Curation and the staged pipeline.

Curation is the difference between an archive and a dataset. In Radial it is a staged pipeline with one rule throughout: Radial proposes, you confirm. Nothing enters the governed database silently.

Imports land in a staging area

Incoming studies queue in a staging area rather than dropping straight into your data. Duplicates and re-exports are detected and reconciled there, so the same plan sent twice becomes one case, not two. A conflicting re-import becomes a decision, not silent corruption.

Propose, then confirm

For each case, Radial proposes the decisions curation actually consists of: a TG-263 name for every structure and a cohort assignment for the case, each with the evidence for the proposal. High-confidence proposals confirm automatically, recorded like the ones you confirm by hand; the rest queue for review. Confirmations aren't just accepted: they teach your deployment's dictionary, so the queue gets shorter as the months pass.

Audit catches what confidence missed

Everything below confidence is flagged, and a structure audit keeps watching for the problems that surface later: unmapped structures, naming drift, and cases whose assignments no longer fit. The audit is a queue you work, not a report you file.

Provenance is the product

Every name and assignment in the governed database records how it got there: proposed on what evidence, confirmed by whom. The metrics computed from them inherit that provenance. That's what makes the downstream numbers defensible: when a reviewer asks why a case is in the cohort, the answer is on the case.