EvidFuse: Writing-Time Evidence Learning for Consistent Text–Chart Data Reporting
Huanxiang Lin, Qianyue Wang, Jinwu Hu, Bailin Chen, Qing Du, Mingkui Tan · 2026
Read EvidFuse: Writing-Time Evidence Learning for Consistent Text–Chart Data Reporting by Huanxiang Lin, Qianyue Wang, Jinwu Hu, Bailin Chen, Qing Du, Mingkui Tan (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-05487.txt.
Description
Huanxiang Lin and 5 co-authors propose EvidFuse, a 2026 framework for writing data reports where text and charts stay consistent. Many automated systems draft text and charts in separate stages. That split can freeze the evidence early, so later claims cannot draw on new visual analysis.
EvidFuse instead builds evidence during writing. An analysis agent with exploratory data tools and raw table access works beside a writer agent that plans and drafts the report. The writer can request fine-grained analysis whenever the narrative needs it, and new evidence then constrains the next claims. In judge-based and human evaluations, EvidFuse ranked first for chart quality, chart-text alignment, and overall usefulness.
Catalog record
| CREATOR | Huanxiang Lin, Qianyue Wang, Jinwu Hu, Bailin Chen, Qing Du, Mingkui Tan |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 73 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-05487