Open Data Infrastructure
dbt Core Exposures and Open Metadata
Where dbt exposures help map downstream use and where open metadata systems still need to own context.
Downstream use is where data models stop being abstract. A model does not merely exist. Someone depends on it, usually in a dashboard, notebook, service, or agent workflow that will break at the worst possible time.
The practical problem
dbt exposures let teams define downstream uses of dbt resources in the DAG, such as dashboards, notebooks, machine learning applications, and other assets. That is useful because it brings consumer context closer to model development.
The open metadata question is what happens outside the dbt project. A real platform also has catalogs, warehouses, query engines, policies, lineage tools, BI systems, APIs, and AI consumers. Exposures help, but they do not become the entire metadata graph.
Exposures should feed the graph
A healthy pattern treats exposures as structured signals. They can identify consumers, owners, maturity, URLs, and dependency paths. Open metadata systems can then connect those signals to cross-engine lineage, data product contracts, policy context, usage, and operational health.
That distinction matters. dbt can tell you a dashboard depends on a model. The broader metadata layer should tell you which table version, policy state, freshness promise, and upstream source shaped the data that dashboard consumed.
Core idea: dbt exposures map downstream use. Open metadata turns that use into platform context.
The ODI boundary
Open Data Infrastructure should not trap consumer context inside one transformation tool. dbt exposures should be portable signals that catalog, lineage, and governance systems can use.
This is especially important as BI consumers and agent consumers start sharing the same data products. The platform needs one consumer map, not one map for analytics and another improvised map for AI.
What breaks first
- Exposures are defined once and then drift away from real downstream use.
- The dbt DAG shows a consumer, but the catalog cannot connect that consumer to policy or freshness.
- Agent tools consume the same data product but never appear in exposure metadata.
- Lineage stops at transformation code and misses serving, API, or retrieval paths.
Questions to ask
Ask which downstream uses are represented as exposures, which are missing, and how exposure metadata flows into the open catalog. Ask whether consumer impact analysis includes BI, services, models, and agents.
For related context, read dbt Core semantic layer and open catalog boundaries, metadata as the infrastructure layer, and context graphs for data product discovery.
Sources to start with
These primary sources anchor the technical claims in this guide.
- dbt exposures documentation
- dbt semantic models documentation
- DataHub documentation
- OpenLineage documentation
The exposure is the clue. The open metadata graph is where the clue becomes useful.