Clinical attribute registry

A public audit of 375 cBioPortal studies found the same clinical concept encoded dozens of ways, 24 distinct encodings of radiation therapy alone, and the tooling to prevent it does not exist: MSK's Clinical Data Dictionary is explicitly advisory, and validateData.py validates one study at a time, never consistency across studies. This recipe builds the missing piece: a governed clinical attribute dictionary your institution owns. Seed it from your best-curated study's headers, refine it (datatypes, descriptions, display names, priorities, allowed values as taxonomies), bind attributes to NCIt terms via ontology suggestions a human confirms, then audit every other study's headers against the same dictionary, so each divergent id, retype, or missing description surfaces as a named finding instead of an archaeology project. Headers only: no patient data ever leaves your site.