System
Design Reference Scenarios

Architecture for AI systems that run *where the data is* — on phones, vehicles, sensors, browsers, and embedded devices — under constraints that do not exist in the cloud, with lifecycles that the cloud's deployment patterns do not address.

Architecture for systems organised around events — facts about things that happened, propagated through subscribers who never need to know each other, ordered or partitioned in ways that the architecture chooses deliberately.

Architecture for systems that must keep running when components fail and recover when entire regions, datacentres, or providers do. High availability is the day-to-day discipline; disaster recovery is the once-a-year discipline; both require explicit design that production traffic never reveals until it is too late to add.

Architecture for systems that absorb growth — more users, more data, more traffic, more concurrent operations — without rewriting from the foundations every time the load doubles. Scalability is not a feature added at the end; it is a set of decisions made early about where bottlenecks are allowed to form, how they are detected, and what is done when they do.