diff --git a/backend/apps/outputs/outputs.py b/backend/apps/outputs/outputs.py
index a5c71936..478f3f2c 100644
--- a/backend/apps/outputs/outputs.py
+++ b/backend/apps/outputs/outputs.py
@@ -476,7 +476,11 @@ def runtime_status_payload(workspace_id: str, instance: int = 1) -> dict:
         "backend_url": f"http://127.0.0.1:{rt.port}" if rt.running and rt.port else None,
         # New-mode only: where the Vite dev server is reachable. Old-mode workspaces report null and the editor falls back to the legacy /api/outputs/workspace/{ws}/serve/... path.
         "frontend_port": rt.frontend_port,
-        "frontend_url": rt.frontend_url if rt.running else None,
+        # Ask the property, do not re-gate on `running`: a serve-static app HAS no process, so the
+        # extra gate blanked its URL and the preview pane had nothing to navigate to, which is the
+        # "app loads forever until I open a second instance" report (a second instance skips serve
+        # mode entirely). The property already handles ready/suspended/dead-vite itself.
+        "frontend_url": rt.frontend_url,
         "is_new_mode": rt.is_new_mode,
         # The boot narration used to be WS-only, so a wedged boot left no queryable trace and one 120s field wedge stayed a shrug (ENG-305); the tail makes any status poll name what the boot is doing.
         "recent_log": [getattr(line, "text", str(line))[-200:] for line in list(rt.log_buffer)[-12:]],
diff --git a/backend/tests/test_serve_static_reports_its_url.py b/backend/tests/test_serve_static_reports_its_url.py
new file mode 100644
index 00000000..ab5f43b2
--- /dev/null
+++ b/backend/tests/test_serve_static_reports_its_url.py
@@ -0,0 +1,41 @@
+"""A serve-static app must report a URL (found 2026-08-14 by a long-run user simulation, and it is
+the "app loads forever until I open a second instance" report).
+
+ENG-209 lets a fresh built bundle be served with NO process. `running` is defined as "has a live
+process", so it is False for those by design, and the status payload re-gated `frontend_url` on
+`running`: the API said there was no URL, the preview pane had nothing to navigate to, and the app
+spun forever. A SECOND instance skips serve mode entirely (`instance == 1` gate), which is exactly
+why opening another window "fixed" it. Measured live: one app in the corpus was permanently
+unopenable this way; after the fix it serves real HTML.
+"""
+
+import inspect
+
+from backend.apps.outputs import outputs as outputs_mod
+from backend.apps.outputs.runtime import AppRuntime
+
+
+def test_status_asks_the_property_instead_of_re_gating_on_running():
+    src = inspect.getsource(outputs_mod.runtime_status_payload)
+    line = next(ln for ln in src.splitlines() if '"frontend_url"' in ln and "None," not in ln)
+    assert "rt.running" not in line, (
+        "re-gating frontend_url on `running` blanks it for serve-static apps, which have no "
+        "process by design; the property already handles ready/suspended/dead-vite"
+    )
+    assert "rt.frontend_url" in line
+
+
+def test_the_property_itself_answers_for_serve_static():
+    src = inspect.getsource(AppRuntime.frontend_url.fget)
+    i_serve = src.find("self.serve_static")
+    i_running = src.find("self.running")
+    assert i_serve != -1, "the property must special-case serve mode"
+    assert i_running == -1 or i_serve < i_running, (
+        "serve mode must be answered BEFORE any running check, or the processless path returns None"
+    )
+
+
+def test_serve_mode_is_still_gated_to_the_primary_instance():
+    # The second-instance escape hatch is what made the bug survivable; keep it working.
+    src = inspect.getsource(AppRuntime.p_start_new_mode)
+    assert "self.instance == 1" in src