During our investigation on HELP-69798, we found that throughput probing was probably making a suboptimal decision about the total ticket amount.

• When there's a steady increase of operations, given the system is not yet overloaded, this usually translates to a steady increase of operation finished processing per seconds. Could that mean the throughput probing could be inclined to repeat the same decision for increasing or decreasing the amount of ticket total during this steady increase until throughput is affected?
• If throughput probing can repeat a decision until throughput is affected, can it happen that we see an increase in latency before the throughput gets affected?
• If both are the case, can it lead to situation similar to the following?
  • A change in workload affects the throughput positively, and the throughput probing repeats a decision to lower the amount of ticket total
  • The workload stabilize, and so the throughput. The throughput probing keep the amount of ticket total steady, despite the increase of latency
  • The workload changes again, causing the throughput probing to correct course and reduce latency by increasing

The question here is can a gradient like a constant increase in workload trick the throughput probing to think throughput is increasing, and keep doing a particular decision that end up being suboptimal? And can it explain nodes taking different decision with the same workload as seen in the help ticket?