17.29Interview answer

“If the business doesn't require immediate Salesforce completion, I'd decouple intake from processing with durable messaging. The API validates and persists or publishes the work, returns 202 with a request ID, and a worker performs an idempotent Salesforce upsert with bounded retry. That makes a 30-minute Salesforce outage a backlog rather than a 30-minute HTTP timeout problem. I'd monitor backlog age and expose status or failure state to the caller.”

That's a very strong architecture answer.