GLOSSARY, read 10 Nov 2026
36 lineage terms, each with an example from a real tool
- data lineageA record of where data comes from and where it goes.
- table-level lineageLineage that links whole tables, views and models to each other.
- column-level lineageLineage that links individual columns across tables.
- field-level lineageLineage down to a single field, usually a column in a table or a field on a BI dashboard.
- transformation lineageLineage that records how a value was changed on its way, not only where it moved.
- BI lineageLineage that continues past the warehouse into dashboards, reports and workbooks.
- metric lineageLineage that follows a business metric, such as revenue, back to the columns and filters that define it.
- upstreamEverything that feeds a given asset: the tables, columns and jobs that come before it in the flow of data.
- downstreamEverything that depends on a given asset: the tables, dashboards and models that come after it.
- impact analysisChecking what will be affected before a change ships, by walking downstream from the thing you plan to alter.
- root cause analysisWorking backward from a bad number or failed check to the step that caused it.
- lineage graphThe network of assets and the links between them that a lineage tool builds.
- SQL parsingReading the text of SQL statements to work out which columns feed which.
- query log lineageLineage built from the history of queries a warehouse has already run.
- runtime lineageLineage emitted by a job while it runs, describing the inputs and outputs of that specific run.
- design-time lineageLineage worked out from code and configuration before anything runs.
- OpenLineageAn open standard for describing lineage events, hosted by the LF AI & Data Foundation.
- OpenLineage facetAn extension block attached to a run, job or dataset in an OpenLineage event.
- MarquezThe open-source reference implementation of the OpenLineage API.
- run eventA single OpenLineage message saying that a job run started, completed or failed, with its input and output datasets.
- jobIn OpenLineage, a process that reads datasets and writes datasets, such as an Airflow task, a Spark application or a dbt model run.
- dataset (OpenLineage sense)In OpenLineage, any named collection of data a job reads or writes: a table, a file, a topic.
- dbt manifestThe manifest.json file dbt writes after a run, describing every model, test, source and the dependencies between them.
- dbt exposureA dbt declaration that something outside the project, such as a dashboard, depends on certain models.
- ref() in dbtThe dbt function one model uses to point at another, as in ref('model_a').
- ACCESS_HISTORYA Snowflake Account Usage view that records which objects, including columns, each query read and wrote.
- GET_LINEAGEA Snowflake SQL function that returns upstream or downstream lineage for an object.
- Unity Catalog lineageThe lineage Databricks captures for tables registered in Unity Catalog.
- tag propagationCopying a tag, such as PII, from one asset to the assets built from it along lineage links.
- semantic layerA shared set of business definitions, such as metrics and dimensions, that sits between raw tables and the tools that query them.
- data observabilityMonitoring data for freshness, volume, schema changes and quality problems, with alerts when something breaks.
- MCP (Model Context Protocol)An open protocol that lets an AI application call tools and read data from another system through a standard interface.
- persona-scoped contextContext for an agent that is filtered by the role of the person asking.
- metadata-only accessA way of connecting a lineage tool to a warehouse so it reads schemas, query logs and usage but not table contents.
- stitching (cross-system lineage)Joining lineage collected from separate systems into one connected graph, for example warehouse queries, a dbt project and a BI tool.
- lineage freshnessHow soon the lineage graph reflects a change in the real pipeline.
Last reviewed 10 Nov 2026.