The completeness of ground-truth labels in evaluation data—incomplete annotations can hide relevant evidence and artificially lower reported retrieval scores.