Recovering Pre-Event Crowd Beliefs in Prediction Markets with a Risk-Calibrated Mixture of Inverse Experts

Develops APD-Safe, a method for reconstructing pre-event crowd beliefs from trader-linked market histories using multiple inverse models and a risk-based selection procedure. Evaluates the method on unseen simulated mechanisms and against independently recorded survey beliefs across 98 real prediction markets. Single-author paper accepted to the 7th ACM International Conference on AI in Finance (ICAIF 2026). This upload is the author-identified manuscript, not the final ACM version of record.