Documentation index: llms.txt. This page is also available as markdown: append .md to this URL or send Accept: text/markdown.
Check Lineage and Impact
A change that looks right inside one Node can still break the Nodes that read from it. Before you accept it, check two directions: what feeds the Node, and what depends on it.
Check What Feeds the Node
As your agent writes files, the graph in Coalesce Desktop redraws. A new {{ ref() }} in a Node's SQL is a new edge in the graph.
- Select the Node in the graph.
- Open the filter menu in the graph controls. It shows the current filter, All by default.
- Choose Upstream.
The graph now shows only the Nodes that feed the selected Node. Check that every source is one you expect. An unexpected upstream Node usually means the agent joined the wrong table or referenced the wrong Storage Location.
Check What Depends on the Node
With the Node still selected, choose Downstream from the same menu. The graph shows every Node that reads from it, whether directly or through other Nodes.
Each one is a Node the change can break. Look harder when the change:
- Renames or removes a column.
- Changes a column's data type.
- Changes the grain, such as adding a join that adds rows.
- Changes a filter, so downstream Nodes see fewer or more rows.
The other filters help too. Related shows upstream and downstream together. Spotlight shows upstream and downstream Nodes up to four levels away. Choose All to clear the filter.
| Filter | Shortcut |
|---|---|
| All | Shift+A |
| Upstream | Shift+U |
| Downstream | Shift+D |
| Related | Shift+R |
| Spotlight | Shift+S |
Every filter except All needs a selected Node.
Trace a Column
Column lineage shows where one column comes from and everywhere it goes.
- On a SQL Node, open the Mapping tab and select the eye button in the column's Lineage cell.
- On a YAML Node, right-click the column in the mapping grid and choose View Column Lineage.
A column lineage tab opens with that column selected. The tab is named after the Node. It shows the column's path through upstream and downstream Nodes. On a SQL Node, a fork icon next to a column means it traces through a CTE. Hover over the column name to see the CTE path.
Use it to answer:
- Is this column built from the source column you expect? A measure built from the wrong source column still compiles.
- Which downstream Nodes use this column? These are the Nodes a rename or type change breaks.
See Column Lineage and Propagation for everything the tab can do.
Confirm Nothing Broke
Lineage shows where to look. These checks prove nothing broke:
-
Run
coa validate. Ask your agent to run it. It scans the Workspace for broken column references, data type mismatches, duplicate names, and circular dependencies. -
Check the Problems panel. Select Problems in the left rail.
-
Apply the change, then validate downstream. Validation compiles against the tables that exist now, so first select Create on the changed Node. Then open each downstream Node and select Validate Select. SQL that refers to a removed or renamed column now fails to compile. See Check the SQL That Will Run.
-
Re-run the chain in development. For a change that alters rows or types, ask your agent to create and run the Node and everything downstream, then compare row counts with the last run:
coa create --include '{ STG_ORDERS }+'coa run --include '{ STG_ORDERS }+'
What's Next?
- Review Checklist for Agent Work
- Column Lineage and Propagation
- The Build Interface for the graph and its controls