Fix HPO gate pass reporting

src/eval/inference/hpo_search_baselines.py

@@ -359,12 +359,14 @@ def primary_metric(metrics: dict[str, Any], scenario: str) -> tuple[float, str |
 
 
 def gate_passed(metrics: dict[str, Any]) -> bool:
+    if metrics.get("error"):
+        return False
     if metrics.get("score") == 0:
         return False
-    qc = metrics.get("quality_check") or {}
-    if qc.get("pass") is False:
+    qc = metrics.get("quality_check")
+    if not isinstance(qc, dict):
         return False
-    return True
+    return qc.get("pass") is True
 
 
 def _run_text(cmd: list[str], timeout_s: float = 20.0) -> str:

tests/test_hpo_search_baselines.py

@@ -142,3 +142,15 @@ def test_gate_passed_ignores_legacy_success_gate_field():
 def test_gate_passed_still_fails_quality_and_zero_score():
     assert not hpo.gate_passed({"score": 0, "quality_check": {"pass": True}})
     assert not hpo.gate_passed({"quality_check": {"pass": False}})
+
+
+def test_gate_passed_requires_explicit_successful_quality_check():
+    assert hpo.gate_passed({"quality_check": {"pass": True}})
+    assert not hpo.gate_passed({})
+    assert not hpo.gate_passed({"quality_check": {}})
+    assert not hpo.gate_passed(
+        {
+            "error": "evaluate.py exited with code 1",
+            "quality_check": {"pass": True},
+        }
+    )