The 50,000-person example also needs the width of the arrival burst before choosing an architecture. Fifty thousand booking requests arriving together is a different workload from the same number spread across ten minutes, which averages about 83 requests per second. The audience size alone does not describe the pressure.

For a beginner's design exercise, I would add an arrival curve and the number of backend requests in one booking attempt. That makes the performance and scalability knobs concrete, and gives the eventual load test something more precise to reproduce than a count of users at ten o'clock.