Fix Issue 4: Make mock heuristic detection more resilient

patches/validate_mock_fix.patch

@@ -0,0 +1,11 @@
+--- a/benchmarks/validate.py
++++ b/benchmarks/validate.py
+@@ -107,7 +107,7 @@
+ def make_mock(task_name):
+     mock = MockLLMBackend()
+     t = TASKS[task_name]
+     def actor(msgs):
+         text = " ".join(m.content for m in msgs)
+-        has_h = "Learned Strategies" in text and "None yet" not in text
++        has_h = ("Learned Strategies" in text or "When:" in text) and "None yet" not in text
+         code = t["good"] if has_h else t["bad"]