Catalog.understandData Catalog & Metadata Intelligence

It needs read-only connections, nothing more, and it works with metadata only: structure and profiles are read in place, and none of your data is copied out or retained. Deployment supports BYOK and air-gapped setups, so it runs inside the controls you already have.

Discovery, rescans, and status computation run on their own, so the daily work is review and decision rather than data entry. Stewards approve term links, definitions, and exceptions through the decision door, with maker-checker where you require it, and owners attest on cycle with second-line review. The catalog stays current because rescans are scheduled, not because someone remembers to update it.

For any asset, an evidence pack shows when it was certified, on what evidence, and who attested, with every decision sealed on a hash-chained ledger. Definitions, mappings, and policies are versioned, so you can show which rule was in force on any reporting date. Coverage is reported honestly, so an unscanned system appears as a named gap you can speak to rather than a surprise an examiner finds.

Agents propose and humans decide: suggested term links, flagged drift, and glossary updates land as claims with their evidence attached, and nothing moves until a steward accepts them. Agents cannot approve anything and they cannot grant certification, because approvals belong to human stewards and status is computed from evidence alone. When a proposal is wrong, the steward rejects it and nothing in the record changes.