The decomposition round hands you a real-world problem with no technical framing — analyse a city’s taxi data, improve emergency response times, unify fraud signals across five systems — and measures whether you scope before you solve. Strong candidates spend the first five minutes on who the user is, what decision changes, and what “better” measures, then decompose into prioritised sub-problems and name what they would not build.
Worked example: given “improve 911 response times”, the winning first move is asking whether response time means call-to-dispatch, dispatch-to-arrival or arrival-to-resolution — three metrics with three owners and three different fixes. Gotcha: interviewers routinely remove a constraint late (“actually you cannot have that data”) to test whether your plan was reasoned or memorised; defending the original plan after its premise is gone is a fail.