One Human, $N$ Agents: Audit-Budget Allocation for LLM Agent Fleets under Miscalibrated, Correlated Confidence

Part of

Machine Learning Model Evaluation and Robustness