Average utilisation hides the cluster that felt the change
A healthy mean can still leave one region waiting on capacity that the pack never named.
Infrastructure teams serving Hong Kong clients often inherit a single utilisation percentile that pre-dates multi-region cutovers. Every feature inherits that blindness.
Separating timers by environment is table stakes. Separating tags by cluster and change class is what lets reporting survive the next launch. Release Report Attribution starts there.
Before celebrating a utilisation drop, list which releases and cohorts changed beside the chart. Pipeline hygiene and product thrift look identical until you can tell them apart.