Identify incorrect denominators, misleading aggregations, inconsistent definitions, data-lineage gaps, workflow bottlenecks, and conclusions not supported by the available evidence. Grade AI-generated responses against structured rubrics for data interpretation, metric consistency, process understanding, feasibility, and communication.