How do you track data lineage, and why does it matter operationally?

Lineage maps which sources feed which tables, models, and dashboards. Capture it automatically by parsing SQL and orchestrator dependencies rather than maintaining it by hand, since manual lineage rots within weeks. It matters because it answers the three urgent questions: what breaks if this source fails, who do I notify about a bad load, and can I safely delete or change this column.

Lineage is the dependency graph of your data, ideally down to column level, and the only version worth having is the one generated automatically. Anything hand maintained is wrong within a month. In practice we got most of it free from dbt's manifest for the transformation layer, plus the BI tool's metadata for dashboards, feeding a catalog. The value shows up in three moments. When a source is late, I can immediately tell which dashboards will be stale and message those owners before they message me. When someone wants to drop a column, impact analysis takes thirty seconds instead of a grep and a prayer. And for deletion requests, I can find every table containing a given identifier. The thing I would add is that lineage without ownership is only half useful. Every dataset needs a named owning team attached, otherwise you know exactly what broke and still have nobody to tell.