, Dong-Ho Lee, Sergey Tulyakov, Mohit Bansal, Francesco Barbieri, and Yuwei Fang. 2024. Evaluating very long-term conversational memory of llm agents. arXiv preprint arXiv:2402.17753. find
  18. André Martins, António Farinhas, Marcos Treviso, Vlad Niculae, Pedro Aguiar, and Mario Figueiredo. 2020. Sparse and continuous attention mechanisms. Advances in Neural Information Processing Systems, 33:20989–21001. find
  19. Noah Shinn, Federico Cassano, Beck Labash, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. Reflexion: Language agents with verbal reinforcement learning. In Advances in Neural Information Processing Systems, volume 36.
  20. Lin Song, Yukang Chen, Shuai Yang, Xiaohan Ding, Yixiao Ge, Ying-Cong Chen, and Ying Shan. 2024. Low-rank approximation for sparse attention in multimodal llms. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13763–13773. find
  21. Weizhi Wang, Li Dong, Hao Cheng, Xiaodong Liu, Xifeng Yan, Jianfeng Gao, and Furu Wei. 2023. Augmenting language models with long-term memory. Advances in Neural Information Processing Systems, 36:74530–74543. find
  22. Xingyao Wang, Yangyi Chen, Lifan Yuan, Yizhe Zhang, Yunzhu Li, Hao Peng, and Heng Ji. 2024. Executable code actions elicit better LLM agents. In International Conference on Machine Learning.
  23. Chaojun Xiao, Pengle Zhang, Xu Hu, Guangxuan Liu, Yichuan Xu, Chenyang Lin, Chengqiang Wang, Yishi Dong, Furu Wei, and Yeyun Cao. 2024. InfLLM: Training-free long-context extrapolation for LLMs with an efficient context memory. In Advances in Neural Information Processing Systems, volume 37.
  24. Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. 2019. How powerful are graph neural networks? In International Conference on Learning Representations.
  25. Michihiro Yasunaga, Hongyu Ren, Antoine Bosselut, Percy Liang, and Jure Leskovec. 2021. QA-GNN: Reasoning with language models and knowledge graphs for question answering. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Lang… find
  26. Qizheng Zhang, Changran Hu, Shubhangi Upasani, Boyuan Ma, Fenglu Hong, Vamsidhar Kamanuru, Jay Rainton, Chen Wu, Mengmeng Ji, Hanchen Li, and 1 others. 2025. Agentic context engineering: Evolving contexts for self-improving language models. arXiv preprint arXiv:2510.04618. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wang-2026-ledger-scaling-agentic-document-editing/