fix(dedup): scope content hashes by account

README.md

@@ -71,8 +71,9 @@ positive) regardless of how the source expressed direction.
   `fitid`, `account_id`, and currency.
 - De-duplicates transactions with deterministic heuristics:
   - Bank-assigned `FITID` (from OFX) is the authoritative key when present.
-  - Otherwise a content hash over (date, signed amount, currency, canonical
-    description), while preserving legitimately-repeated same-day charges.
+  - Otherwise a content hash over (account ID, date, signed amount, currency,
+    canonical description), while preserving legitimately-repeated same-day
+    charges.
 - Merges multiple statements (e.g. overlapping monthly exports) into one
   de-duplicated list.
 - Ships a CLI and a small Python API.

statement_normalizer/dedup.py

@@ -10,11 +10,12 @@
 
 * If two transactions share a non-empty ``fitid`` (and account), they are the
   same transaction. FITID is bank-assigned and authoritative.
-* Otherwise fall back to a content hash over (date, signed amount, currency,
-  canonical description). To avoid collapsing genuinely-repeated charges (e.g.
-  two identical $4.75 coffees on the same day), identical content rows are only
-  treated as duplicates beyond the count seen on a *single* source statement —
-  i.e. we keep the max multiplicity observed within any one input, not the sum.
+* Otherwise fall back to a content hash over (account, date, signed amount,
+  currency, canonical description). To avoid collapsing genuinely-repeated
+  charges (e.g. two identical $4.75 coffees on the same day), identical content
+  rows are only treated as duplicates beyond the count seen on a *single* source
+  statement — i.e. we keep the max multiplicity observed within any one input,
+  not the sum.
 
 The first occurrence is kept; order is otherwise preserved (stable).
 """

statement_normalizer/schema.py

@@ -120,18 +120,19 @@ def canonical_description(self) -> str:
     def content_hash(self) -> str:
         """Stable hash over the identity-bearing fields.
 
-        Two rows with the same date, signed amount, currency and canonical
-        description are considered the same transaction when no FITID is
-        available. Used by the dedup heuristics.
+        Two rows from the same account with the same date, signed amount,
+        currency and canonical description are considered the same transaction
+        when no FITID is available. Used by the dedup heuristics.
         """
-        key = "|".join(
-            [
-                self.date.isoformat(),
-                f"{self.amount:.2f}",
-                self.currency,
-                self.canonical_description,
-            ]
-        )
+        fields = [
+            self.date.isoformat(),
+            f"{self.amount:.2f}",
+            self.currency,
+            self.canonical_description,
+        ]
+        if self.account_id:
+            fields.insert(0, self.account_id)
+        key = "|".join(fields)
         return hashlib.sha256(key.encode("utf-8")).hexdigest()
 
     def to_dict(self) -> dict[str, Any]:

tests/test_dedup.py

@@ -25,12 +25,21 @@ def test_fitid_dedup():
 
 
 def test_content_dedup_collapses_without_multiplicity():
-    a = _txn(5, "-4.75", "COFFEE ROASTERS")
-    b = _txn(5, "-4.75", "Coffee   Roasters")  # same after canonicalization
+    a = _txn(5, "-4.75", "COFFEE ROASTERS", acct="CHECKING")
+    b = _txn(5, "-4.75", "Coffee   Roasters", acct="CHECKING")
     out = dedup_transactions([a, b])
     assert len(out) == 1
 
 
+def test_content_dedup_preserves_transactions_from_different_accounts():
+    a = _txn(5, "-4.75", "COFFEE ROASTERS", acct="CHECKING")
+    b = _txn(5, "-4.75", "Coffee   Roasters", acct="SAVINGS")
+
+    out = dedup_transactions([a, b])
+
+    assert out == [a, b]
+
+
 def test_content_dedup_preserves_legit_repeats_within_source():
     # Two identical coffees on the same day within ONE statement are legitimate.
     a = _txn(5, "-4.75", "COFFEE")