diff --git a/docs/4-implementation/scheduler/observability.md b/docs/4-implementation/scheduler/observability.md
index 9aa7ccae..0c4125d3 100644
--- a/docs/4-implementation/scheduler/observability.md
+++ b/docs/4-implementation/scheduler/observability.md
@@ -208,6 +208,12 @@ fixes the collector port every hive and its agents share,
 That's the reference; this page covers what a per-option listing can't: the
 two-tier architecture, the security model, and how the options interact.
 
+The host collector's `hostmetrics` filesystem scraper reports only the mount
+points in `services.hyperhive.otel.collector.filesystemMountPoints` (default
+`/`, `/boot` and `/nix`), as `system.filesystem.*` with a `mountpoint` label. A
+listed mount the host lacks reports nothing; `/nix` appears only where it's its
+own mount.
+
 ## Built-in resource labels
 
 The harness sets the OTLP variables (`OTEL_EXPORTER_OTLP_ENDPOINT`, `_PROTOCOL`,
diff --git a/nix/host-modules/otel.nix b/nix/host-modules/otel.nix
index 8505eb9d..bfe7da4d 100644
--- a/nix/host-modules/otel.nix
+++ b/nix/host-modules/otel.nix
@@ -189,6 +189,29 @@ in
       '';
     };
 
+    collector.filesystemMountPoints = lib.mkOption {
+      type = lib.types.listOf lib.types.str;
+      default = [
+        "/"
+        "/boot"
+        "/nix"
+      ];
+      example = [
+        "/"
+        "/boot"
+        "/data"
+      ];
+      description = ''
+        Mount points the host collector's `filesystem` scraper reports
+        (`system.filesystem.*`), matched exactly. The list is strict because
+        the scraper has to include virtual filesystems to see a btrfs
+        subvolume root, and that would otherwise report every tmpfs, proc
+        and cgroup mount plus each per-container btrfs subvolume, which
+        repeats the same device's usage. A mount point on the list that a
+        host does not have reports nothing.
+      '';
+    };
+
     httpChecks = lib.mkOption {
       type = lib.types.listOf lib.types.str;
       default = [ ];
@@ -480,7 +503,19 @@ in
               scrapers = {
                 cpu = { };
                 memory = { };
-                filesystem = { };
+                # `include_virtual_filesystems` turns gopsutil's `all` on.
+                # With it off, a mount whose mountinfo root is not `/`
+                # counts as a bind mount and is dropped, which includes a
+                # btrfs `subvol=@` root. The strict list then keeps the
+                # now-visible tmpfs, proc and per-container subvolume
+                # mounts out of the store.
+                filesystem = {
+                  include_virtual_filesystems = true;
+                  include_mount_points = {
+                    match_type = "strict";
+                    mount_points = otel.collector.filesystemMountPoints;
+                  };
+                };
                 load = { };
                 disk = { };
                 network = { };
diff --git a/nix/module-eval/matrix-core.nix b/nix/module-eval/matrix-core.nix
index 8bea9c56..6a392307 100644
--- a/nix/module-eval/matrix-core.nix
+++ b/nix/module-eval/matrix-core.nix
@@ -207,6 +207,31 @@ let
         && (settings matrixElsewhere).receivers ? hostmetrics
         && lib.elem "hostmetrics" (pipeline matrixElsewhere);
     }
+    {
+      # The root of a btrfs `subvol=@` host has a mountinfo root other than
+      # `/`, which the scraper drops as a bind mount unless virtual
+      # filesystems are included. The strict list is what keeps that from
+      # reporting every tmpfs and per-container subvolume. The lookup is
+      # read on a host that has the receiver (the pipeline arm above is its
+      # control).
+      name = "the host collector's filesystem scraper includes virtual filesystems and reports only the listed mount points";
+      ok =
+        let
+          fs =
+            matrixElsewhere.services.opentelemetry-collector.settings.receivers.hostmetrics.scrapers.filesystem;
+        in
+        fs.include_virtual_filesystems == true
+        && fs.include_mount_points.match_type == "strict"
+        &&
+          fs.include_mount_points.mount_points
+          == matrixElsewhere.services.hyperhive.otel.collector.filesystemMountPoints
+        &&
+          fs.include_mount_points.mount_points == [
+            "/"
+            "/boot"
+            "/nix"
+          ];
+    }
     {
       # The size is read from the database directory, as the dynamic user the
       # homeserver runs as, and pushed to the forwarder's OTLP port. Without a