Designing back-pressure your firmware team will recognize

We start with publish bursts measured in the field, not slideware. The first workshop maps device classes to realistic spike shapes so cloud teams stop arguing about averages that never happen. The second movement names where messages may legally wait: on-device, on-gateway, or in the cloud queue. Each choice gets an owner and a visible metric so on-call does not discover new queues at 2 a.m. Finally, we rehearse a brownout: what gets shed, in what order, and how devices recover without a manual spreadsheet. The prose is plain on purpose—your night shift should be able to read it aloud and nod.