Twenty-interview sprint

Opening

“I am researching how teams operate production inference. I am not selling a platform today. I want to understand the last real deployment.”

Evidence questions

1. Walk me through the last model you put in production.
2. Where did the schedule slip or the service become unpredictable?
3. Tell me about the most recent latency, capacity, cost, or rollout incident.
4. What did it consume: engineer time, GPU spend, delayed launch, errors, or customer trust?
5. What did you try? What remains unsolved?
6. Who owns the problem and who approves spend?
7. What traffic trace or test corpus could reproduce it safely?
8. If a four-week pilot removed the failure, what would you need to commit?

Do not ask

• Would you use this?
• Is lower latency valuable?
• Which features should we build?

Synthesis

Record observed event, consequence, workaround, owner, data access, urgency, and commitment. Do not count compliments.