diff --git a/deploy/kubernetes/easyai-ai-gateway-cluster-release b/deploy/kubernetes/easyai-ai-gateway-cluster-release
index f2b9bf6..a2c115a 100755
--- a/deploy/kubernetes/easyai-ai-gateway-cluster-release
+++ b/deploy/kubernetes/easyai-ai-gateway-cluster-release
@@ -712,6 +712,65 @@ EOF
   return 1
 }
 
+prepare_postgres_single_instance() {
+  local source_sha=$1
+  local current_instances current_primary primary_node synchronous_enabled
+  if ! grep -Eq '^  instances: 1$' "$DESIRED_STATE_DIR/database.yaml" ||
+     ! grep -Eq '^    synchronous: null$' "$DESIRED_STATE_DIR/database.yaml"; then
+    echo 'production PostgreSQL desired state must declare one instance with synchronous replication disabled' >&2
+    return 1
+  fi
+  current_instances=$("${kubectl[@]}" get cluster easyai-postgres -n "$NAMESPACE" \
+    -o jsonpath='{.spec.instances}')
+  synchronous_enabled=$("${kubectl[@]}" get cluster easyai-postgres -n "$NAMESPACE" -o json |
+    jq -r '.spec.postgresql.synchronous != null')
+  if [[ $current_instances == 1 && $synchronous_enabled == false ]]; then
+    return 0
+  fi
+  current_primary=$("${kubectl[@]}" get cluster easyai-postgres -n "$NAMESPACE" \
+    -o jsonpath='{.status.currentPrimary}')
+  [[ $current_primary =~ ^easyai-postgres-[0-9]+$ ]] || {
+    echo "production PostgreSQL primary is unavailable: primary=${current_primary:-unknown}" >&2
+    return 1
+  }
+  primary_node=$("${kubectl[@]}" get pod "$current_primary" -n "$NAMESPACE" \
+    -o jsonpath='{.spec.nodeName}')
+  [[ $primary_node == easyai-ningbo ]] || {
+    echo "refusing PostgreSQL downscale while primary is outside Ningbo: primary=$current_primary node=$primary_node" >&2
+    return 1
+  }
+  run_backup "$source_sha"
+  # A distinct preflight field manager cannot remove the existing synchronous
+  # object by omission. Remove it explicitly before applying instances=1 so
+  # CNPG never validates the old number=1 setting against the new topology.
+  "${kubectl[@]}" patch cluster easyai-postgres -n "$NAMESPACE" --type=merge \
+    -p '{"spec":{"postgresql":{"synchronous":null}}}' >/dev/null
+}
+
+verify_postgres_single_instance() {
+  local state primary primary_node replicas
+  for _ in $(seq 1 120); do
+    state=$("${kubectl[@]}" get cluster easyai-postgres -n "$NAMESPACE" -o json |
+      jq -r '[.spec.instances, (.status.readyInstances // 0)] | @tsv')
+    if [[ $state == $'1\t1' ]]; then
+      primary=$("${kubectl[@]}" get cluster easyai-postgres -n "$NAMESPACE" \
+        -o jsonpath='{.status.currentPrimary}')
+      primary_node=$("${kubectl[@]}" get pod "$primary" -n "$NAMESPACE" \
+        -o jsonpath='{.spec.nodeName}')
+      replicas=$("${kubectl[@]}" exec -n "$NAMESPACE" "$primary" -c postgres -- \
+        psql -X -U postgres -d easyai_ai_gateway -At -c \
+        'SELECT count(*) FROM pg_stat_replication;' 2>/dev/null || true)
+      if [[ $primary_node == easyai-ningbo && $replicas == 0 ]]; then
+        echo "production_postgres=PASS topology=ningbo-single-instance primary=$primary"
+        return 0
+      fi
+    fi
+    sleep 5
+  done
+  echo "production PostgreSQL did not converge to Ningbo single-instance topology: state=${state:-unknown}" >&2
+  return 1
+}
+
 run_migrator() {
