Open Data Infrastructure
dbt Core Semantic Layer and Open Catalog Boundaries
Where dbt semantic definitions help and where open catalogs still need to own metadata, policy, and control.
A semantic layer can define a metric. It should not pretend to own every boundary around the data.
The practical problem
dbt semantic models and metrics help teams define business meaning as code. That is valuable. A metric definition with entities, dimensions, measures, and versioned project context is a big improvement over everyone recreating revenue in a dashboard.
The mistake is treating the semantic layer as the whole governance layer. Open Data Infrastructure still needs a catalog boundary for table identity, access control, lineage, storage ownership, and cross-engine interoperability.
Semantic definitions solve a real problem
The dbt Semantic Layer centralizes metric definitions on top of models and exposes them to downstream tools. Semantic models define entities, dimensions, and measures. That helps prevent business logic from scattering across BI tools, notebooks, and agents.
For analytics teams, this is a strong application-plane pattern. It gives consumers a governed way to ask for metrics without rewriting SQL every time.
Core idea: semantic layers define business meaning. Open catalogs define the governed infrastructure boundary around the data.
The catalog still owns infrastructure truth
The open catalog has a different job. It needs to expose table identity, schema, snapshots, permissions, lineage hooks, storage credentials, and engine-neutral metadata. A metric can reference a model, but the catalog should still know which table exists, who owns it, and which engines can safely use it.
This distinction matters for AI systems. An agent asking for a metric needs semantic meaning, but it also needs policy, freshness, lineage, and source identity. That is why semantic layers and context graphs should connect rather than collapse into one tool.
What breaks first
- Metrics are governed, but the source table permissions are copied manually across engines.
- Semantic definitions move with dbt, but catalog metadata remains trapped in one platform.
- AI tools receive metric names without source freshness or policy context.
- Lineage stops at model dependencies and misses access decisions or table state.
Questions to ask
- Which layer owns metric meaning?
- Which layer owns table identity and access?
- How do semantic models connect to catalog metadata?
- Can an agent retrieve metric meaning and governance context together?
For adjacent articles, read dbt Core Model Contracts and Open Catalogs, dbt Core in the ODI Stack, and The Context Graph.
Sources to start with
These primary sources anchor the technical claims in this guide.
Meaning belongs in the semantic layer, but control still needs the catalog.