fix(curator): query Douban only by Chinese title or ISBN, never a Latin title

Douban's English-language book records are sparse and low quality. BookPageProvider now picks a CJK title from the title or aliases for Douban (and still allows a precise ISBN), and skips Douban entirely when only a Latin title is available; Goodreads continues to cover Latin titles. Aliases now reach the page search so a translated book's Chinese title is used. New unit test covers the four cases; ISBN cross-title matching is preserved.

scenarios/curator/backend/curator/book_pages.py

@@ -10,6 +10,8 @@ from typing import Any
 
 from .media_catalog import title_key
 
+_CJK = re.compile(r"[\u3400-\u4dbf\u4e00-\u9fff\uf900-\ufaff]")
+
 
 class _SearchParser(HTMLParser):
     def __init__(self) -> None:
@@ -89,10 +91,27 @@ class BookPageProvider:
         "goodreads.com": "goodreads",
     }
 
-    def search(self, title: str, author: str = "", isbn: str = "", limit: int = 6) -> list[dict[str, Any]]:
-        terms = " ".join(value for value in (f'"{title}"' if title else "", author, isbn) if value)
+    def search(self, title: str, author: str = "", isbn: str = "", limit: int = 6,
+               aliases: list[str] | None = None) -> list[dict[str, Any]]:
+        candidates = [title, *(aliases or [])]
+        douban_title = next((c.strip() for c in candidates if c and _CJK.search(c)), "")
+        goodreads_terms = " ".join(value for value in (f'"{title}"' if title else "", author, isbn) if value)
         urls: list[str] = []
         for provider in ("douban", "goodreads"):
+            if provider == "douban":
+                # Douban's English-language records are sparse and low quality, so it
+                # is queried only with a Chinese title (or a precise ISBN), never with
+                # a Latin title. Without either, Goodreads covers the book alone.
+                parts = [f'"{douban_title}"'] if douban_title else []
+                if author and _CJK.search(author):
+                    parts.append(author)
+                if isbn:
+                    parts.append(isbn)
+                if not parts:
+                    continue
+                terms = " ".join(parts)
+            else:
+                terms = goodreads_terms
             for url in self._search_site(provider, terms):
                 if url not in urls:
                     urls.append(url)

scenarios/curator/backend/curator/book_reviews.py

@@ -32,8 +32,8 @@ class BookReviewProvider:
         )
         return hashlib.sha256(value.encode("utf-8")).hexdigest()
 
-    def _public_pages(self, title: str, author: str, isbn: str) -> list[dict[str, Any]]:
-        return BookPageProvider().search(title, author, isbn, 8)
+    def _public_pages(self, title: str, author: str, isbn: str, aliases: list[str]) -> list[dict[str, Any]]:
+        return BookPageProvider().search(title, author, isbn, 8, aliases=aliases)
 
     @staticmethod
     def _text_key(value: str) -> str:
@@ -95,7 +95,7 @@ class BookReviewProvider:
         for name, loader in (("douban-goodreads-pages", self._public_pages),):
             result["providers_checked"].append(name)
             try:
-                result["results"].extend(self._matched_results(loader(title, author, isbn), title, aliases, author, isbn))
+                result["results"].extend(self._matched_results(loader(title, author, isbn, aliases), title, aliases, author, isbn))
             except Exception as exc:
                 result["errors"].append(f"{name}: {exc}")
         if result["results"]:

scenarios/curator/backend/tests/test_curator.py

@@ -368,6 +368,35 @@ class CuratorTest(unittest.TestCase):
             "非洲运转之道", "乔·斯塔威尔", "978-1-84668-407-9",
         ))
 
+    def test_douban_is_queried_only_with_a_chinese_title(self) -> None:
+        provider = BookPageProvider()
+        calls: list[tuple[str, str]] = []
+        provider._search_site = lambda prov, terms: calls.append((prov, terms)) or []  # type: ignore[method-assign]
+        # English-only title, no ISBN, no alias: Douban is skipped, Goodreads runs.
+        calls.clear()
+        provider.search("Clean Code", "Robert Martin")
+        self.assertEqual([p for p, _ in calls], ["goodreads"])
+        # A Chinese alias routes Douban to the Chinese title, not the English one.
+        calls.clear()
+        provider.search("Sapiens", "Yuval Noah Harari", aliases=["人类简史"])
+        douban = [t for p, t in calls if p == "douban"]
+        self.assertEqual(len(douban), 1)
+        self.assertIn("人类简史", douban[0])
+        self.assertNotIn("Sapiens", douban[0])
+        # A precise ISBN still reaches Douban even without a Chinese title.
+        calls.clear()
+        provider.search("How Africa Works", "", isbn="9781846684079")
+        douban = [t for p, t in calls if p == "douban"]
+        self.assertEqual(len(douban), 1)
+        self.assertIn("9781846684079", douban[0])
+        self.assertNotIn("How Africa Works", douban[0])
+        # A Chinese title is used directly (no regression).
+        calls.clear()
+        provider.search("三体", "刘慈欣")
+        douban = [t for p, t in calls if p == "douban"]
+        self.assertEqual(len(douban), 1)
+        self.assertIn("三体", douban[0])
+
     def test_book_web_review_falls_back_and_caches_attributed_evidence(self) -> None:
         provider = BookWebReviewProvider(self.settings, self.database)
         calls: list[str] = []