   local source_sha=$1
   local api_image=$2
@@ -929,6 +988,9 @@ activate_manifest() {
   deployment_snapshot=$(mktemp "$RELEASES_DIR/.deployments.XXXXXX")
   snapshot_application_deployments "$deployment_snapshot"
 
+  if [[ $action == deploy ]]; then
+    prepare_postgres_single_instance "$source_sha"
+  fi
   if [[ $action == deploy && $migrations_changed == true ]]; then
     run_backup "$source_sha"
     run_migrator "$source_sha" "$api_image"
@@ -943,6 +1005,12 @@ activate_manifest() {
     rm -f -- "$deployment_snapshot" "$current_file"
     return 1
   fi
+  if [[ $action == deploy ]] && ! verify_postgres_single_instance; then
+    echo '[cluster-release] PostgreSQL single-instance verification failed; restoring exact deployment snapshot' >&2
+    restore_application_deployments "$deployment_snapshot" || true
+    rm -f -- "$deployment_snapshot" "$current_file"
+    return 1
+  fi
   if [[ $action == deploy && $api_changed == true ]] &&
      ! run_capacity_controller_preflight "$source_sha" "$api_image"; then
     echo '[cluster-release] capacity controller preflight failed; restoring exact deployment snapshot' >&2
diff --git a/deploy/kubernetes/production/database.yaml b/deploy/kubernetes/production/database.yaml
index 8dd6c38..d50d7e4 100644
--- a/deploy/kubernetes/production/database.yaml
+++ b/deploy/kubernetes/production/database.yaml
@@ -63,7 +63,9 @@ metadata:
     app.kubernetes.io/name: easyai-postgres
     app.kubernetes.io/part-of: easyai-ai-gateway
 spec:
-  instances: 2
+  # Production currently uses one Ningbo primary. Cross-region replicas are
+  # intentionally disabled until a separately certified HA topology exists.
+  instances: 1
   imageName: ghcr.io/cloudnative-pg/postgresql:18.4-standard-trixie@sha256:4587df73024408f5b2be9b4dd6ba2ccee8c9e5dc0c9a87c274c292291cc8a68c
   imagePullPolicy: IfNotPresent
   env:
@@ -106,6 +108,7 @@ spec:
     topologyKey: easyai.io/site
     nodeSelector:
       easyai.io/database: "true"
+      easyai.io/site: ningbo
   resources:
     requests:
       cpu: 500m
@@ -114,10 +117,9 @@ spec:
       cpu: "2"
       memory: 3Gi
   postgresql:
-    synchronous:
-      method: any
-      number: 1
-      dataDurability: preferred
+    # Keep this explicit null so server-side apply atomically removes the old
+    # two-instance synchronous setting during preflight field management.
+    synchronous: null
     parameters:
       # A 16 MiB WAL segment every 60 seconds requires at least 2.24 Mbit/s
       # of continuous cross-region throughput even while the database is idle.
@@ -129,10 +131,9 @@ spec:
       wal_compression: "on"
       wal_keep_size: 1GB
       # CNPG defaults these cross-region liveness timers to 5 seconds. A short
-      # control-plane or pod-network stall can otherwise tear down a healthy
-      # WAL stream and make preferred synchronous replication fall back to
-      # async. Keep failure detection bounded without amplifying a transient
-      # multi-second stall into repeated reconnects.
+      # control-plane or pod-network stall can otherwise tear down a future
+      # WAL stream. Keep failure detection bounded without amplifying a
+      # transient multi-second stall into repeated reconnects.
