diff --git a/kubernetes/apps/monitoring/helmrelease-prometheus.yaml b/kubernetes/apps/monitoring/helmrelease-prometheus.yaml
index 3a2d2c46..55ed2a46 100644
--- a/kubernetes/apps/monitoring/helmrelease-prometheus.yaml
+++ b/kubernetes/apps/monitoring/helmrelease-prometheus.yaml
@@ -74,7 +74,36 @@ spec:
           # instance queries each PVE node's API on the user's behalf.
           # The targets list is the PVE node addresses; the actual
           # scrape goes to pve-exporter.monitoring.svc:9221.
+          #
+          # Cluster collectors (pve_up, pve_guest_info, …) return the SAME
+          # cluster-wide resource set no matter which node answers, so they
+          # are scraped from a single node only. Scraping them from all three
+          # (the pre-2026-07 config) ingested every node/guest series 3x —
+          # identical `id`, differing `instance` — a real TSDB cardinality
+          # cost that dashboards had to paper over with max by (id).
+          # Trade-off: cluster metrics pause if .201 is down; node-local
+          # metrics below keep flowing from the surviving nodes.
           - job_name: pve
+            static_configs:
+              - targets:
+                  - "10.2.40.201"
+            metrics_path: /pve
+            params:
+              cluster: ["1"]
+              node: ["0"]
+            relabel_configs:
+              - source_labels: [__address__]
+                target_label: __param_target
+              - source_labels: [__param_target]
+                target_label: instance
+              - target_label: __address__
+                replacement: pve-exporter.monitoring.svc.cluster.local:9221
+
+          # Node collectors (pve_version_info, replication status) are
+          # node-local and genuinely distinct per target, so every node is
+          # scraped — this is the split recommended by prometheus-pve-exporter
+          # upstream to avoid the cluster-collector duplication above.
+          - job_name: pve-node
             static_configs:
               - targets:
                   - "10.2.40.201"
@@ -82,7 +111,7 @@ spec:
                   - "10.2.40.203"
             metrics_path: /pve
             params:
-              cluster: ["1"]
+              cluster: ["0"]
               node: ["1"]
             relabel_configs:
               - source_labels: [__address__]