Story2Proposal: A Scaffold for Structured Scientific Paper Writing
Zhuoyang Qian, Wei Shi, Xu Lin, Li Ling, Meng Luo, Ziming Wang, Zhiwei Zhang, Tengyue Xu, Gaoge Liu, Zhentao Zhang, Shuo Zhang, Ziqi Wang, Zheng Feng, Yan Luo, Shu Xu, Yongjin Chen, Zhibo Feng, Zhuo Chen, Bruce Yuan, Biao Wu, Harry Wang, Kris Chen
2026-03-31
Summary
This paper introduces a new system called Story2Proposal that aims to automatically write scientific papers in a more organized and accurate way, keeping the text, data, and figures consistent throughout the document.
What's the problem?
Currently, when AI tries to write scientific papers, it often struggles to maintain a logical flow and ensure everything fits together. The AI generates text first and *then* checks for errors, which leads to problems like sections that don't connect well, missing important figures or tables, and inconsistencies between what the text says and what the visuals show. Basically, the AI writes a draft and *then* tries to fix it, instead of building it correctly from the start.
What's the solution?
Story2Proposal uses a team of AI 'agents' working together, all guided by a central 'contract'. This contract keeps track of the paper's structure and what figures and tables are included. There's an 'architect' to plan the paper, a 'writer' to create the text, a 'refiner' to improve it, and a 'renderer' to create the visuals. 'Evaluation agents' constantly check the work and provide feedback, updating the contract as they go. This 'generate, evaluate, adapt' loop ensures everything stays aligned and consistent during the writing process.
Why it matters?
This research is important because it makes AI-generated scientific papers much more reliable and useful. By focusing on structure and consistency from the beginning, Story2Proposal produces papers that are closer in quality to those written by human experts, as shown by the improved scores compared to other AI writing tools. This could eventually help scientists automate parts of the writing process and accelerate research.
Abstract
Generating scientific manuscripts requires maintaining alignment between narrative reasoning, experimental evidence, and visual artifacts across the document lifecycle. Existing language-model generation pipelines rely on unconstrained text synthesis with validation applied only after generation, often producing structural drift, missing figures or tables, and cross-section inconsistencies. We introduce Story2Proposal, a contract-governed multi-agent framework that converts a research story into a structured manuscript through coordinated agents operating under a persistent shared visual contract. The system organizes architect, writer, refiner, and renderer agents around a contract state that tracks section structure and registered visual elements, while evaluation agents supply feedback in a generate evaluate adapt loop that updates the contract during generation. Experiments on tasks derived from the Jericho research corpus show that Story2Proposal achieved an expert evaluation score of 6.145 versus 3.963 for DirectChat (+2.182) across GPT, Claude, Gemini, and Qwen backbones. Compared with the structured generation baseline Fars, Story2Proposal obtained an average score of 5.705 versus 5.197, indicating improved structural consistency and visual alignment.