dbt Core Contracts and Catalog Metadata Drift

A model contract can pass while the catalog story slowly drifts away from the system people actually use.

Contracts protect the model boundary

dbt model contracts help teams define expected columns and data types for a model. That is valuable because it turns part of the model interface into something that can be checked.

Catalog metadata drift starts when the contract, documentation, source freshness, exposures, and external catalog no longer tell the same story. The model may still build. The product meaning may still be wrong.

Metadata drift is a product problem

Drift shows up when owners change but metadata does not, when source freshness rules lag behind new dependencies, when exposures no longer represent real consumers, or when an open catalog sees tables but not the semantic contract around them.

The fix is not another documentation sprint. The fix is to connect dbt contract checks, catalog metadata, lineage, and product review into one operating loop.

The ODI pattern closes the metadata loop

Open Data Infrastructure depends on metadata that travels. dbt can own transformation intent. The catalog can own discovery and policy context. Lineage can connect producers and consumers. The product owner has to review the combined picture.

For adjacent context, read dbt model versions and open data contracts, dbt Core source freshness, and MetricFlow and open catalog semantics.

What breaks first

• The model contract passes, but catalog owners are stale.
• Source freshness checks exist for pipelines but not for product SLAs.
• Exposures list dashboards but miss agent tools or APIs.
• The open catalog receives table metadata without the transformation contract that explains it.

Questions to ask

Ask which metadata fields are contractual, which are documentation, and which are operational evidence. Ask how often dbt state, catalog state, and product-owner state are compared.

Sources to start with

These primary sources anchor the technical claims in this guide.

• dbt model contracts documentation
• dbt contract configuration reference
• dbt source freshness documentation
• DataHub lineage documentation

The contract is only as durable as the metadata loop around it.