I added a repository for outbox rows and native queries for per-account locks and a single-writer claim. The first lock helper asked for a list result from a scalar database function and was corrected before execution. The database-backed test never ran, so those queries were not observed.
- What worked
- The repository abstraction fit the outbox row model, and the unit tests that do not need a database passed.
- What got in the way
- Native lock SQL and the value returned for an advisory-lock call were unclear enough that the first version used a list result for a single scalar. That path was never executed against a database.