diff --git a/monitoring/alert_rules.yml b/monitoring/alert_rules.yml
index 2debefb..5baa112 100644
--- a/monitoring/alert_rules.yml
+++ b/monitoring/alert_rules.yml
@@ -142,26 +142,25 @@ groups:
           description: "{{ $labels.name }} has restarted more than 2 times in 15 minutes."
 
       - alert: ContainerHighCPU
+        # Our containers run WITHOUT a CPU quota (CpuQuota=0), so the old
+        # `rate(...) * 100 > 80` treated 1 full core = 100% and fired every time
+        # forgejo/erpnext briefly used ~0.8 of a core — on a 12-core host that is
+        # only ~7% of the machine, pure noise. Normalize against the whole host
+        # (machine_cpu_cores) so the number means "share of the entire server this
+        # one container is eating." A single unlimited container sustained above
+        # half the host for 30m is a genuine runaway worth paging on.
         expr: |
-          (
-            sum by (instance, name) (
-              rate(container_cpu_usage_seconds_total{name=~"performancewest-.*"}[5m])
-            )
-            /
-            sum by (instance, name) (
-              container_spec_cpu_quota{name=~"performancewest-.*"}
-              / container_spec_cpu_period{name=~"performancewest-.*"}
-            )
-            * 100 > 80
+          sum by (instance, name) (
+            rate(container_cpu_usage_seconds_total{name=~"performancewest-.*"}[5m])
           )
-          and on (instance, name)
-          sum by (instance, name) (container_spec_cpu_quota{name=~"performancewest-.*"}) > 0
+          / scalar(max(machine_cpu_cores))
+          * 100 > 50
         for: 30m
         labels:
           severity: warning
         annotations:
-          summary: "Container {{ $labels.name }} high CPU ({{ $value | printf \"%.0f\" }}%)"
-          description: "Container CPU usage above 80% of its CPU quota for 30 minutes."
+          summary: "Container {{ $labels.name }} high CPU ({{ $value | printf \"%.0f\" }}% of host)"
+          description: "Container using more than 50% of the whole host's CPU for 30 minutes."
 
       - alert: ContainerHighMemory
         expr: |