Drawing queues before choosing brokers

Teams often reverse the decision stack: they pick tooling, then retrofit semantics. In our cohorts we flip that by spending the first night on paper timelines—who emits, who acknowledges, and where idempotency keys live. The exercise is deliberately low-tech so nobody hides behind vendor wizards. When the diagram stabilizes, mapping to Kafka topics or AMQP queues becomes clerical. The hard part—deciding who owns retries and poison handling—is already visible. We also annotate cloud cost ops signals early: fan-out that looks innocent on a whiteboard can explode spend under burst traffic. Finally, we ask teams to write a one-page degradation story. If the broker slows, what user-visible behavior changes? That narrative keeps operational discussions grounded. The post mirrors what we teach in Messaging Choreography Lab without promising a single silver-bullet topology.