Which idle-timeout setting minimizes cost for Bamboo elastic agents in low-traffic workloads?

For a low-traffic Bamboo Server deployment that uses AWS elastic agents, the goal is to choose an agent idle-timeout value that reduces compute charges without noticeably delaying build feedback. The uncertainty lies in the trade-off between the 30-90 second provisioning latency and the cost of keeping an agent running idle; build frequency can vary from a few per day to sporadic bursts, making a fixed timeout risky.

What idle-timeout duration yields the lowest monthly AWS/Azure compute expense while keeping the average queue wait time below an acceptable threshold? How does setting the minimum idle agent count to zero versus one influence the total cost when build spikes occur?