System design: a product tour
Can this engineer design something that holds up under load, within budget?
You’re Priya Raman
Candidate
Priya Raman designs a fix for a notification outage on a board, then runs traffic through it. Then you see what the hiring team sees: every run, scenarios Priya never saw, and how each level is set.
8 steps · about 2 minutes
Read the tour as text
Do they start from the facts?
Each task starts with a brief and facts from the business: the traffic to handle, what must not fail, prices and a monthly budget. Which facts Priya reads before submitting is recorded.
Do they design, or just write?
Priya designs on a board, not in code: services, queues, caches, databases and the guards between them. Each component shows its monthly price.
Does it hold up under load?
Run traffic sends the task’s traffic through the design and names any part that cannot keep up. Traffic runs are in practice tasks now, not yet in assessments. On the starting design both scenarios fail: the dispatcher pool gets 5,000 notifications a second against a capacity of 1,600.
Is cost part of the answer?
Priya adds a per-workspace rate limit where notifications are enqueued, caps retries and runs again. Both scenarios pass, and the cost meter shows $4,400 a month against the $5,000 target.
Do they know what it costs?
Priya writes down what the design gives up and why it is worth it, then submits. Submitting freezes the board and the rationale.
What did they try, and in what order?
You see every traffic run: what changed on the board since the one before, the cost, and which scenarios failed. Assessed tasks have no traffic runs yet, so this view is built from the practice.
Does it hold up beyond what they tested?
After submit, every scenario runs on the final design, including ones Priya never saw. Here a larger burst hits only the change marked to ship this week.
How is each level set?
Rules set each level from the final design and the facts read before submitting. Traffic runs are evidence for you, not points. Tradeoff reasoning waits for you: for each cost, you answer whether the rationale explains it. An AI label and a keyword check are only suggestions.