Human-Mediated Cross-Model Deliberation: A Jury Protocol for AI-Assisted Research

We describe a protocol for structured multi-model inquiry — human-mediated cross-model deliberation — and release, to our knowledge, the first public dataset of multi-model deliberation transcripts with claim-level provenance: verbatim jury transcripts, a consolidated verdict, and take/reject lists from a formal deliberation round in which four independent commercial AI systems were consulted separately on the same research question, with a human operator and a coordinator agent ferrying all material between them and no direct AI-to-AI contact. The protocol's components (multi-agent debate, AI juries, human-in-the-loop orchestration) are individually precedented; the contribution is the assembly as a persistent, documented institution, and the dataset as a public research artifact.