diff --git a/tests/test_import_hygiene.py b/tests/test_import_hygiene.py
index 4129e348..43d85977 100644
--- a/tests/test_import_hygiene.py
+++ b/tests/test_import_hygiene.py
@@ -11,20 +11,23 @@
 
 import subprocess
 import sys
+import textwrap
 
-_PROBE = """
-import sys
-import aai_cli.main
-heavy = {"jiwer", "rapidfuzz"}
-loaded = sorted(heavy & {name.split(".")[0] for name in sys.modules})
-assert not loaded, f"CLI import eagerly loaded: {loaded}"
-"""
+_PROBE = textwrap.dedent(
+    """
+    import sys
+    import aai_cli.main
+    heavy = {"jiwer", "rapidfuzz"}
+    loaded = sorted(heavy & {name.split(".")[0] for name in sys.modules})
+    assert not loaded, f"CLI import eagerly loaded: {loaded}"
+    """
+).strip()
 
 
 def test_cli_import_does_not_load_eval_scoring_stack():
     # A fresh interpreter: the suite's own process has long since imported the
     # scoring stack, so sys.modules can only be probed in a child.
     result = subprocess.run(
-        [sys.executable, "-c", _PROBE], capture_output=True, text=True, timeout=120, check=False
+        [sys.executable, "-c", _PROBE], capture_output=True, text=True, timeout=15, check=False
     )
     assert result.returncode == 0, result.stderr