diff --git a/sgl-model-gateway/src/observability/metrics.rs b/sgl-model-gateway/src/observability/metrics.rs
index 5523f5c75..de380a864 100644
--- a/sgl-model-gateway/src/observability/metrics.rs
+++ b/sgl-model-gateway/src/observability/metrics.rs
@@ -190,6 +190,10 @@ pub fn init_metrics() {
         "smg_worker_errors_total",
         "Worker-level errors by worker_type, connection_mode, error_type"
     );
+    describe_gauge!(
+        "smg_manual_policy_cache_entries",
+        "Number of routing entries in manual policy cache"
+    );
 
     // Layer 3: Worker resilience metrics (circuit breaker)
     describe_gauge!(
@@ -810,6 +814,11 @@ impl Metrics {
         .increment(1);
     }
 
+    /// Set manual policy cache entries count
+    pub fn set_manual_policy_cache_entries(count: usize) {
+        gauge!("smg_manual_policy_cache_entries").set(count as f64);
+    }
+
     /// Record consistent hashing policy execution branch for routing decisions
     pub fn record_worker_consistent_hashing_policy_branch(branch: &'static str) {
         counter!(
diff --git a/sgl-model-gateway/src/policies/manual.rs b/sgl-model-gateway/src/policies/manual.rs
index f4d55828d..55d3c6cb2 100644
--- a/sgl-model-gateway/src/policies/manual.rs
+++ b/sgl-model-gateway/src/policies/manual.rs
@@ -161,6 +161,7 @@ impl LoadBalancingPolicy for ManualPolicy {
     fn select_worker(&self, workers: &[Arc<dyn Worker>], info: &SelectWorkerInfo) -> Option<usize> {
         let (result, branch) = self.select_worker_impl(workers, info);
         Metrics::record_worker_manual_policy_branch(branch.as_str());
+        Metrics::set_manual_policy_cache_entries(self.routing_map.len());
         result
     }