diff --git a/sdk/app/services/normalize.py b/sdk/app/services/normalize.py
index fc3d40d..0fd94b6 100644
--- a/sdk/app/services/normalize.py
+++ b/sdk/app/services/normalize.py
@@ -20,6 +20,16 @@ def pick(obj: dict, *names: str, default=None):
 # ---------------------------------------------------------------------------
 # §8 Source 归一化（缺失字段键必须存在，值为 null）
 # ---------------------------------------------------------------------------
+# title 回填优先级（缺陷 #27，2026-09-11）：title/paper_title 缺失时，
+# 依次从以下字段取可读文本（论文标题优先于实体名），全部缺失才保持 null。
+# 排除 content/evidence/text/snippet 正文字段——正文片段不是标题。
+_TITLE_FALLBACK_TOPLEVEL = (
+    "publication_title", "article_title", "journal_title", "name",
+)
+_TITLE_FALLBACK_METADATA = (
+    "paper_title", "publication_title", "article_title", "journal_title",
+    "source_entity", "name", "source",
+)
 
 def normalize_source(item: dict[str, Any] | None) -> dict[str, Any]:
     """单个 source item → API.md §8 形态（所有键存在，缺失为 null）。
@@ -41,6 +51,23 @@ def normalize_source(item: dict[str, Any] | None) -> dict[str, Any]:
 
     source = pick(item, "source")            # 断言主体实体名（上游语义）
     title = pick(item, "title", "paper_title")
+    # 缺陷 #27：title 为空时按固定优先级回填——
+    # 顶层标题类字段 → metadata 标题类字段 → 顶层 source（断言主体实体名）。
+    # 正文字段绝不参与（正文片段不是标题）。
+    if not title:
+        for _key in _TITLE_FALLBACK_TOPLEVEL:
+            _candidate = pick(item, _key)
+            if _candidate:
+                title = _candidate
+                break
+    if not title and isinstance(pick(item, "metadata"), dict):
+        for _key in _TITLE_FALLBACK_METADATA:
+            _candidate = pick(item["metadata"], _key)
+            if _candidate:
+                title = _candidate
+                break
+    if not title:
+        title = source
 
     doi = pick(item, "doi")
     url = pick(item, "url", "doi_url")
diff --git a/sdk/app/services/test_normalize.py b/sdk/app/services/test_normalize.py
index 8723245..f51a207 100644
--- a/sdk/app/services/test_normalize.py
+++ b/sdk/app/services/test_normalize.py
@@ -84,6 +84,52 @@ def test_normalize_source_only_doi_derives_url():
     assert src["metadata"] == {}
 
 
+# ---------------------------------------------------------------------------
+# §8 title 回填（缺陷 #27，2026-09-11）
+# ---------------------------------------------------------------------------
+
+def test_normalize_source_title_fill_from_source_entity():
+    """title 为空 -> 回填 source（断言主体实体名），正文片段绝不参与。"""
+    src = normalize_source({
+        "source": "OsSPL14",
+        "title": "",
+        "evidence": "## 正文片段开头（不是标题）",
+        "paper_id": "P1",
+    })
+    assert src["title"] == "OsSPL14"
+    # title 即 source 回填：两者相等，source_entity 不落 metadata（不重复）
+    assert src["metadata"].get("source_entity") is None
+    assert src["paper_id"] == "P1"
+
+
+def test_normalize_source_title_fill_from_metadata():
+    """title/paper_title 均空 -> metadata 标题类字段回填。"""
+    src = normalize_source({
+        "source": "entity",
+        "metadata": {"paper_title": "水稻抗病基因研究进展", "source_entity": "entity"},
+    })
+    assert src["title"] == "水稻抗病基因研究进展"
+
+
+def test_normalize_source_title_not_overridden():
+    """title 非空 -> 不回填（既有语义不变）。"""
+    src = normalize_source({
+        "title": "Real Title",
+        "source": "Entity",
+        "paper_title": "Other",
+        "metadata": {"paper_title": "Meta"},
+    })
+    assert src["title"] == "Real Title"
+    assert src["metadata"]["source_entity"] == "Entity"
+
+
+def test_normalize_source_title_all_empty_stays_none():
+    """全空 -> 保持 null（前端有 fallback）。"""
+    src = normalize_source({"content": "snippet only", "paper_id": "P1"})
+    assert src["title"] is None
+    assert src["snippet"] == "snippet only"
+
+
 def test_normalize_source_alias_chains():
     src = normalize_source({
         "content": "snippet text",