Roughly constant load with light noise.
The raw formula, recomputed every simulated second — before stabilization windows or rate limits are applied. See the ledger below for what actually happens.
Every scale event lands here. A rose flag means a scale-down was allowed sooner than its configured period implies — the effect described in Controlling the speed: Autoscaling Policy. Try the Flapping preset with a short scale-down period to trigger it on purpose.
One simulated day is 600s. Inside the window, pods never drop below the guaranteed minimum, whatever the metric says. Outside it, standard (or zero) scaling takes back over — this is the real Schedule-Based Autoscaling feature, not just a traffic shape.
Scale-down period is shortened for the demo — real deployments allow up to 1800s. The math is identical.