Role: Cloud Architect, existing estateFull interview loop

For a team that already has a platform in the cloud, or is half in and half out, and is hiring the person who decides what that estate becomes. It assumes other people will build most of what the architect decides, so there is no round on writing infrastructure code and none on running the delivery pipeline or carrying the pager; those belong to other roles' loops.

Every round starts from a document the architect would be handed in their first month rather than from a blank whiteboard: a board's request to survive the loss of a region, a month's bill with its flow logs, a migration plan written before anyone read the dependency map, and a decision record already marked accepted with objections underneath it. The order runs from a design the candidate owns, to what the estate costs, to moving what already exists, to getting a decision through the people it lands on.

Boundaries and blast radius is read three times, at three different layers, because boundaries are the decisions most expensive to reverse and the least visible while they are right: where one customer's data may be stored in the first round, how two overlapping networks meet in the third, and how accounts and identity separate fourteen teams in the fourth. The other must-haves are read twice, each from two directions. Resilience is designed in the first round and priced in the second, where finance offers to save money by giving up a zone. Cost is read off a bill in the second round and forecast as a three-year managed-or-self-run choice in the third. Written decisions and buy-in is read as a paragraph a board has to sign in the first round, and as a record under review, with four comments to answer and a story of the candidate's own, in the fourth.

Two trade-offs are deliberate. Migration planning is a nice-to-have read once, because a team whose estate is already in the cloud still gets a full round of network and database decisions out of the data centre exercise; if the job is the exit itself, make it a must and add a second read, such as a migration the candidate has run, to the fourth round. And no round measures depth in one provider: this loop will pass an architect who reasons carefully across providers and writes decisions down over one who knows a single provider's services by heart and cannot price a design. If your architect will also build the foundations with their own hands, add a hands-on round for that and score it separately.