diff --git a/custom_components/localthings/coordinator.py b/custom_components/localthings/coordinator.py
index 64445f5..692fa22 100644
--- a/custom_components/localthings/coordinator.py
+++ b/custom_components/localthings/coordinator.py
@@ -834,6 +834,20 @@ class LocalThingsCoordinator(DataUpdateCoordinator[dict[str, Any]]):
                     resources = await self.hass.async_add_executor_job(self._poll_once)
                 except Exception as e2:
                     self._log.error("poll failed after reconnect: %s", e2)
+                    # The reconnect itself proved the old session is gone, so
+                    # any OBSERVE subscription on it is too -- this cycle has
+                    # no live session to resubscribe on, unlike the success
+                    # branch below, so downgrade without setting
+                    # just_downgraded_from_observe (that would attempt a
+                    # resubscribe this same cycle against a session we just
+                    # confirmed is unreachable). Without this, a device that
+                    # drops off the network entirely leaves the connection-mode
+                    # sensor reporting "Push" forever, since only the success
+                    # path below ever changed it (issue #287) -- the poll-mode
+                    # retry timer (_maybe_retry_observe_mode) still re-attempts
+                    # observe once the device is actually reachable again.
+                    if self._observe.mode == MODE_OBSERVE:
+                        self._observe.downgrade_to_poll()
                     snapshot = self._cache.snapshot()
                     # Same precondition as _defer_reconnect_for (issue #254):
                     # degraded-but-successful data only makes sense once
diff --git a/tests/localthings/test_coordinator.py b/tests/localthings/test_coordinator.py
index 1711150..3338079 100644
--- a/tests/localthings/test_coordinator.py
+++ b/tests/localthings/test_coordinator.py
@@ -478,6 +478,57 @@ async def test_reconnect_while_observe_mode_downgrades_to_poll(
     assert coordinator._observe.mode == MODE_POLL
 
 
+async def test_total_poll_failure_downgrades_observe_mode_to_poll(
+    hass: HomeAssistant, mock_entry, mock_coordinator_observe_session, fridge_resources
+) -> None:
+    """A device that drops off the network entirely -- both the poll and its
+    reconnect retry fail -- must not leave the connection-mode sensor
+    reporting 'Push' forever (issue #287). Only the *successful* reconnect
+    branch used to touch observe mode (see
+    test_reconnect_while_observe_mode_downgrades_to_poll); this covers the
+    branch where the device stays unreachable."""
+    fake = mock_coordinator_observe_session
+    await hass.config_entries.async_setup(mock_entry.entry_id)
+    await hass.async_block_till_done()
+
+    coordinator: LocalThingsCoordinator = hass.data[DOMAIN][mock_entry.entry_id]
+    hrefs = coordinator._hot_hrefs + coordinator._warm_hrefs
+
+    fake.notify_on_subscribe = {"notified": True}
+    entered = await hass.async_add_executor_job(
+        coordinator._observe.try_enter_observe_mode,
+        fake,
+        hrefs,
+        0.02,
+        0.8,
+    )
+    assert entered is True
+    assert coordinator.observe_mode == MODE_OBSERVE
+
+    last_notify_ts = coordinator._observe._last_notify_ts
+    assert last_notify_ts is not None
+    coordinator._observe._last_notify_ts = last_notify_ts - (PUSH_HEALTH_WINDOW_S + 1)
+
+    with (
+        patch(
+            "custom_components.localthings.coordinator.LocalThingsCoordinator._poll_once",
+            side_effect=[RuntimeError("connection lost"), RuntimeError("still lost")],
+        ),
+        patch(
+            "custom_components.localthings.coordinator.asyncio.sleep",
+            new=AsyncMock(),
+        ),
+    ):
+        await coordinator.async_request_refresh()
+        await hass.async_block_till_done()
+
+    # The update still "succeeds" with the last-known snapshot (issue #254's
+    # degraded-data path) -- but the connection mode must reflect reality
+    # now, not the stale OBSERVE state from before the outage.
+    assert coordinator.observe_mode == MODE_POLL
+    assert coordinator.last_update_success is True
+
+
 async def test_poll_timeout_skips_reconnect_when_push_is_healthy(
     hass: HomeAssistant, mock_entry, mock_coordinator_observe_session
 ) -> None: