Each component has one verb: API stores, scheduler picks, kubelet runs.

The control plane splits work into narrow jobs. The API server is the front door, and etcd just holds the stored state. The scheduler notices a Pod with no node and chooses one using requests, constraints and spreading. Only after a node is named does that node's kubelet start the containers, so the kubelet never decides placement.

Read it once, then a question tests whether the trick stuck.