T3.1Interactive artifact · P3 · Data quality & lineage
Column-level data lineage viewer
An interactive lineage graph of a small analytics stack: five sources, three staging models, four marts and four consumers. Click a table to trace its upstream sources and downstream impact, or pick a column to follow it on its own.
- Type
- Data tool
- Drawn
- Runs
- In your browser, nothing to install
- Tags
- data-engineering · lineage · data-quality · interactive
Detail A · Reading the drawing
What it shows
Data lineage answers two questions: where did this number come from, and what breaks if I change this table? Table-level lineage shows which models depend on each other. Column-level lineage goes a step further and follows a single field, such as revenue, from the Postgres orders table through staging and the fact table into the dashboards that read it.
The graph is laid out like a dbt project: sources feed staging models, staging feeds dimensional marts, and marts feed dashboards, a feature store and a CDP export. Tracing a table highlights its full upstream and downstream path, which is the impact analysis you want before renaming a column or changing a model's grain.
Detail B · Procedure
How to use it
- Click any table to pin it. Its full upstream and downstream path is traced and everything else fades.
- Read the detail panel for the table's columns, freshness and row count, plus its direct upstream and downstream tables.
- Click a column to trace that column alone through the graph.
- Drag tables to rearrange them, scroll to zoom, and use find to jump to a table by name.
- Press Escape, or reset, to clear the trace.
Detail C · Glossary
Key concepts
- Upstream
- Everything a table is built from. If an upstream table is late or wrong, so is this one.
- Downstream (impact)
- Everything built from a table: the blast radius of a change to it.
- Column-level lineage
- Lineage tracked per field rather than per table, so you can see which consumers read a specific column.
- Staging model
- A thin cleaning layer over one raw source: casting, renaming and deduplication, one-to-one with the source.
- Mart
- A modelled fact or dimension table, shaped for consumers such as BI and machine learning.
Detail D · Cited by
Related writing
Posts that use this drawing as a worked example.
- D3.1Column-level data lineage in practice
Table-level lineage tells you two models are connected. Column-level lineage tells you that changing one field breaks exactly these three downstream columns and nothing else. That precision is the difference between a confident change and a prayer.