diff --git a/MANUAL.md b/MANUAL.md
index a60530c..67202dd 100644
--- a/MANUAL.md
+++ b/MANUAL.md
@@ -9,7 +9,11 @@ A short guide to installing ICC on Kubernetes with this Helm chart.
 - PostgreSQL reachable from the cluster (one server; ICC uses several databases
   on it, listed in section 2).
 - Valkey (or Redis), reachable from the cluster.
-- Prometheus, reachable from the cluster.
+- Prometheus, reachable from the cluster. It must scrape kubelet
+  `/metrics/cadvisor` and kube-state-metrics so ICC can read container resource
+  usage, requests, limits, and Pod labels. kube-state-metrics must export the
+  `platformatic.dev/monitor` Pod label.
+- A Kubernetes resource metrics API (`metrics.k8s.io`) for the ICC HPA.
 - Prometheus Operator CRDs (`PodMonitor`, `ServiceMonitor`) installed. The
   chart renders both by default; to install without them, disable the monitors
   (`watt.monitor.enable` and the per-service `monitor.enable`).
@@ -151,7 +155,8 @@ kubectl port-forward -n platformatic svc/icc 8080:80
 ```sh
 helm upgrade platformatic oci://ghcr.io/platformatic/helm \
   --version "^4.1.0" -n platformatic \
-  -f my-values.yaml -f my-secrets.yaml
+  -f my-values.yaml -f my-secrets.yaml \
+  --wait --timeout 10m
 
 helm uninstall platformatic -n platformatic
 ```
@@ -169,6 +174,10 @@ helm uninstall platformatic -n platformatic
   capabilities) is below 1.30. Upgrade the cluster.
 - `no matches for kind "PodMonitor"` (or `"ServiceMonitor"`): the Prometheus
   Operator CRDs are missing. Install them, or disable the monitors (section 1).
+- ICC CPU metrics are empty or zero: confirm Prometheus contains
+  `container_cpu_usage_seconds_total`, `kube_pod_container_resource_limits` or
+  `kube_pod_container_resource_requests`, and
+  `kube_pod_labels{label_platformatic_dev_monitor="prometheus"}`.
 - ICC pod not ready: check `kubectl logs -n platformatic deploy/icc`; usually
   the database, Valkey, or Prometheus URL is unreachable.
 
diff --git a/MIGRATING-v3-to-v4.md b/MIGRATING-v3-to-v4.md
index ddb8ea0..1267fc3 100644
--- a/MIGRATING-v3-to-v4.md
+++ b/MIGRATING-v3-to-v4.md
@@ -14,6 +14,9 @@ Test the upgrade in a non-production cluster first.
 - Kubernetes 1.30 or newer. The chart refuses to install on older clusters.
 - Helm 3.13.2 or newer.
 - PostgreSQL, Valkey or Redis, and Prometheus reachable from the ICC Pod.
+  Prometheus must scrape kubelet `/metrics/cadvisor` and kube-state-metrics.
+  kube-state-metrics must export the `platformatic.dev/monitor` Pod label.
+- A Kubernetes resource metrics API (`metrics.k8s.io`) for the ICC HPA.
 - A replacement for any Ingress or persistent storage created by chart v3. See
   [Resources removed or changed](#resources-removed-or-changed).
 
@@ -114,8 +117,11 @@ services:
 
     scaler:
       algorithm_version: v1
-      # Preserve the v3 values. The v3 defaults were 300 and 60.
+      # 300 preserves the chart v3 behavior. Chart v4 defaults to 15.
+      # With scaler v1, scale-up is blocked for this period after a scale
+      # operation; scale-down uses a shorter derived cooldown (60-180s).
       cooldown: 300
+      # Preserve the chart v3 default. Chart v4 also defaults to 60.
       periodic_trigger: 60
 
   machinist:
@@ -281,7 +287,8 @@ helm upgrade "$RELEASE" oci://ghcr.io/platformatic/helm \
 helm upgrade "$RELEASE" oci://ghcr.io/platformatic/helm \
   --version "$CHART_VERSION" \
   --namespace "$RELEASE_NAMESPACE" \
-  -f v4-values.yaml
+  -f v4-values.yaml \
+  --wait --timeout 10m
 ```
 
 Review the dry-run for unexpected namespace changes. Compare it with the saved
@@ -296,6 +303,24 @@ kubectl rollout status deployment/machinist --namespace platformatic
 helm test "$RELEASE" --namespace "$RELEASE_NAMESPACE"
 ```
 
+Run these queries in Prometheus. Each must return a value greater than zero:
+
+```promql
+count(container_cpu_usage_seconds_total{container!="POD"})
+```
+
+```promql
+count(
+  kube_pod_container_resource_limits{resource="cpu", unit="core"}
+  or
+  kube_pod_container_resource_requests{resource="cpu", unit="core"}
+)
+```
+
+```promql
+count(kube_pod_labels{label_platformatic_dev_monitor="prometheus"})
+```
+
 Then confirm:
 
 - ICC `GET /api/status` returns 200.