Guides agents through profiling, validating, and documenting the quality of person-level and health-adjacent datasets before any analysis or modeling. Use when loading any new dataset, merging tables from different sources, inheriting data collected by someone else, or whenever results look suspiciously good or bad. Use even when the data "looks clean" — health-adjacent data hides impossible values, unit mix-ups, sentinel codes, and duplicate people that silently corrupt every downstream conclusion.
