The underlying issue was due to resource re-allocation, with the service switching servers mid-request. Issue is now resolved and latency should retain to normal.

An Internal API service that powers Elo's answer generation experienced temporary high latency (>30s), leading to a misfire of our downtime detector.