# Ground Truth Annotation Ground-truth annotation records a human judgment that can be used to evaluate or supervise a later computational system. Atomic Markdown contains author-declared semantic boundaries and, in transformational instances, paired contextual and autonomous realizations of the same proposition. Those pairs were not created as machine-learning data, but they now provide unusually direct human supervision for boundary selection and faithful decontextualization. Related: [[wiki/Faithful Decontextualization|Faithful Decontextualization]], [[wiki/Semantic Annotation|Semantic Annotation]], [[wiki/Semantic Retrieval|Semantic Retrieval]], [[wiki/Retrieval-Augmented Generation|Retrieval-Augmented Generation]].