fix(dictionary): strip Latin romanization debris from Russian translations and sanitize dictionary cache

app/database.py

@@ -432,6 +432,22 @@ def repair_mismatched_cefr_contexts(conn: sqlite3.Connection):
     except Exception:
         pass
 
+    # Clean up any leftover Latin transliteration artifacts from offline_dictionary
+    try:
+        dirty_check = conn.execute(
+            "SELECT 1 FROM offline_dictionary WHERE translation GLOB '*[a-zA-Záéíóúýšžčñ/\\\\]*' LIMIT 1"
+        ).fetchone()
+        if dirty_check:
+            rows = conn.execute(
+                "SELECT rowid, translation FROM offline_dictionary WHERE translation GLOB '*[a-zA-Záéíóúýšžčñ/\\\\]*'"
+            ).fetchall()
+            from app.dictionary import clean_russian_translation
+            cleaned = [(clean_russian_translation(r[1]), r[0]) for r in rows]
+            conn.executemany("UPDATE offline_dictionary SET translation = ? WHERE rowid = ?", cleaned)
+            conn.execute("DELETE FROM dictionary_cache")
+    except Exception:
+        pass
+
 
 def _seed_offline_dictionary(conn: sqlite3.Connection):
     """Seeds the offline dictionary from data/offline_dict.tsv.gz if the table is empty."""
@@ -452,12 +468,14 @@ def _seed_offline_dictionary(conn: sqlite3.Connection):
         return
 
     import gzip
+    from app.dictionary import clean_russian_translation
 
     batch = []
     with gzip.open(archive_path, "rt", encoding="utf-8") as f:
         for line in f:
             parts = line.rstrip("\r\n").split("\t")
             if len(parts) == 4:
+                parts[1] = clean_russian_translation(parts[1])
                 batch.append(parts)
                 if len(batch) >= 10000:
                     conn.executemany(