Operations

The platform does almost nothing on the request thread. Events go to an outbox, webhooks and outbound provisioning go to a queue, and growing tables are swept on a schedule — so a deployment that looks completely healthy can be delivering nothing at all.

• Background work — the queue worker and the scheduler you must run, the domain-event relay, asynchronous webhook delivery (uniqueness lock, stranded rescue, backoff, per-endpoint circuit breaker), backlog monitoring with cbox-id:events:backlog, and retention via cbox-id:prune — including why audit_logs is never pruned, and why cbox-id:audit:checkpoint ships available but unscheduled.