diff --git a/cmd/felis/operator.go b/cmd/felis/operator.go
index 2ac0e92..38e2a46 100644
--- a/cmd/felis/operator.go
+++ b/cmd/felis/operator.go
@@ -79,6 +79,15 @@ func cmdOperator(args []string, _, stderr io.Writer) int {
 		return 1
 	}
 
+	// Republish felis_servers_total from a periodic full List of the fleet. A
+	// per-object reconcile can never maintain a fleet-wide gauge correctly, so a
+	// snapshot Runnable owns it; it shares the manager's cached client and stops
+	// with the manager.
+	if err := mgr.Add(&operator.GaugeSyncer{Client: mgr.GetClient()}); err != nil {
+		fmt.Fprintf(stderr, "felis operator: add gauge syncer: %v\n", err)
+		return 1
+	}
+
 	if err := mgr.Start(ctrl.SetupSignalHandler()); err != nil {
 		fmt.Fprintf(stderr, "felis operator: manager exited: %v\n", err)
 		return 1
diff --git a/internal/metrics/metrics.go b/internal/metrics/metrics.go
index 805eb11..7db2331 100644
--- a/internal/metrics/metrics.go
+++ b/internal/metrics/metrics.go
@@ -57,6 +57,27 @@ var (
 	})
 )
 
+// SyncServerGauge republishes felis_servers_total from a full snapshot of the
+// fleet's per-server states. states holds one entry per MinecraftServer the
+// operator knows about (its desiredState).
+//
+// It Resets the GaugeVec before Setting one child per distinct state, so a state
+// that has drained to zero reports 0 rather than its stale last value. That is
+// the whole reason a periodic full-snapshot is used instead of inc/dec on
+// reconcile transitions: a snapshot is self-correcting and cannot drift on a
+// missed event. Producing the states slice (a cached List of MinecraftServers)
+// is the untestable I/O edge; this Reset+tally+Set logic is pure and unit-tested.
+func SyncServerGauge(states []string) {
+	ServersTotal.Reset()
+	counts := make(map[string]int, len(states))
+	for _, s := range states {
+		counts[s]++
+	}
+	for state, n := range counts {
+		ServersTotal.WithLabelValues(state).Set(float64(n))
+	}
+}
+
 // Collectors returns every felis_* collector in a stable order. Production and
 // tests register the same slice, so the test asserting the full set is exposed
 // also pins the production surface.
diff --git a/internal/metrics/metrics_test.go b/internal/metrics/metrics_test.go
index 5e27b2a..42bbade 100644
--- a/internal/metrics/metrics_test.go
+++ b/internal/metrics/metrics_test.go
@@ -54,6 +54,30 @@ func TestRegisterExposesNamedFelisMetrics(t *testing.T) {
 	}
 }
 
