Average lag hides the cohort that feels stuck
A healthy mean can still leave one device class waiting minutes for catch-up.
Application teams serving Hong Kong clients often inherit a single lag percentile that pre-dates multi-device editing. Every feature inherits that blindness.
Separating timers by path is table stakes. Separating tags by device class and tenant is what lets sync health survive the next launch. Lag Path Attribution starts there.
Before celebrating a lag drop, list which paths and cohorts changed beside the chart. Pipeline hygiene and product thrift look identical until you can tell them apart.