ScienceIDE: Scaling Scientific Experience toward 1,000 Executable Environments
Abstract
Scientific software and expertise do not readily become experience that agents can learn from: repositories depend on fragile toolchains and unwritten conventions, and correctness rests on scientific behaviour and numerical tolerances. ScienceIDE is infrastructure that connects scientific practice with agent development by making expert knowledge reusable. Domain experts define scientific cases and acceptance criteria, agents help turn those decisions into executable environments, and task factories generate diverse challenges whose validity is established through execution and scientific verification. The same environments yield graded trajectories for supervised fine-tuning, online rewards for reinforcement learning, and evidence for evaluation.
Authors
* Equal contribution
Venue
Cite
BibTeX
@techreport{geng2026scienceide,
title = {{ScienceIDE}: Scaling Scientific Experience toward 1,000 Executable Environments},
author = {Geng, Hejia and Huang, Zesen and Li, Haoyang and Li, Wenbin and Wu, Koutian and Zhou, Zihan and Pang, Yuanbo and Liu, Weihao and Xu, Zigong and Shi, Yueheng and Ma, Yue and Zhang, Zongzheng and Zheng, Tianzhe and Dong, Chuanfei and Xie, Fengyu and Li, Zhiping and Sun, Jiankai and Gao, Qucheng and Pan, Jiaming and Zhu, Zhenlin and Zhang, Peijin and Yuan, Lanqing and Xu, Liuwei and Liu, Xianrong and Xie, Tong and Wang, Junkai and Di, Zonglin and Ma, Xiao-Han and Liang, Kangkai and Liu, Ziang and Huang, Sheng and Zhao, Zehong and Xu, Ziyang and Xie, Jingxu and Xing, Yaopeng and Xian, Jiayi and Meng, Xing and Yin, Zhenfei and Wu, Yingcheng and Yang, Ling},
institution = {PhAI Labs},
type = {Technical Report},
number = {PHAI-TR-2026-03},
month = {September},
year = {2026},
url = {https://phai-labs.com/papers/scienceide/},
note = {Version v1. Equal contribution: Hejia Geng, Zesen Huang}
}