fix: track new transactions explicitly for post-import categorization

session.new is emptied by auto-flush during duplicate detection queries,
so we track new Transaction objects in a list instead.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/services/importer.py

@@ -5,6 +5,7 @@ from sqlalchemy import func
 from sqlalchemy.orm import Session
 
 from src.models.transaction import Transaction
+from src.services.categorizer import RuleBasedCategorizer
 from src.services.csv_reader import read_csv
 from src.services.normalizer import normalize_description
 
@@ -23,6 +24,7 @@ class ImportService:
         rows = read_csv(file_path)
         imported = 0
         duplicates = 0
+        new_transactions: list[Transaction] = []
 
         # Track how many times each key appears in the current batch so that
         # legitimate repeated transactions (same date/amount/description) in a
@@ -65,8 +67,18 @@ class ImportService:
                 source_category=source_cat,
             )
             self.session.add(txn)
+            new_transactions.append(txn)
             imported += 1
 
+        categorizer = RuleBasedCategorizer(self.session)
+        for txn in new_transactions:
+            if txn.category_id is None:
+                result = categorizer.categorize(txn)
+                if result:
+                    txn.category_id = result.category_id
+                    txn.tag = result.tag
+                    txn.attributed_to_id = result.attributed_to_id
+
         self.session.commit()
         return {"imported": imported, "duplicates": duplicates, "total_rows": len(rows)}