Entity Registry

The admin Entity Registry is where the catalogue is actually made trustworthy. Analysts read the Registry; admins build it.

Introspect a source

Open a connector in the Entity Registry and run Introspect. SEMANTIX reads the database catalogue and registers what it finds as entities and columns.

Introspection does not overwrite your work: descriptions you have written and trust levels you have assigned survive a re-introspect, so you can re-run it safely after a schema change.

Caution

Introspection is a real read of the source. On a large warehouse, point it at the databases you care about rather than the whole instance — and remember it uses the same read-only path as everything else.

The bulk trust action

Trusting one table at a time is tedious. Select entities and apply a trust level to all of them at once. This is how you get a newly introspected source to a usable state quickly, and then spend your curation effort on the tables that actually matter.

Entity trust

Each entity carries a trust level from 0 to 5 — a human judgement of how much you believe the descriptions and structure. The full scale, and what each level implies about the metrics built on it, is in the entity trust reference.

Trust is separate from the system trust percentage shown on metrics, which is machine-computed. They answer different questions and can disagree.

Certification

Certified entities are treated as automatically certified when a metric is built on them, which means their metrics inherit certification rather than needing their own review. Raising trust on a source table is therefore the fastest way to lift a family of metrics.

Metric certification

Metrics are certified individually. Approving a pending metric request certifies the metric it creates — that is the review step that says someone looked at the SQL and agreed with it.

Pruning and review

Entities are marked for review when a re-introspect suggests the source has changed underneath them, and stale entities can be pruned. A description that no longer matches its table is worse than no description, so pruning matters as much as adding.

Column roles

Set each column’s role — dimension, measure, or time — during curation. This is the cheapest accuracy improvement available: a time column marked as time is not aggregated, and a measure marked as a measure is.