Data Lineage: The Family Tree of Your Data

Upstream and Downstream

If the Catalog tells you what the data is, Lineage tells you where it came from and where it is going. It represents your entire data ecosystem as a directed graph — nodes are data assets (tables, files, dashboards, ML models), and edges are transformations (ETL jobs, SQL queries, API calls).

Every node in this graph has two directions:

• Upstream (Provenance): Tracing backward to find the origin. "This dashboard shows $4.2M revenue — where does that number actually come from?" You follow the arrows backward: Dashboard ← revenue_summary view ← orders_cleaned table ← raw_orders table ← Stripe API ingest job.
• Downstream (Impact Analysis): Tracing forward to find what depends on something. "If I rename the column cust_id to customer_id in the users table, what breaks?" You follow the arrows forward and discover it feeds 12 downstream tables, 3 dashboards, and 2 ML feature pipelines.

Debugging with Lineage

Lineage is a superpower for root-cause analysis. Consider this scenario:

• Symptom: The CEO's revenue dashboard shows $0 at 9 AM.
• Without Lineage: You inspect the dashboard SQL — it looks fine. You check the revenue_summary view — it's empty. You now have to manually trace through 50 ETL scripts to find which one failed. This takes hours.
• With Lineage: You open the lineage graph for revenue_summary. You see the node is highlighted red (failed). You click "trace upstream" and see the failure originated in Payment_Service_Ingest_Job, which failed at 3:14 AM due to a Stripe API timeout. Root cause identified in under 60 seconds.

The graph makes the invisible visible. Instead of reasoning about your pipeline from memory or documentation (which is always out of date), you reason from the actual execution record.

Column-level lineage takes this further. Instead of tracking which tables depend on each other, it tracks which columns flow into which other columns. This is critical for compliance: you can answer "show me every place the email column is used, transformed, or copied across the entire warehouse."