fix: avoid tracked mojibake literals

Author: Henry

backend_api_python/app/professional_report/narrative.py

@@ -7,17 +7,29 @@
 from typing import Any, Iterable, Mapping
 
 
-_MOJIBAKE = ("�", "—", "–", "’", "锟斤拷", "�")
+_MOJIBAKE = (
+    "\u00ef\u00bf\u00bd",
+    "\u00e2\u20ac\u201d",
+    "\u00e2\u20ac\u201c",
+    "\u00e2\u20ac\u2122",
+    "\u951f\u65a4\u62f7",
+    "\ufffd",
+)
 
 
 def normalize_report_text(value: Any) -> Any:
     """Normalize Unicode and replace known mojibake without changing numbers."""
     if isinstance(value, str):
         text = unicodedata.normalize("NFKC", value).replace("\x00", "")
-        replacements = {"—": "—", "–": "–", "’": "’", "�": ""}
+        replacements = {
+            "\u00e2\u20ac\u201d": "—",
+            "\u00e2\u20ac\u201c": "–",
+            "\u00e2\u20ac\u2122": "’",
+            "\u00ef\u00bf\u00bd": "",
+        }
         for broken, fixed in replacements.items():
             text = text.replace(broken, fixed)
-        return text.replace("�", "").strip()
+        return text.replace("\ufffd", "").strip()
     if isinstance(value, dict):
         return {key: normalize_report_text(item) for key, item in value.items()}
     if isinstance(value, list):

backend_api_python/tests/test_professional_report_builder.py

@@ -75,7 +75,7 @@ def _analysis(payload, evidence_claims=None):
         "language": "zh-CN",
         "decision": "BUY",
         "confidence": 88,
-        "summary": "趋势改善—但仍需确认�",
+        "summary": "趋势改善\u00e2\u20ac\u201d但仍需确认\ufffd",
         "timeframe": "medium",
         "detailed_analysis": {"technical": "动量改善", "fundamental": "数据可用", "sentiment": "中性"},
         "scores": {"technical": 68, "fundamental": 60, "sentiment": 52},
@@ -102,8 +102,8 @@ def test_professional_builder_produces_valid_contract_for_supported_markets(mark
     assert report["contract_validation"]["valid"] is True
     assert report["instrument"]["market"] == market
     assert report["claims"][0]["evidence_refs"] == [ref]
-    assert "—" not in report["executive_summary"]
-    assert "�" not in report["executive_summary"]
+    assert "\u00e2\u20ac\u201d" not in report["executive_summary"]
+    assert "\ufffd" not in report["executive_summary"]
     assert report["data_quality"]["coverage_ratio"] == 1