Science One Framework: A verifiable autonomous research framework via Chain-of-Evidence

Google Cloud researchers Rui Meng and Tomas Pfister introduced the Science One Framework, an experimental autonomous research prototype that eliminates hallucinations by building verifiable evidence chains via Chain-of-Evidence (CoE), achieving zero phantom references and fully verifiable scores on MLE-Bench and Parameter-Golf while baseline systems hallucinate up to 21% of references.

July 30, 2026 Rui Meng, Research Scientist, and Tomas Pfister, Director, Google Cloud Introducing the Science One Framework, an experimental research prototype designed to eliminate hallucinations by natively building verifiable evidence chains, and CoE Audit, an automated protocol to evaluate the integrity of AI-generated papers. Large language models LLMs are increasingly being deployed not just as coding assistants but as autonomous agents capable of conducting end-to-end scientific research workflows. Recent systems e.g., Sakana’s AI-Scientist https://arxiv.org/pdf/2504.08066 , AutoResearchClaw https://arxiv.org/pdf/2605.20025 , DeepScientist https://arxiv.org/pdf/2509.26603 , AI-Researcher https://arxiv.org/pdf/2505.18705 can review literature, formulate hypotheses, execute experiments and write complete manuscripts that are comparable to human-authored papers. However, as the surface-level quality of these AI-generated manuscripts improves, a critical structural problem has emerged: verifiability. Because current autonomous research pipelines generate text iteratively, errors introduced at any stage are amplified. Some existing systems can generate non-existent citations https://arxiv.org/pdf/2601.18724v1 , exhibit misalignments https://arxiv.org/pdf/2506.01372 between the described methods and the actual code, and report experimental scores that are not fully reproducible https://arxiv.org/pdf/2509.08713v1 from the provided code. In our paper https://arxiv.org/pdf/2605.26340 , we tackle this problem by introducing Chain-of-Evidence CoE , a new verifiability framework for AI-driven research. We instantiate CoE with the Science One Framework, an autonomous research prototype that natively builds and maintains evidence chains, and the CoE Audit, a set of automated evaluation metrics that measures the integrity of AI-generated papers against their underlying code and evidence. Our results show that baseline systems hallucinate up to 21% of their references and frequently misalign their code and text, whereas the Science One Framework achieves zero phantom references and fully verifiable scores while achieving state-of-the-art performance on frontier benchmarks like MLE-Bench https://arxiv.org/pdf/2410.07095 and Parameter-Golf https://github.com/openai/parameter-golf . The CoE is a conceptual framework that defines what makes a research artifact trustworthy, much as ACID https://dl.acm.org/doi/10.1145/289.291 defines what makes a database transaction reliable. Rather than prescribing how to build a research agent, the framework specifies the properties its outputs must have. It follows a single principle with two halves: every claim in a research artifact must carry a recorded evidence chain completeness , and each chain must genuinely support the claim it is attached to correctness . A claim may be a reference, a reported number, a method description, or a conclusion, that must link back to corresponding evidence, such as a peer-reviewed paper, an experimental log line, the code that actually ran, or the results table. A hallucinated reference points to a paper that does not exist. An unreproducible score does not reappear when the code is re-run. A misdescribed method claims one algorithm in the paper while the code implements another. Each is a claim whose chain back to its evidence is broken; the CoE Audit makes these breaks measurable. To demonstrate that verifiable AI research is possible without sacrificing problem-solving performance, we designed the Science One Framework https://arxiv.org/pdf/2605.26340 . Unlike previous agents that generate a paper and try to link facts retroactively, the Science One Framework instantiates the CoE framework by construction through three main modules: To rigorously evaluate the Science One Framework prototype against state-of-the-art baselines e.g., Sakana AI's AI Scientist v2 https://arxiv.org/pdf/2504.08066 , AutoResearchClaw https://arxiv.org/pdf/2605.20025 , DeepScientist https://arxiv.org/pdf/2509.26603 , AI-Researcher https://arxiv.org/pdf/2505.18705 , we developed the CoE Audit. This post-hoc evaluation protocol acts as an automated forensic reviewer, running four strict integrity checks on the generated artifacts paper, solution, code and references : We applied the CoE Audit to 75 papers generated across five systems-optimization tasks Prism https://arxiv.org/abs/2505.04021 , Cloudcast https://dl.acm.org/doi/10.5555/3691825.3691842 , EPLB https://github.com/deepseek-ai/eplb , LLM-SQL https://arxiv.org/abs/2403.05821 , and transaction scheduling https://www.vldb.org/pvldb/vol17/p2694-cheng.pdf from the Automated Design of Research Systems https://arxiv.org/pdf/2510.06189 ADRS benchmark. The Science One Framework significantly outperformed existing baselines in verifiability. The CoE Audit applies the same independent protocol to every system, re-checking each reference against live scholarly databases, and under it, Science One Framework led on all four integrity checks. None of its references were phantom: every one pointed to a real, retrievable paper, compared to hallucination rates as high as 21% in baselines, because the Problem Investigator retrieves every reference rather than generating it from memory. It also achieved perfect score verification and the highest method-code alignment. In contrast, baseline systems frequently described sophisticated algorithms like "hybrid neuro-symbolic solvers" when their submitted code was a simple, deterministic heuristic. Crucially, implementing strict verifiability did not compromise the agent's scientific capabilities. The Science One Framework matched or exceeded https://arxiv.org/pdf/2605.26340 human expert performance on all five ADRS tasks, achieving the best overall score among all systems on two of them Cloudcast and EPLB . To test its generalization, we deployed the Science One Framework on six highly complex external tasks: As autonomous research systems scale to solve increasingly difficult scientific problems, solver quality alone will no longer be enough to differentiate them. What will separate their outputs is whether the resulting research can be trusted. Our findings demonstrate that verifiability must be treated as a first-class architectural constraint. By building evidence chains at the time a claim is produced rather than attempting to reconstruct grounding after the fact, The Science One Framework demonstrates that AI agents can produce rigorous, trustworthy and highly competitive scientific research. We hope the Chain-of-Evidence framework and its audit will serve as valuable tools for the community as we continue to build the next generation of AI scientists. We would like to thank Bhavana Dalvi Mishra, Jiefeng Chen, Chun-Liang Li, Palash Goyal, Mihir Parmar, Yiwen Song, Yale Song, Raj Sinha, Parthasarathy Ranganathan, Burak Gokturk and Jinsung Yoon for their valuable contributions to this work. The Science One Framework is an experimental research prototype, not production-ready tools.