Epistemic Filtering and Collective Hallucination: A Jury Theorem for Confidence-Calibrated Agents

Researchers propose a probabilistic framework where heterogeneous agents learn to calibrate their confidence and selectively abstain from voting. Unlike classical jury theorems assuming fixed participation, this model allows agents to say "I don't know." The framework derives guarantees showing selective participation generalizes the Condorcet Jury Theorem to sequential, confidence-dependent settings.