Variance-Bounded Evaluation of Entity-Centric AI Systems Without Ground Truth: Theory and Measurement

Kaihua Ding

Reliable evaluation of AI systems remains a fundamental challenge when ground truth labels are unavailable, particularly for systems generating natural language outputs like AI cha…