Each card below is a real RippleProof run against a demo warehouse, recorded in full and committed to this repository. You can read the lineage it was given, the repairs it wrote, the builds it ran, and the exact points where it stopped and refused. They differ on purpose: some propagate the rename downstream, some preserve the downstream names, one of them fails its build.
The same graph behaves differently depending on whether the rename is carried downstream. See preserve and propagate side by side
The complete campaign. A human resolved the ambiguous model, every repair was built in an isolated sandbox, the migration was applied and rolled back in a disposable PostgreSQL, and review-only pull requests were opened.
The same five-asset graph, but the dbt build did not pass. The campaign stopped instead of publishing, and the unproven assets went to manual review.
In preserve mode the downstream column names stay as they are, so assets covered by a data contract are protected rather than rewritten.
A three-asset graph with the rename carried downstream, captured before the project was renamed and still bound to its manifest under the old namespace.
The smallest run in the corpus: three assets, preserve mode, one repaired project. Useful as the floor to compare the richer runs against.