Compare the lineage evidence Embrasure collects from direct connectors, artifact imports, and runtime events.
This matrix covers lineage evidence, not every system Embrasure can catalog or query. Embrasure can combine several paths for one asset and keeps their provenance separate.
| Source | How Embrasure reads it | Table lineage | Column lineage |
|---|---|---|---|
| BigQuery | BigQuery catalog and query history, plus the Google Cloud Data Lineage API during connector sync | Native Google lineage links and observed SQL | Google column links when supplied, with SQL-derived mappings as supporting evidence |
| Snowflake | Catalog and query history, plus ACCOUNT_USAGE.ACCESS_HISTORY when available | Observed warehouse SQL | Native ACCESS_HISTORY mappings and SQL-derived mappings |
| Databricks | Unity Catalog, query history, and system.access.column_lineage | Observed SQL and system lineage events | Native Unity Catalog system lineage |
| StarRocks | Catalog metadata and view definitions, plus AuditLoader when installed | Declared views, observed CTAS and insert statements, and exact-location Iceberg bridges | SQL-derived mappings when the statement resolves |
| Fivetran | Fivetran metadata API with source and destination connector bindings (currently Postgres to BigQuery) | Enabled replication mappings | Enabled column mappings |
Google Cloud Data Lineage is part of the BigQuery connector path, not a separate catalog source. Snowflake native column lineage requires ACCESS_HISTORY; Databricks native column lineage requires access to Unity Catalog system tables.
| Producer | Intake path | Table lineage | Column lineage |
|---|---|---|---|
| dbt | manifest.json upload | Declared model dependencies | Compiled SQL when dependencies resolve exactly |
| SQLMesh | Embrasure project export | Environment-aware model dependencies | Rendered SQL when dependencies resolve exactly |
| DataHub | DataHub lineage export | Imported relationships | Fine-grained lineage when included in the export |
| Airflow, Spark, Flink, and custom applications | Standard OpenLineage HTTP events | Runtime inputs and outputs | columnLineage facets when supplied |
| Connector | Catalog metadata comes from | Native lineage feed |
|---|---|---|
| Amazon Athena | AWS Glue Data Catalog | No |
| Amazon Redshift | Redshift system catalog | No |
| PostgreSQL and Supabase | PostgreSQL system catalog | No |
| MySQL and Azure MySQL | MySQL system catalog | No |
| Azure SQL | Azure SQL system catalog | No |
| Azure PostgreSQL | PostgreSQL system catalog | No |
| Azure Cosmos DB | Cosmos DB databases, containers, and fields | No |
These connectors can catalog and query their sources, but they do not currently sync provider-native lineage. Attach dbt, SQLMesh, DataHub, Fivetran, or OpenLineage evidence when you need transformation lineage for their assets.
Runtime and warehouse-native observations carry more authority than declarations or imports. A lower-authority source can fill a gap, but it cannot silently replace stronger contradictory evidence.
Evidence reaches the served graph only after its endpoints resolve to catalog objects inside the source binding. Unresolved evidence stays durable and can resolve after a later catalog sync or identity improvement.