+func TestSyncServerGaugeResetsStaleStates(t *testing.T) {
+	// Self-contained (Reset->sync->assert within this one test) so it never
+	// clobbers another test's ServersTotal children. The correctness point is the
+	// Reset inside SyncServerGauge: a state present in one snapshot but absent from
+	// the next must drain to 0, not retain its stale last value forever.
+	SyncServerGauge([]string{"Running", "Running", "Stopped"})
+	if got := testutil.ToFloat64(ServersTotal.WithLabelValues("Running")); got != 2 {
+		t.Errorf("Running = %v, want 2", got)
+	}
+	if got := testutil.ToFloat64(ServersTotal.WithLabelValues("Stopped")); got != 1 {
+		t.Errorf("Stopped = %v, want 1", got)
+	}
+
+	// Next snapshot: the two Running servers are gone. Without the Reset the gauge
+	// would still report Running=2; with it, the child drains to 0.
+	SyncServerGauge([]string{"Stopped"})
+	if got := testutil.ToFloat64(ServersTotal.WithLabelValues("Running")); got != 0 {
+		t.Errorf("Running after drain = %v, want 0", got)
+	}
+	if got := testutil.ToFloat64(ServersTotal.WithLabelValues("Stopped")); got != 1 {
+		t.Errorf("Stopped = %v, want 1", got)
+	}
+}
+
 func TestRegisterIsIdempotent(t *testing.T) {
 	reg := prometheus.NewRegistry()
 	if err := Register(reg); err != nil {
diff --git a/internal/operator/gauge.go b/internal/operator/gauge.go
new file mode 100644
index 0000000..ca51b6a
--- /dev/null
+++ b/internal/operator/gauge.go
@@ -0,0 +1,90 @@
+package operator
+
+import (
+	"context"
+	"time"
+
+	"felis.lolicon.best/internal/apis/felis/v1alpha1"
+	"felis.lolicon.best/internal/metrics"
+	ctrl "sigs.k8s.io/controller-runtime"
+	"sigs.k8s.io/controller-runtime/pkg/client"
+)
+
+// defaultGaugeInterval is the republish cadence used when none is configured.
+const defaultGaugeInterval = 30 * time.Second
+
+// GaugeSyncer periodically republishes felis_servers_total (spec §23) from a
+// full List of MinecraftServers.
+//
+// felis_servers_total is a fleet-wide gauge partitioned by desiredState, which a
+// per-object Reconcile fundamentally cannot maintain: one reconcile observes a
+// single server, so it could never Set a correct fleet count and inc/dec on
+// transitions would drift on any missed event. A periodic full-snapshot sidesteps
+// that — every tick re-derives the entire gauge from the live list (and
+// metrics.SyncServerGauge Resets first), so the published value is always exactly
+// the current fleet and never accumulates stale state.
+//
+// It is registered as a manager.Runnable (mgr.Add), sharing the manager's cached,
+// namespace-scoped client and lifecycle; it stops when the manager's context is
+// cancelled.
+type GaugeSyncer struct {
+	// Client is the manager's cached client (namespace-scoped like the operator).
+	Client client.Client
+	// Interval is the republish cadence; defaults to defaultGaugeInterval.
+	Interval time.Duration
+}
+
+// Start runs the republish loop until ctx is cancelled, satisfying
+// manager.Runnable. The ticker loop and the live List are the untestable I/O
+// edge; the List->translate->gauge work it delegates to SyncOnce is exercised
+// against a fake client. A failed sync is logged best-effort, not fatal — the
+// next tick re-derives the gauge from scratch regardless.
+func (g *GaugeSyncer) Start(ctx context.Context) error {
+	interval := g.Interval
+	if interval <= 0 {
+		interval = defaultGaugeInterval
+	}
+	t := time.NewTicker(interval)
+	defer t.Stop()
+	// Publish once up front so the gauge is populated before the first tick.
+	if err := g.SyncOnce(ctx); err != nil {
+		ctrl.LoggerFrom(ctx).Error(err, "gauge: initial fleet sync failed")
+	}
+	for {
+		select {
+		case <-ctx.Done():
+			return nil
+		case <-t.C:
+			if err := g.SyncOnce(ctx); err != nil {
+				ctrl.LoggerFrom(ctx).Error(err, "gauge: fleet sync failed")
+			}
+		}
+	}
+}
+
+// SyncOnce Lists the fleet once and republishes felis_servers_total from it.
+// Separated from Start so the List->translate->gauge path is unit-testable
+// against a fake client, leaving only the ticker loop untested.
+func (g *GaugeSyncer) SyncOnce(ctx context.Context) error {
+	var list v1alpha1.MinecraftServerList
+	if err := g.Client.List(ctx, &list); err != nil {
+		return err
+	}
+	metrics.SyncServerGauge(serverStates(list.Items))
+	return nil
+}
+
+// serverStates maps each server to its desiredState, defaulting an unset state
+// to Stopped (the CRD default, spec §4). Pure, so the labelling rule is covered
+// by SyncOnce's fake-client test rather than only at runtime.
+func serverStates(items []v1alpha1.MinecraftServer) []string {
+	states := make([]string, 0, len(items))
+	for i := range items {
+		s := string(items[i].Spec.DesiredState)
+		if s == "" {
+			s = string(v1alpha1.DesiredStopped)
+		}
+		states = append(states, s)
+	}
+	return states
+}
diff --git a/internal/operator/gauge_test.go b/internal/operator/gauge_test.go
new file mode 100644
index 0000000..75f392a
--- /dev/null
+++ b/internal/operator/gauge_test.go
@@ -0,0 +1,50 @@
+package operator_test
+
+import (
+	"context"
+	"testing"
+
+	"felis.lolicon.best/internal/apis/felis/v1alpha1"
+	"felis.lolicon.best/internal/metrics"
+	"felis.lolicon.best/internal/operator"
+	"github.com/prometheus/client_golang/prometheus/testutil"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"sigs.k8s.io/controller-runtime/pkg/client/fake"
+)
+
+func gaugeServer(name string, state v1alpha1.DesiredState) *v1alpha1.MinecraftServer {
+	return &v1alpha1.MinecraftServer{
+		ObjectMeta: metav1.ObjectMeta{Name: name, Namespace: "minecraft"},
+		Spec:       v1alpha1.MinecraftServerSpec{Subdomain: name, DesiredState: state},
+	}
+}
+
+// TestGaugeSyncerPublishesFleetStateFromCluster drives SyncOnce against a fake
+// client to verify the whole production path bar the ticker: List the fleet,
+// translate each server to its desiredState (an unset state defaulting to
+// Stopped), and republish felis_servers_total partitioned by that state.
+func TestGaugeSyncerPublishesFleetStateFromCluster(t *testing.T) {
+	scheme := newScheme(t)
+	c := fake.NewClientBuilder().
+		WithScheme(scheme).
+		WithObjects(
+			gaugeServer("a", v1alpha1.DesiredRunning),
+			gaugeServer("b", v1alpha1.DesiredRunning),
+			gaugeServer("c", v1alpha1.DesiredStopped),
+			gaugeServer("d", ""), // unset desiredState must be counted as Stopped
+		).
+		Build()
+
+	g := &operator.GaugeSyncer{Client: c}
+	if err := g.SyncOnce(context.Background()); err != nil {
+		t.Fatalf("SyncOnce: %v", err)
+	}
+
+	if got := testutil.ToFloat64(metrics.ServersTotal.WithLabelValues("Running")); got != 2 {
+		t.Errorf("servers_total{state=Running} = %v, want 2", got)
+	}
+	// One explicit Stopped plus one unset (defaulted to Stopped) = 2.
+	if got := testutil.ToFloat64(metrics.ServersTotal.WithLabelValues("Stopped")); got != 2 {
+		t.Errorf("servers_total{state=Stopped} = %v, want 2", got)
+	}
+}