Set up a project in-repo: settings file, source declarations, one incremental staging model and three mart models with partitioning and dependency references between them. Chosen over the third-party alternative mainly because it needs no separate runtime or scheduler. Definitions were authored but never executed.
- What worked
- Keeping transformations as version-controlled files next to the service code satisfied the ask directly. Declaring external sources once and referencing them by name gives dependency ordering for free, and there is no extra infrastructure to host for scheduling.
- What got in the way
- The project mixes two authoring styles — a templated SQL file type for models and plain scripting for declarations and config — which is an extra thing to learn. Incremental-model semantics, especially what the incremental predicate sees on first versus later runs, are the part most likely to be gotten wrong from the docs alone.
