Ground-truth reliability in AIED evaluation asks whether the human labels used to judge AI are themselves trustworthy: ground-truth-reliability-aied, vetting-dual-llm-safety-education, and data-annotations-pedagogical-hints show expert disagreement and noisy annotations that can mislead benchmark conclusions (ai-ed-evaluation, benchmark).