Entity Registry
The Entity Registry is the catalogue of what SEMANTIX knows about your data. It has two halves: entities (tables and their columns) and metrics (defined, reusable calculations).
If it is not in the Entity Registry, Chat cannot ground an answer in it.
Entities
Section titled “Entities”An entity is a registered table. It carries:
- its schema and table name, and the source it came from;
- a description — the highest-value field on the page;
- its columns, each with a type, a description, and a role;
- its trust level (0–5), which is a human judgement;
- its lineage back to the connector it was introspected from.
Column roles
Section titled “Column roles”Each column can be marked as a dimension, a measure, or time. These
are expectations about how the column should be used, and they materially reduce
plausible-but-wrong groupings — telling the agent that order_date is a time
column is what stops it being summed.
Finding things
Section titled “Finding things”Search covers entities, and you can filter by certification and by trust level. The admin view adds a filter for which data source an entity came from.
Data Explorer
Section titled “Data Explorer”An entity’s page is also where the Data Explorer lives: its columns with optional sample values, and a read-only SQL playground scoped to that entity.
Metrics
Section titled “Metrics”A metric is a defined calculation with a label, a description, the query behind it, a run history, and a certification state.
Certified means reviewed and cleared for decisions. Uncertified metrics are visible but carry no claim on anyone’s trust. The percentage shown on a metric is its system trust score — machine-computed and distinct from the human-assigned entity trust.
Certifying
Section titled “Certifying”Only a trusted metric is worth watching on a schedule, so certification is the gate between “a query someone liked” and “a number the team relies on”.
Certification propagates upward: an entity with a high enough trust level counts as certified automatically, so governing a source table promotes the metrics built on it. See grounding in practice.
Lineage and staleness
Section titled “Lineage and staleness”Entities record when they were last introspected. If a source table changes shape, SEMANTIX can mark an entity as needing review, and admins can prune entities that have gone away. Provenance is what lets you trust a description — you can always see which source and when.
Suggesting a metric
Section titled “Suggesting a metric”If a number you need is not in the Registry, you do not need an admin to write it for you. Write the query in the Data Explorer and Request metric — an admin reviews it and, if approved, it becomes a Registry metric.