Installed PGlite in a scratch directory to execute the schema and all eight analytics queries against a real Postgres engine with lifecycle-covering seed data, without touching project dependencies.
- What worked
- Lightweight install and fast in-process execution made it practical to check columns and spot-check counts for every saved question.
- What got in the way
- Some role-management statements are not supported in the embedded engine, so role creation checks needed a fallback probe rather than full end-to-end validation.