ScientistOne: Towards Human-Level Autonomous Research via Chain-of-Evidence

arXiv:2605.26340 · 2026-05-28 공개 · arXiv · PDF

benchmarking reproducibility autonomous-research research-agents score-verification reference-verification frontier-research chain-of-evidence

Abstract

Autonomous research agents produce competitive solutions and professional-looking manuscripts, yet their outputs contain verifiability failures undetectable by surface-level evaluation: fabricated citations, unreproducible scores, and method descriptions that diverge from the implementation. We address this through three contributions. First, Chain-of-Evidence (CoE), a verifiability framework requiring every claim to be traceable to its evidence source. Second, ScientistOne, an end-to-end autonomous research system that maintains evidence chains by construction throughout literature review, solution discovery, and paper writing. Third, CoE Audit, a post-hoc audit whose four integrity checks -- score verification, specification violation, reference verification, and method-code alignment -- apply uniformly to all systems. Across 75 papers spanning five systems and five frontier research tasks, every baseline exhibits at least one systematic failure mode: hallucinated reference rates reach 21%, score verification passes in as few as 42% of papers, and method-code alignment ranges from 20% to 80%. ScientistOne achieves zero hallucinated references (0/337), perfect score verification (12/12), and the highest method-code alignment (14/15), while matching or exceeding human expert performance on all five tasks. ScientistOne further generalizes to six additional tasks spanning medical imaging, fine-grained recognition, 3D perception, and language modeling, achieving state-of-the-art on Parameter Golf and gold medals on MLE-Bench tasks where baselines fail entirely.

한국어 요약

📋 한 줄 요약

**[자율 연구 에이전트 / Chain-of-Evidence]** ScientistOne이 모든 claim을 evidence source에 traceable하게 강제하는 CoE 프레임워크로 75 논문·11 task에서 hallucinated reference 0/337, score verification 12/12 달성, 인간 expert 수준·MLE-Bench 금메달.

🎯 핵심 기여도

💡 핵심 아이디어

자율 연구의 신뢰성은 결과물의 외형이 아니라 모든 claim이 evidence까지 거슬러 추적 가능한지에 의해 결정되며, evidence chain을 시스템 설계 단계부터 by construction으로 유지하면 surface-level 검증으로는 잡히지 않는 fabrication·reproducibility failure를 원천 차단할 수 있다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: 자율 연구 시스템의 "looks-good but unverifiable" 문제를 정량 노출, evidence chain을 시스템 설계 원리로 격상, audit가 모든 시스템에 적용 가능한 보편 도구 제공, 인간 expert 수준 도달의 첫 신뢰 가능 사례. **한계**: 11 task가 주로 ML·과학 응용 — 인문·실험과학 일반화는 미검증, evidence chain 유지의 운영 비용, audit의 4 check 외 failure mode 가능성.

🚀 실용적 활용