It feeds the catalog you already bought.

Publishes to your catalog

One registration, six destinations: OpenMetadata, DataHub, Collibra, Atlan, Apache Atlas, and OpenLineage. Tables, columns, tags, classifications, and lineage flow out as first-class catalog assets, from the admin console or from provisa metadata export on a cron.

The point is not that Provisa has a catalog. It is that the metadata is a byproduct of the thing that already has to be correct — the model that executes and governs your queries — rather than a parallel description maintained by hand and drifting from the moment it is written.

Lineage that survives re-platforming

Assets are stitched on business identity, not physical location. Move a table from Oracle to Postgres and its catalog entry, lineage, and history follow the business entity instead of shattering into a new asset with no past. A migration doesn't reset your catalog to zero.

Data quality, without a second engine

Point Soda at a governed table and register the contract as a source. You get scan history, a quality scorecard, and lineage from each failing check back to the table it guards — with scans triggered by change rather than by a cron guess. Provisa ships no data-quality engine and doesn't ask you to retire yours.

Every value has an address

A governed value is addressable as provisa://<org>/<domain>/<table>/pk=<id>/<column> and resolves over REST or MCP — optionally as of a point in time. Exported documents, lineage nodes, audit records, and AI-agent citations carry links that resolve back to the live governed value, under the same policy as any other read.

Only semantic addresses dereference. Physical URIs are refused, so a citation can never leak a location — which matters most when the thing doing the citing is an agent whose output you cannot review line by line.