Compounding Research Loops: Automated Follow-On Session Generation from Knowledge Gaps and Unresolved Objections in Multi-Agent AI Deliberation

Multi-agent AI deliberation systems explicitly characterize what they cannot resolve: knowledge gaps where evidence is insufficient to support a finding, and unresolved objections where adversarial challenges could not be definitively rebutted within the session. These structured outputs represent the deliberation’s own map of its limitations—yet in existing systems they have no functional connection to subsequent research. We describe an automated Follow-On Session Generation mechanism that converts knowledge gaps and unresolved objections from a completed deliberation session into structured follow-on session proposals, each comprising a proposed research question derived from the gap or objection’s resolution condition, a priority classification, a recommended session depth, and a prerequisite field. A session lineage tracking mechanism maintains parent-child relationships between sessions, enabling users to navigate the full research history of a topic across multiple deliberation cycles and enabling corpus systems to identify successor session records for comparative calibration analysis. The mechanism creates a compounding research loop in which the limitations of one deliberation automatically drive the inputs of the next—transforming isolated sessions into nodes in a self-deepening research graph.