Grouped by endpoint. Each chart is RPS over time, one line per server (three lean-tea backends — reactor / fast / libuv — plus nginx), measured on the same CI runner in the same job (see the docs · methodology).