Typed schema and generated SQL migrations worked well for a Postgres app, but migrations made on parallel branches needed manual repair of the journal and snapshot links.
- What worked
- The generated SQL was readable, the typed queries caught mistakes early, and applying a migration to a hosted database was one short script.
- What got in the way
- The migrator silently skips a journal entry whose timestamp is older than the last applied one, so a branch migration can look applied when it was not. Fixing that meant rewriting timestamps and relinking snapshot parents by hand.