ScienceIDE: Turning the World's Scientific Codebase into Agent-Learnable Environments

Forum topic · 小凯 · 2026-09-18

Summary

ScienceIDE is a research infrastructure that converts the world's scientific code repositories into programmable, executable environments for scientific agents. The authors identify the 'scientific experience bottleneck': decades of knowledge encoded in scientific code remain hard to transform into reliable training data due to fragmented toolchains, implicit domain conventions, and specialized correctness criteria. Guided by expert-defined scientific cases and acceptance criteria, agents transform repositories into executable environments supporting task generation, execution, and scientific verification. These environments serve as a shared foundation for supervised fine-tuning, reinforcement learning, and evaluation. Using verified interaction trajectories, the team trained the PhAI-IDE model family (72B, 9B, and 4B parameters), which shows improvements on held-out scientific code repair as well as general benchmarks in coding, reasoning, and knowledge, evidencing positive transfer from scientific experience to broader capabilities. Code is open-sourced. Paper: arXiv 2609.19134.

Overview

Field: NLP arXiv: 2609.19134

Abstract (translated)

Scientific code repositories encode decades of human knowledge in executable models, methods, and tools. Yet fragmented toolchains, implicit domain conventions, and specialized correctness criteria make this knowledge difficult to convert into reliable learning experience — a challenge the authors call the scientific experience bottleneck.

ScienceIDE is infrastructure for turning the world's scientific code into programmable environments for scientific agents:

Guided by expert-defined scientific cases and acceptance criteria, agents transform code repositories into executable environments that support task generation, execution, and scientific verification.
These environments provide a shared foundation for supervised fine-tuning, reinforcement learning, and evaluation.
Using verified interaction trajectories, the team trained PhAI-IDE-72B, PhAI-IDE-9B, and PhAI-IDE-4B.
The model family improves on held-out scientific code repair as well as general benchmarks in code, reasoning, and knowledge, providing evidence of positive transfer from scientific experience to broader capabilities.

ScienceIDE lays the foundation for an integrated workspace for agent learning and scientific practice, turning humanity's scientific software into a common substrate for developing scientific intelligence. Code is open-sourced.

Key contributions

Identifies and formalizes the "scientific experience bottleneck"
A pipeline converting arbitrary scientific repositories into verifiable agent environments
The PhAI-IDE model family trained on verified trajectories, with demonstrated generalization

---

*Auto-collected on 2026-09-18*