Until today, the latency metric used was mtail-computed average Apache response time across the appservers (a mean of means).

This has been pretty inaccurate/unreliable for a while now in the transition to mw-on-k8s, but with today marking the 100% transition, it's definitely time to switch.

After some experimenting with queries and also taking into consideration the load they place on Thanos, we'll be moving to a similar metric computed by Benthos.

On the Statuspage side, we'll also be purging the history of the current latency metric so we can backfill it with the new metric.
This will cause a temporary (~20 minute?) lack of data in the metric, but that's fine.
(It *is* possible to do this user-invisibly, but that requires multiple back and forth updates to both the Statuspage admin UI and also patching the config file -- on the UI create a new, invisible metric; on statograph add it to the config file and start populating data; wait a while; etc)