A payment processing system failed in production. Investigation revealed that the service had been designed, reviewed, and deployed with no discussion of what would happen if the payment provider’s API was slow. Under normal conditions, the API responded in 80ms. Under load — which happened on the last day of the quarter, when every enterprise customer ran end-of-year billing — it responded in 3000ms. The payment service’s thread pool filled. It became unavailable. Every service that depended on it failed.
The question “What is the failure mode when an external dependency is slow?” is the fourth question in the architecture review checklist. Nobody had asked it.
The seven architecture review questions are the checklist that catches what individual engineers miss. They are ordered by dependency: you cannot fully answer a later question without first answering the earlier ones. Apply them in sequence, and no major failure surface remains unchecked.
The questions were developed from observing what architecture reviews consistently missed. Each question addresses a class of failure that was found in production because it was not asked during design.