diff --git a/modules/observability.nix b/modules/observability.nix
index 247d2e5..22e4086 100644
--- a/modules/observability.nix
+++ b/modules/observability.nix
@@ -477,13 +477,14 @@
 
       services.alloy.enable = true;
       systemd.services.alloy.serviceConfig.SupplementaryGroups = [ "systemd-journal" ];
-      services.prometheus.exporters.node = {
-        # enabledCollectors = [
-        #   "node"
-        #   "postgres"
-        # ];
-        enable = true;
-      };
+      services.prometheus.exporters.node.enable = true;
+
+      # SMART disk metrics. smartctl is its own exporter (:9633), NOT a
+      # node_exporter collector — passing it via enabledCollectors makes
+      # node_exporter fail on an unknown --collector.smartctl flag. It only
+      # yields data on hosts whose disks actually expose SMART; plain virtio
+      # disks in VMs don't, so expect empty results there.
+      services.prometheus.exporters.smartctl.enable = true;
 
       # Process exporter for per-service CPU/memory metrics.
       # Runs as root so it can read /proc for all processes.
@@ -517,6 +518,15 @@
               forward_to = [prometheus.remote_write.nixvms.receiver]
             }
 
+            // Scrape the standalone smartctl exporter (port 9633). Without a
+            // scrape job its metrics never reach Mimir — the exporter just
+            // sits there answering /metrics to nobody.
+            prometheus.scrape "smartctl" {
+              targets    = [{"__address__" = "127.0.0.1:9633"}]
+              job_name   = "integrations/smartctl"
+              forward_to = [prometheus.remote_write.nixvms.receiver]
+            }
+
             // Scrape the NixOS-native process exporter (runs as root, port 9256).
             // Produces namedprocess_namegroup_cpu_seconds_total{groupname="<exe>"}.
             prometheus.scrape "process" {