AGORA

An Archive-Grounded Benchmark for Agentic Workplace Document Reasoning

AGORA is a benchmark for archive-grounded agentic workplace document reasoning, with 362 questions across eight domains, 9,664 documents and 372M tokens. Agents search a fixed document collection, combine evidence across files, and return a numerical answer.

Files

• tasks/*.jsonl: task ID, domain, question (query), gold answer, reference_reasoning_path and reference_files.
• documents.zip: the document collection, organized as documents/<domain>/.
• documents.manifest.jsonl: document paths, byte sizes and SHA-256 digests.

Task keys are (domain, id). Documents include Markdown, Excel and CSV files; questions are in English.

reference_reasoning_path contains the reference solution steps: step, input, output and feeds_into (a list of following step IDs, empty for the final step). reference_files lists path, page_start and page_end. Paths are relative to the dataset root (data/ after download), for example documents/Healthcare & Medicine/example.md. Page ranges are inclusive, use the original PDF page numbers, and match the <!-- pages N-M --> markers in Markdown. A null page range refers to the whole document.

Reference solutions are evaluator-side data. Agents should receive only the question and the permitted document collection.

Download

uv run scripts/download_data.py

Run this command from the evaluation-code repository. It downloads the dataset and extracts the document collection.

Use

The code is licensed under the MIT License.

The data is intended for academic research. Terms and rights of the underlying source documents apply.

Citation

@article{guo2026agora,
  author       = {Honglin Guo and Qi Zhang and Yu Zhang and Weijie Li and Rui Zheng and Zhikai Lei and Qiyuan Peng and Zhiheng Xi and Tao Gui and Qi Zhang},
  title        = {{AGORA:} An Archive-Grounded Benchmark for Agentic Workplace Document
                  Reasoning},
  journal      = {CoRR},
  volume       = {abs/2606.24526},
  year         = {2026},
  url          = {https://doi.org/10.48550/arXiv.2606.24526},
  doi          = {10.48550/ARXIV.2606.24526},
  eprinttype   = {arXiv},
  eprint       = {2606.24526},
  timestamp    = {Thu, 09 Jul 2026 07:25:37 +0200},
}