Sending LLM requests twice beats priority tier on tail latency

In a 50-request replay of production traffic, HOAi cut worst-case time-to-complete from 9.8s to 3.5s by sending each LLM request twice and keeping the faster reply, matching the median of paid priority tiers at no extra cost.