YouRA architecture keeps research claims matched with real evidence
YouRA: A Persistent-State Architecture for Evidence-Traceable Autonomous Research Agents
Artificial Intelligence
Summary
Scientific papers sometimes say things that aren’t fully backed up by the experiments they describe because the computer systems don’t keep track of all the details and failures throughout the whole research process. The authors present YouRA, a system that carefully tracks experiments, evidence, and mistakes over time so that claims in research papers can be traced back to real data. Their system has several parts that manage the research state, separate decision-making from execution, and learn from failures. Tests show YouRA improves over other AI research agents in producing trustworthy results.
What this means in practice
- •For ai research platform developers: Create AI systems that run lengthy research tasks while maintaining verifiable evidence and error histories for more reliable outputs.
- •For automated scientific experiment teams: Track hypotheses and failures in AI-run experiments to ensure findings reported by autonomous agents are backed by real data and clear evidence.
Authors
Yoonkyu Woo, Woojin Lee, Jin-Xia Huang
Abstract
End-to-end research agents can now produce complete scientific papers, yet manuscript claims often diverge from executed experiments. This gap is structural: research state, failure histories, and claim-evidence alignment are not maintained as persistent, verifiable state across long-horizon pipelines. We present YouRA (Your Research Agent), an architecture for stateful, evidence-traceable autonomous research. YouRA preserves research state, execution evidence, and failure history across the research trajectory by integrating three components: a Verification State Architecture (VSA) that tracks hypotheses, gates, and evidence pointers; an Independent Controller that turns state and reflection records into lifecycle, recovery, and debate/review control while separating control from execution; and Stateful Reflection that logs failures as structured lessons and routes recovery through bounded repair, redesign, or reset. On MLR-Bench's predefined ten-task end-to-end subset, YouRA improves over both MLR-Agent and AI Scientist V2 on scalar Overall across all three matched backbones. An automated diagnostic using MLR-Bench's hallucination taxonomy reports intersection/union counts for four fact-based failure types, and data-provenance diagnostic shows more real-data-based outputs. Ablating each of the four components (the VSA, the Independent Controller, MCP tool access, and reflection-guided recovery) supports their separable contributions. Removing either core-state component drops YouRA below the full system. Code: https://github.com/PrayPrey/Your-Research-Agent.