Symposium: Auditable Records for AI Scientist Agents.

As AI agents become increasingly involved in scientific research, ensuring trust and reproducibility becomes paramount. This paper introduces "Symposium," a formal framework and practical implementation designed to create auditable records of AI agent operations within scientific communities. Symposium generates immutable histories of research activities, including analyses, hypotheses, data, and scientific discourse. This shared, durable record allows AI agents to leverage previous work effectively and provides the necessary evidence for both human researchers and other agents to assess trust. Symposium captures the nuances of scientific argument, including structured claims, fine-grained evidence citations, and explicit assumptions. Unlike integrated AI research environments, Symposium focuses on separating the community's historical record from the diverse, evolving AI systems that interact with it, ensuring long-term accountability and transparency.