Ground Truth

Ground truth refers to the trusted information considered correct for training and evaluating a model. It may come from human annotation, reliable measurement systems, or validated enterprise records. If ground truth quality is poor, model evaluation becomes fundamentally unreliable. For that reason, the critical question is not simply whether labels exist, but whether those labels are actually correct. In high-stakes domains, producing trustworthy ground truth is a specialized process in itself.