Lineage and Impact Analysis

Lineage and Impact Analysis

Track where every value came from and which downstream consumers depend on it.

Why it matters

Lineage answers 'if I change / break this upstream column, what blows up downstream?' Tools (OpenLineage, dbt, Atlas) parse SQL to auto-extract it.

Going deeper

Three lineage layers worth distinguishing:

• Static lineage — parsed from SQL / dbt DAGs. Cheap, complete, but blind to runtime branching (CASE WHEN, dynamic SQL).
• Runtime lineage — emitted by the engine (OpenLineage events from Airflow, Spark, dbt). Captures what actually ran.
• BI / consumer lineage — extends lineage past the warehouse into Looker, Tableau, ML feature stores. The hardest tier, and the one that closes the loop.

Lineage is the supply-chain receipt for a number on a dashboard.

Imagine a chocolate bar in a supermarket. The packaging tells you the brand; the receipt tells you the shop. Lineage is the bill of materials underneath — which cocoa farm, which roasting batch, which factory line, which truck. When a contamination alert fires for one cocoa batch, the supply chain instantly knows every bar that contains it. Column-level data lineage does the same: when an upstream currency_rate column changes meaning, you can list every dashboard, alert, and ML feature that consumed it.