Six months from now, a new developer opens your project.
They see an outbox table. A relay process. Sync services in an async-capable Django. A ForeignKey to the user model in a codebase that forbids cross-context dependencies.
New developer: “Why is this so complicated? Let’s just publish events after save.”
They delete the outbox. Tests pass. The deploy succeeds.
You can read this article for free via this link: Click Here 👈
Three weeks later:
Finance: "Revenue in our reports doesn't match Stripe." Support: "Customers paid, but never got a confirmation email." You: "...who removed the outbox?" New dev: "Nobody told me why it was there."
They weren’t careless. They were missing one thing: the reason.
In the last article, we made a lot of decisions:
→ Sync application services, async only at the edges → Events written to an outbox, in the same transaction → Repositories for writes, plain ORM queries for reads → Contexts talk only through api.py → One allowed exception: ForeignKey(settings.AUTH_USER_MODEL)
Two of these rules run on trust. “Never edit, supersede” is a habit, and a PR that rewrites an accepted decision still passes CI. And “revisit when” usually happens outside the repo, where nobody’s watching. Who notices that a premise has changed before the agent acts on a stale “Accepted”?
