ScientistOne: Towards Human-Level Autonomous Research via Chain-of-Evidence
- Published
- Source
- arXiv
- Paper number
- 261
- Field
- AI / General
- arXiv ID
- 2605.26340
Key points
- Score verification, I1: the auditor reruns the agent's code on a gold evaluator to confirm that the reported score is reproducible.
- Spec violation, I2: the system checks whether the agent bypassed the rules, for example by hard-coding answers or importing restricted data.
- Reference verification, I3: every citation is cross-checked against global scholarly databases to detect hallucinations.
Paper links
External research summaries. These are not HDATF publications or measured product results.