       wal_sender_timeout: 30s
       wal_receiver_timeout: 30s
   monitoring:
diff --git a/scripts/cluster/monitor-production-release.sh b/scripts/cluster/monitor-production-release.sh
index 11e5af4..48e0004 100755
--- a/scripts/cluster/monitor-production-release.sh
+++ b/scripts/cluster/monitor-production-release.sh
@@ -246,11 +246,11 @@ while (( $(date +%s) - started_epoch < AI_GATEWAY_RELEASE_MONITOR_TOTAL_SECONDS
       if (value>max) max=value
     } END {printf "%.0f",max}')
   (( pod_max_memory < 1536 )) || failure_gate_id=${failure_gate_id:-pod_memory_hard}
-  [[ $(remote_kubectl get cluster easyai-postgres -n "$namespace" \
-    -o 'jsonpath={.status.readyInstances}') == 2 ]] || failure_gate_id=${failure_gate_id:-postgres_ready}
-  sync_state=$(database_query "SELECT COALESCE(string_agg(sync_state,','),'') FROM pg_stat_replication;")
-  [[ ",$sync_state," == *,sync,* || ",$sync_state," == *,quorum,* ]] ||
-    failure_gate_id=${failure_gate_id:-postgres_replication}
+  postgres_state=$(remote_kubectl get cluster easyai-postgres -n "$namespace" -o json |
+    jq -r '[.spec.instances, (.status.readyInstances // 0)] | @tsv')
+  [[ $postgres_state == $'1\t1' ]] || failure_gate_id=${failure_gate_id:-postgres_ready}
+  replica_count=$(database_query "SELECT count(*) FROM pg_stat_replication;")
+  [[ $replica_count == 0 ]] || failure_gate_id=${failure_gate_id:-postgres_replication}
   connections=$(database_query "SELECT count(*) FROM pg_stat_activity WHERE backend_type='client backend';")
   max_connections=$(database_query "SELECT setting::int FROM pg_settings WHERE name='max_connections';")
   (( connections < 150 && connections * 4 < max_connections * 3 )) ||
diff --git a/scripts/cluster/verify-cluster.sh b/scripts/cluster/verify-cluster.sh
index 1f15518..720e0a3 100755
--- a/scripts/cluster/verify-cluster.sh
+++ b/scripts/cluster/verify-cluster.sh
@@ -96,22 +96,23 @@ echo "k3s_control_plane=PASS servers=3 worker_nodes=$((expected_nodes - 3)) witn
 database_state=$(cluster_ssh "$CLUSTER_NINGBO_HOST" 'bash -s' <<'REMOTE'
 set -euo pipefail
 kubectl='k3s kubectl'
+configured_instances=$($kubectl get cluster easyai-postgres -n easyai -o jsonpath='{.spec.instances}')
 ready_instances=$($kubectl get cluster easyai-postgres -n easyai -o jsonpath='{.status.readyInstances}')
-[[ $ready_instances -eq 2 ]] || {
-  echo "postgres ready instance gate failed: ready_instances=$ready_instances expected=2" >&2
+[[ $configured_instances -eq 1 && $ready_instances -eq 1 ]] || {
+  echo "postgres single-instance gate failed: configured_instances=$configured_instances ready_instances=$ready_instances expected=1/1" >&2
   exit 1
 }
 primary=$($kubectl get cluster easyai-postgres -n easyai -o jsonpath='{.status.currentPrimary}')
 primary_node=$($kubectl get pod "$primary" -n easyai -o jsonpath='{.spec.nodeName}')
-[[ $primary_node == easyai-ningbo || $primary_node == easyai-hongkong ]] || {
+[[ $primary_node == easyai-ningbo ]] || {
   echo "postgres primary placement gate failed: primary_node=$primary_node" >&2
   exit 1
 }
 replication=$($kubectl exec -n easyai "$primary" -c postgres -- \
   psql -X -U postgres -d easyai_ai_gateway -At -c \
-  "SELECT count(*) FROM pg_stat_replication WHERE sync_state IN ('sync','quorum');")
-[[ $replication -eq 1 ]] || {
-  echo "postgres synchronous replication gate failed: synchronous_replicas=$replication expected=1" >&2
