Data Lineage

Definition

Data lineage is the recorded path a piece of data took: where it originated, which steps transformed it, and which outputs depend on it. With lineage you can trace a number back to its sources, and forward to everything it affects.

In plain terms

Lineage answers three questions about any figure:

1. Where did it come from? The original sources.
2. What happened to it on the way? Filters, joins, corrections, calculations.
3. What relies on it? The reports, models and decisions built on top.

The first two let you judge whether to trust a number. The third tells you what to re-check when something upstream changes. The related word provenance is often used for the same idea, especially for documents and research artifacts.

Example

The chain below is synthetic.

If the duplicate rule changes, lineage shows at once which downstream figures need re-checking. Without it, someone has to rediscover the chain by hand, usually after a number has already been questioned.

Common misunderstandings

• “Lineage is a diagram.” A diagram drawn once goes stale. Useful lineage is captured automatically as the work runs.
• “Lineage proves the data is right.” It shows the path, not the quality. A perfectly documented path can still carry a wrong number.
• “Table-level lineage is enough.” Often it is not. When one column is wrong, you need to know which outputs used that column.