diff --git a/agent/qa/src/rag/qa_v1_1_73/direct_flow_retriever.py b/agent/qa/src/rag/qa_v1_1_73/direct_flow_retriever.py
index 56bb4d9..8ae9ceb 100644
--- a/agent/qa/src/rag/qa_v1_1_73/direct_flow_retriever.py
+++ b/agent/qa/src/rag/qa_v1_1_73/direct_flow_retriever.py
@@ -117,6 +117,7 @@ class DirectFlowRetriever:
         use_qdrant: bool = True,
         use_neo4j: bool = True,
         use_graph_expansion: bool = True,
+        qdrant_score_threshold: float | None = None,
     ) -> Tuple[List[EvidenceItem], Dict[str, Any]]:
         effective_use_neo4j = bool(use_neo4j) and self.graph_schema in {
             "canonical",
@@ -138,6 +139,7 @@ class DirectFlowRetriever:
                 "use_graph_expansion": use_graph_expansion,
                 "effective_use_neo4j": effective_use_neo4j,
                 "effective_use_graph_expansion": effective_use_graph_expansion,
+                "qdrant_score_threshold": qdrant_score_threshold,
             },
             "neo4j_status": (
                 "enabled"
@@ -153,7 +155,12 @@ class DirectFlowRetriever:
             per_query = max(10, min(qdrant_top_k // max(1, len(all_queries)), 50))
             if self.vector_query_workers == 1 or len(all_queries) == 1:
                 query_rows = [
-                    (query, self.qdrant_search(query, per_query))
+                    (
+                        query,
+                        self.qdrant_search(
+                            query, per_query, score_threshold=qdrant_score_threshold
+                        ),
+                    )
                     for query in all_queries
                 ]
             else:
@@ -162,7 +169,12 @@ class DirectFlowRetriever:
                     thread_name_prefix="qa-vector",
                 ) as executor:
                     futures = [
-                        executor.submit(self.qdrant_search, query, per_query)
+                        executor.submit(
+                            self.qdrant_search,
+                            query,
+                            per_query,
+                            score_threshold=qdrant_score_threshold,
+                        )
                         for query in all_queries
                     ]
                     query_rows = [
diff --git a/agent/qa/src/rag/qa_v1_1_73/test_score_gate.py b/agent/qa/src/rag/qa_v1_1_73/test_score_gate.py
index eb71835..bc82e64 100644
--- a/agent/qa/src/rag/qa_v1_1_73/test_score_gate.py
+++ b/agent/qa/src/rag/qa_v1_1_73/test_score_gate.py
@@ -224,6 +224,46 @@ def test_chat_mode_passes_no_threshold():
     assert p._retriever_stub.last_kwargs["qdrant_score_threshold"] is None
 
 
+def test_direct_flow_retriever_threshold_passthrough():
+    """真实 retriever 的 qdrant_score_threshold 参数链路（不触发 IO）。
+
+    用对象伪造构造 DirectFlowRetriever 实例，直接调用 qdrant_search：
+    - 传阈值时透传给 qdrant.query_points（同域向量阈值）；
+    - 不传阈值时 score_threshold=None（chat 保召回）。
+    """
+    import rag.qa_v1_1_73.direct_flow_retriever as dfr_mod
+
+    class _FakeQdrant:
+        def __init__(self):
+            self.calls = []
+
+        def query_points(self, **kwargs):
+            self.calls.append(kwargs)
+            from qdrant_client.http import models
+
+            return models.QueryResponse(points=[])
+
+    class _MinimalRetriever(dfr_mod.DirectFlowRetriever):
+        """绕过 __init__（不建连接），只测 qdrant_search 参数透传。"""
+
+        def __init__(self):
+            self.qdrant = _FakeQdrant()
+            self.collection = "plant_srt_pc2025_v3_20260817_normalized"
+            self.embed_url = ""
+            self.embed_model = ""
+
+        def embed(self, text):
+            return [0.0] * 1024
+
+    r = _MinimalRetriever()
+    r.qdrant_search("景甜是谁", 10, score_threshold=0.45)
+    assert r.qdrant.calls[-1]["score_threshold"] == 0.45
+    assert r.qdrant.calls[-1]["collection_name"] == r.collection
+
+    r.qdrant_search("arginine", 10)
+    assert r.qdrant.calls[-1]["score_threshold"] is None
+
+
 def test_env_overrides(monkeypatch):
     monkeypatch.setenv("QA_MIN_VECTOR_SCORE", "0.5")
     assert score_gate._env_value("QA_MIN_VECTOR_SCORE", 0.45) == 0.5