One latency tile creates shared blindness
A single aggregate for every route makes latency tracking optional theatre.
UK product teams often inherit one APM tile that pre-dates the current API surface. Every feature inherits that blindness.
Separating environments is table stakes. Separating tags by route, version, and caller is what lets latency tracking survive the next launch. Endpoint Latency Attribution starts there.
Before celebrating a p95 drop, list which tags and sampling rules changed beside the chart. Sampling thrift and genuine speed look identical until you can tell them apart.