dbt Core MetricFlow and Open Catalog Semantics

A metric that only works inside one tool is not a business definition. It is a product feature with a business name.

Metrics are contracts

dbt documentation describes MetricFlow and semantic models as part of defining and querying metrics. That makes metrics more portable than scattered dashboard formulas, but the metric still needs ownership, lineage, policy context, and compatibility rules.

A metric has a name, calculation, grain, dimensions, filters, freshness expectations, and owner. If those pieces do not travel with catalog metadata, the metric can move across tools while losing the meaning that made it useful.

Semantic models need catalog context

The semantic layer should not float above the data platform like a separate universe. It should connect to source tables, transformations, lineage, quality checks, and access policy. That connection lets teams understand not only how a metric is calculated, but whether the underlying data is allowed and trustworthy for a given use.

This matters more when AI systems ask metric questions. An agent needs the definition, the allowed dimensions, the freshness state, and the data product owner before it repeats an answer.

Open catalogs need meaning, not labels

Open Data Infrastructure should connect dbt semantic definitions to catalog entries, lineage, and data contracts. A catalog should expose metric ownership and meaning, not only table names. A contract should state which changes break downstream consumers.

For adjacent context, read dbt semantic layer and open catalog boundaries, semantic contracts in ODI, and semantic layers to context graphs.

What breaks first

• A metric name is shared across tools, but filters and grain differ.
• Metric owners approve business wording but not lineage or policy impact.
• Catalog search finds the table but not the semantic model that defines use.
• AI answers cite a metric without checking freshness or allowed dimensions.

Questions to ask

Ask where the metric definition lives, who owns compatibility, and how the catalog exposes lineage and policy. Ask whether an agent can retrieve metric meaning without bypassing the semantic layer.

Sources to start with

These primary sources anchor the technical claims in this guide.

• dbt MetricFlow documentation
• dbt semantic models documentation
• dbt metrics documentation
• OpenLineage object model documentation

A metric becomes infrastructure when its meaning can survive the tool boundary.