fix(recurring): group by merchant and account, not merchant alone

src/data/recurring_detector.py

@@ -27,23 +27,27 @@ def detect_recurring(
     """
     cutoff = date.today() - timedelta(days=lookback_days)
 
-    # Group by merchant name
-    by_merchant: dict[str, list[dict]] = defaultdict(list)
+    # Group by (merchant name, account id). A merchant string like "Ameriprise"
+    # can cover independent recurring streams on different accounts (e.g. two
+    # household members' insurance debits) — merging them produces a polluted
+    # bag whose median amount and average interval describe neither stream.
+    by_group: dict[tuple[str, str], list[dict]] = defaultdict(list)
     for txn in transactions:
         merchant = (txn.get("merchant") or {}).get("name", "")
         if not merchant:
             continue
+        account_id = (txn.get("account") or {}).get("id", "")
         txn_date_str = txn.get("date", "")
         try:
             txn_date = date.fromisoformat(txn_date_str[:10])
         except (ValueError, TypeError):
             continue
         if txn_date < cutoff:
             continue
-        by_merchant[merchant].append(txn)
+        by_group[(merchant, account_id)].append(txn)
 
     items: list[RecurringItem] = []
-    for merchant, txns in by_merchant.items():
+    for (merchant, _account_id), txns in by_group.items():
         if len(txns) < min_occurrences:
             continue
 

tests/test_recurring_detector.py

@@ -86,6 +86,28 @@ def test_detects_biweekly(self):
         assert items[0].frequency == "biweekly"
         assert items[0].amount == 2000.0
 
+    def test_same_merchant_on_different_accounts_stays_split(self):
+        # Two genuinely-monthly Ameriprise streams on different accounts —
+        # one on the 16th for $1100, one on the 25th for $1000. Grouping by
+        # merchant alone would interleave them into a bag whose median is
+        # $1050 and whose avg interval lands in the biweekly bucket.
+        today = date.today()
+        txns = [
+            _make_txn("Ameriprise", -1000.0, today - timedelta(days=89), account_id="karen"),
+            _make_txn("Ameriprise", -1100.0, today - timedelta(days=66), account_id="rex"),
+            _make_txn("Ameriprise", -1000.0, today - timedelta(days=58), account_id="karen"),
+            _make_txn("Ameriprise", -1100.0, today - timedelta(days=39), account_id="rex"),
+            _make_txn("Ameriprise", -1000.0, today - timedelta(days=30), account_id="karen"),
+            _make_txn("Ameriprise", -1100.0, today - timedelta(days=8), account_id="rex"),
+        ]
+        items = detect_recurring(txns)
+        assert len(items) == 2
+        by_account = {item.account_id: item for item in items}
+        assert by_account["karen"].frequency == "monthly"
+        assert by_account["karen"].amount == -1000.0
+        assert by_account["rex"].frequency == "monthly"
+        assert by_account["rex"].amount == -1100.0
+
     def test_includes_account_info(self):
         today = date.today()
         txns = [