Numbers are pretty much unusable for the next 24 hours after deployment. On stage, this is almost every day, on prod this skews the numbers.

Let’s calculate these numbers properly via a background job on regular intervals and export those values as prometheus gauges.