Build the system behind a ticket drop from parts, press Run, and see whether it can hold seats, show which seats are free and confirm payments when 8,000 requests a second hit it at ten o'clock, and whether it keeps going when an app server, the seat-map cache or the payment worker dies.
A simplified model, not a real cloud. Times are in model milliseconds, the cache is always full of useful answers (warm), and prices are made up to compare designs. The board counts capacity and waiting, not correctness: it cannot show two buyers getting the same seat, because it never looks at which seat a request asks for. About 1 admitted request in 8 becomes a payment (chapter 25.2: half the admitted fans buy and each makes 4 requests), so the queue and the worker see a small share of the traffic. The waiting room shows as over 100% busy because it turns people away on purpose.