diff --git a/tests/helpers/runtime.py b/tests/helpers/runtime.py
index 975a6ddd2..5c9c8076f 100644
--- a/tests/helpers/runtime.py
+++ b/tests/helpers/runtime.py
@@ -38,6 +38,8 @@ from vllm_omni.outputs import OmniRequestOutput
 
 logger = init_logger(__name__)
 
+SERVER_STARTUP_TIMEOUT_S = 1200
+
 PromptAudioInput = list[tuple[Any, int]] | tuple[Any, int] | None
 PromptImageInput = list[Any] | Any | None
 PromptVideoInput = list[Any] | Any | None
@@ -246,9 +248,10 @@ class OmniServer:
             cwd=_omni_subprocess_cwd(),
         )
 
-        max_wait = 1200
-        start_time = time.time()
-        while time.time() - start_time < max_wait:
+        max_wait = SERVER_STARTUP_TIMEOUT_S
+        # System clock corrections must not shorten or extend startup waits.
+        start_time = time.monotonic()
+        while time.monotonic() - start_time < max_wait:
             ret = self.proc.poll()
             if ret is not None:
                 raise RuntimeError(f"Server processes exited with code {ret} before becoming ready.")
@@ -550,9 +553,10 @@ class OmniServerStageCli(OmniServer):
             for replica_id in range(self.stage_replica_counts.get(stage_id, 1)):
                 self._launch_stage(stage_id, headless=True, replica_id=replica_id)
 
-        max_wait = 1200
-        start_time = time.time()
-        while time.time() - start_time < max_wait:
+        max_wait = SERVER_STARTUP_TIMEOUT_S
+        # System clock corrections must not shorten or extend startup waits.
+        start_time = time.monotonic()
+        while time.monotonic() - start_time < max_wait:
             self._ensure_stage_processes_alive()
             with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as sock:
                 sock.settimeout(1)
diff --git a/tests/helpers/tests/test_runtime_startup_timeout.py b/tests/helpers/tests/test_runtime_startup_timeout.py
new file mode 100644
index 000000000..0eb6a91bf
--- /dev/null
+++ b/tests/helpers/tests/test_runtime_startup_timeout.py
@@ -0,0 +1,90 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+
+"""Startup deadlines must be independent of system clock corrections."""
+
+from types import SimpleNamespace
+from unittest.mock import MagicMock
+
+import pytest
+
+from tests.helpers import runtime
+
+pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
+
+
+@pytest.fixture(params=["server", "stage_cli"])
+def startup_server(request, tmp_path, monkeypatch):
+    monkeypatch.setattr(runtime, "cleanup_test_environment", lambda: None)
+    proc = MagicMock(spec=runtime.subprocess.Popen)
+    proc.poll.return_value = None
+    monkeypatch.setattr(runtime.subprocess, "Popen", lambda *args, **kwargs: proc)
+    if request.param == "server":
+        # Exercise the wait-loop clock without reserving or inspecting real ports.
+        monkeypatch.setattr(runtime.OmniServer, "_reserve_port", lambda self: None)
+        monkeypatch.setattr(runtime.OmniServer, "_owns_listening_port", lambda self: True)
+        return runtime.OmniServer("fake-model", [])
+
+    config = tmp_path / "stages.yaml"
+    config.write_text("stages:\n  - stage_id: 0\n  - stage_id: 1\n")
+    server = runtime.OmniServerStageCli("fake-model", str(config), [])
+
+    def launch(stage_id, *, headless, replica_id=0):
+        server.stage_procs[(stage_id, replica_id)] = proc
+
+    monkeypatch.setattr(server, "_launch_stage", launch)
+    return server
+
+
+@pytest.mark.parametrize("wall_clock_jump", [3600, -3600])
+@pytest.mark.parametrize("becomes_ready", [True, False], ids=["ready", "timeout"])
+def test_startup_deadline_ignores_wall_clock_changes(startup_server, monkeypatch, wall_clock_jump, becomes_ready):
+    elapsed = 0.0
+    wall_offset = 0.0
+    first_probe_at = None
+    probes = 0
+
+    def sleep(seconds):
+        nonlocal elapsed
+        elapsed += seconds
+
+    # Replace only this module's clock, leaving pytest's own timing untouched.
+    monkeypatch.setattr(
+        runtime,
+        "time",
+        SimpleNamespace(
+            time=lambda: 1_700_000_000 + elapsed + wall_offset,
+            monotonic=lambda: elapsed,
+            perf_counter=lambda: elapsed,
+            sleep=sleep,
+        ),
+    )
+
+    def connect(address):
+        nonlocal probes, wall_offset, first_probe_at
+        assert address == (startup_server.host, startup_server.port)
+        if first_probe_at is None:
+            first_probe_at = elapsed
+        # Even a backwards clock correction must not extend the deadline.
+        assert elapsed - first_probe_at < runtime.SERVER_STARTUP_TIMEOUT_S, "startup exceeded its elapsed-time budget"
+        probes += 1
+        if probes == 1:
+            wall_offset += wall_clock_jump
+            return 111
+        return 0 if becomes_ready else 111
+
+    sock = MagicMock()
+    sock.__enter__.return_value = sock
+    sock.connect_ex.side_effect = connect
+    monkeypatch.setattr(runtime.socket, "socket", lambda *args, **kwargs: sock)
+
+    if becomes_ready:
+        startup_server._start_server()
+        assert probes == 2
+        assert first_probe_at is not None
+        assert elapsed - first_probe_at == 2
+    else:
+        with pytest.raises(RuntimeError, match=f"failed to start within {runtime.SERVER_STARTUP_TIMEOUT_S} seconds"):
+            startup_server._start_server()
+        assert first_probe_at is not None
+        assert elapsed - first_probe_at == runtime.SERVER_STARTUP_TIMEOUT_S