diff --git a/gateway/run.py b/gateway/run.py
index bcee403b1ce..2071950df9d 100644
--- a/gateway/run.py
+++ b/gateway/run.py
@@ -3595,6 +3595,13 @@ class GatewayRunner(GatewayAuthorizationMixin, GatewayKanbanWatchersMixin, Gatew
         """
         if any(not t.done() for t in self._background_tasks):
             return True
+        try:
+            from tools.async_delegation import active_count
+
+            if active_count() > 0:
+                return True
+        except Exception:  # noqa: BLE001 - never let the idle check raise
+            logger.debug("scale-to-zero async-delegation check failed", exc_info=True)
         try:
             from tools.process_registry import process_registry
 
@@ -3653,6 +3660,23 @@ class GatewayRunner(GatewayAuthorizationMixin, GatewayKanbanWatchersMixin, Gatew
             has_live_background_work=self._scale_to_zero_has_live_background_work(),
         )
 
+    def _scale_to_zero_note_real_inbound(self) -> None:
+        """Stamp real inbound and restore lifecycle after a dormant wake.
+
+        The watcher marks runtime status `draining` as it quiesces the relay, but
+        dormancy is not the stop/restart drain path: the process remains alive and
+        should present as running once real traffic wakes it and re-enters the
+        gateway. Internal completion/replay events intentionally do not call this
+        helper, so they do not keep an otherwise idle gateway awake.
+        """
+        self._last_inbound_at = time.time()
+        if getattr(self, "_scale_to_zero_cooldown_until", 0.0) > 0:
+            try:
+                self._update_runtime_status("running")
+            except Exception:  # noqa: BLE001 - status restoration is best-effort
+                logger.debug("scale-to-zero: status restore failed", exc_info=True)
+            self._scale_to_zero_cooldown_until = 0.0
+
     def _relay_adapter_for_dormancy(self):
         """Return the connected RELAY adapter, if any (the one go_dormant targets)."""
         try:
@@ -7504,7 +7528,7 @@ class GatewayRunner(GatewayAuthorizationMixin, GatewayKanbanWatchersMixin, Gatew
         # traffic — counting them would keep a genuinely idle gateway awake. This
         # clock is what the idle predicate (gateway/scale_to_zero.is_idle) reads.
         if not is_internal:
-            self._last_inbound_at = time.time()
+            self._scale_to_zero_note_real_inbound()
 
         # Fire pre_gateway_dispatch plugin hook for user-originated messages.
         # Plugins receive the MessageEvent and may return a dict influencing flow:
diff --git a/tests/gateway/test_scale_to_zero_watcher.py b/tests/gateway/test_scale_to_zero_watcher.py
index 5048928b285..bb959e810c6 100644
--- a/tests/gateway/test_scale_to_zero_watcher.py
+++ b/tests/gateway/test_scale_to_zero_watcher.py
@@ -113,6 +113,37 @@ def test_bg_work_blocks_idle_via_background_tasks(monkeypatch):
         loop.close()
 
 
+def test_bg_work_blocks_idle_via_async_delegation(monkeypatch):
+    """delegate_task(background=true) lives in tools.async_delegation, not the
+    process registry. An active background delegation must block suspend too."""
+    r = GatewayRunner.__new__(GatewayRunner)
+    r._background_tasks = set()
+
+    monkeypatch.setattr("tools.async_delegation.active_count", lambda: 1)
+
+    assert r._scale_to_zero_has_live_background_work() is True
+
+
+def test_real_inbound_after_dormancy_restores_running_status(monkeypatch):
+    """Once a dormant gateway receives real inbound after wake, the runtime
+    lifecycle must not remain stuck in the watcher-written `draining` state."""
+    r = GatewayRunner.__new__(GatewayRunner)
+    r._last_inbound_at = 0.0
+    r._scale_to_zero_cooldown_until = time.time() + 60.0
+    status_updates = []
+    monkeypatch.setattr(
+        r,
+        "_update_runtime_status",
+        lambda state=None, *a, **k: status_updates.append(state),
+        raising=False,
+    )
+
+    r._scale_to_zero_note_real_inbound()
+
+    assert r._last_inbound_at > 0.0
+    assert status_updates == ["running"]
+
+
 def test_bg_work_false_when_quiet():
     r = GatewayRunner.__new__(GatewayRunner)
     r._background_tasks = set()