What Candidates Reported
A February 2026 candidate report describes an OpenAI system-design round for a GitHub Actions-like CI/CD platform with high reliability as the first priority. A separate May 2026 Software Engineer report names a devbox and CI/CD pipeline design. The independent accounts support this prompt family without revealing an internal OpenAI rubric.
Problem and Scope
Design event-triggered workflow runs, versioned DAGs, isolated job execution, logs, artifacts, caches, secrets, cancellation, and retry. Clarify job duration, worker types, trust boundaries, acceptable duplicate side effects, retention, and queue-delay objectives before selecting orchestration or compute technology.
Critical Design Decisions
A strong answer separates durable control-plane state from worker execution. Jobs need explicit attempts, leases, heartbeats, and monotonic transitions. Worker pools need fair matching and isolation. Secrets must be short-lived and policy-scoped, while artifacts and caches require provenance, trust-domain boundaries, and retention controls.
Failure Scenarios to Practise
Test worker death after an external side effect, duplicate repository events, cancellation races, a malicious fork, cache poisoning, a huge monorepo burst, exhausted specialized workers, artifact-upload failure, and orchestration-region loss. State which record is authoritative and how uncertain work is reconciled.
How the Answer Is Evaluated
MockMe scores scope and execution semantics (15%), orchestration and workers (25%), security and isolation (20%), scale and reliability (25%), and communication and trade-offs (15%). High scores require durable workflow state, retry-safe attempts, enforceable security boundaries, tenant fairness, and credible regional recovery.