+  "SELECT count(*) FROM pg_stat_replication;")
+[[ $replication -eq 0 ]] || {
+  echo "postgres single-instance replication gate failed: replicas=$replication expected=0" >&2
   exit 1
 }
 archived_count=$($kubectl exec -n easyai "$primary" -c postgres -- \
@@ -146,17 +147,17 @@ for pod in $($kubectl get pods -n easyai -l cnpg.io/cluster=easyai-postgres \
     exit 1
   fi
 done
-[[ $readable_backup_configs -eq 2 ]] || {
-  echo "postgres projected backup config count failed: readable=$readable_backup_configs expected=2" >&2
+[[ $readable_backup_configs -eq 1 ]] || {
+  echo "postgres projected backup config count failed: readable=$readable_backup_configs expected=1" >&2
   exit 1
 }
-printf 'primary=%s archived=%s backups=%s ready=%s synchronous=%s continuous_archiving=%s backup_configs=%s\n' \
+printf 'primary=%s archived=%s backups=%s ready=%s replicas=%s continuous_archiving=%s backup_configs=%s\n' \
   "$primary" "$archived_count" "$completed_backups" "$ready_instances" \
   "$replication" "$continuous_archiving" "$readable_backup_configs"
 REMOTE
 )
 [[ $database_state == primary=easyai-postgres-* ]]
-echo "postgres_preflight=PASS $database_state synchronous_replicas=1"
+echo "postgres_preflight=PASS $database_state replicas=0 topology=ningbo-single-instance"
 
 if [[ $mode == postcutover ]]; then
   application_state=$(cluster_ssh "$CLUSTER_NINGBO_HOST" 'bash -s' <<'REMOTE'
diff --git a/tests/release/cluster-release-helper-test.sh b/tests/release/cluster-release-helper-test.sh
index a862bbf..3289c9a 100755
--- a/tests/release/cluster-release-helper-test.sh
+++ b/tests/release/cluster-release-helper-test.sh
@@ -102,6 +102,11 @@ fi
 grep -Fq 'AI_GATEWAY_WORKER_REPLICAS_HONGKONG=1' \
   "$root/deploy/kubernetes/easyai-ai-gateway-cluster-release.conf.example"
 grep -Fq "hongkong) replicas=\$AI_GATEWAY_WORKER_REPLICAS_HONGKONG" "$helper"
+grep -Fq "prepare_postgres_single_instance \"\$source_sha\"" "$helper"
+grep -Fq "'production PostgreSQL desired state must declare one instance with synchronous replication disabled'" "$helper"
+grep -Fq 'refusing PostgreSQL downscale while primary is outside Ningbo' "$helper"
+grep -Fq 'production_postgres=PASS topology=ningbo-single-instance' "$helper"
+grep -Fq -- "-p '{\"spec\":{\"postgresql\":{\"synchronous\":null}}}'" "$helper"
 grep -Fq "\"\${kubectl[@]}\" replace -f -" "$helper"
 if grep -Fq "\"\${kubectl[@]}\" apply -f \"\$snapshot\"" "$helper"; then
   echo 'deployment rollback still uses merge semantics instead of exact restore' >&2
diff --git a/tests/release/manual-release-test.sh b/tests/release/manual-release-test.sh
index 829da40..f4bfd57 100755
--- a/tests/release/manual-release-test.sh
+++ b/tests/release/manual-release-test.sh
@@ -314,6 +314,9 @@ if grep -Fq 'archive_timeout: 60s' "$root/deploy/kubernetes/production/database.
   echo 'production archive timeout would exceed the certified cross-region bandwidth floor' >&2
   exit 1
 fi
+grep -Fq '  instances: 1' "$root/deploy/kubernetes/production/database.yaml"
+grep -Fq '      easyai.io/site: ningbo' "$root/deploy/kubernetes/production/database.yaml"
+grep -Fq '    synchronous: null' "$root/deploy/kubernetes/production/database.yaml"
 grep -Fq 'wal_sender_timeout: 30s' "$root/deploy/kubernetes/production/database.yaml"
 grep -Fq 'wal_receiver_timeout: 30s' "$root/deploy/kubernetes/production/database.yaml"
 bash -n "$root/scripts/cluster/configure-control-plane-io.sh"