'Clear data lineage' is the line in here that does the most work, and it's the one most pipelines treat as a catalog field instead of a property of every record. The version we settled on for a house: each entry carries its source, an evidence grade, and a verification state, and a suspect entry is demoted rather than deleted — the same instinct as 'metadata stays available even if the data is removed.' An agent reading that record can see how each value got its standing, which matters more for agentic use than whether the value is currently clean.