The application was for a senior engineering role at a web development agency. What came back was a mandatory assignment, due in five days, sent before any introductory call or recruiter screen.
The scenario: a fictional six-person agency managing around fifty client sites. The candidate would build a working monitoring service from scratch — checking sites in a local Docker setup for availability, response times, keywords and TLS expiry, with check frequencies that scale by criticality, debouncing to prevent alert storms, state history, automated tests and notification stubs.
Then the cloud architecture: deploy it to Azure with cost estimates and repeatable deployment definitions, and design a multi-region high-availability strategy for privacy-sensitive clients. All data must stay in Canada. The system must survive the total loss of an entire Azure region, with explicit recovery-point and recovery-time commitments across compute, databases, files, backups, secrets, networking and edge.
Then an eight-page written explanation covering trade-offs, scaling to five hundred sites, self-monitoring, security, cost and a roadmap — all before a human being had spoken to them.