fix kfintech conflict resolver dropping group columns

src/openreversefeed/core/conflict.py

@@ -5,24 +5,36 @@
 
 
 def resolve_kfintech_conflicts(df: pd.DataFrame) -> pd.DataFrame:
+    """If a `(transaction_number, folio_number)` group contains both a plain
+    ``P`` (purchase) and any ``*SIN`` marker row, keep only the first ``P``
+    and drop the SIN rows. Groups without that collision are untouched.
+
+    Implemented as a manual row-index scan rather than ``groupby.apply`` so
+    that the returned DataFrame keeps every column of the input — using
+    ``groupby.apply`` with ``include_groups=False`` strips the group columns
+    from the output, and ``include_groups=True`` is deprecated in pandas
+    2.2.
+    """
     if df.empty:
         return df.copy()
 
     required = {"transaction_number", "folio_number", "transaction_purred"}
     if not required.issubset(df.columns):
         return df.copy()
 
-    def _pick(group: pd.DataFrame) -> pd.DataFrame:
+    kept_indices: list = []
+    for _key, group in df.groupby(
+        ["transaction_number", "folio_number"], sort=False
+    ):
         purreds = set(group["transaction_purred"])
         has_sin = any(
-            isinstance(p, str) and (p == "SIN" or p.endswith("SIN")) for p in purreds
+            isinstance(p, str) and (p == "SIN" or p.endswith("SIN"))
+            for p in purreds
         )
         if "P" in purreds and has_sin:
-            return group[group["transaction_purred"] == "P"].head(1)
-        return group
+            first_p = group[group["transaction_purred"] == "P"].head(1)
+            kept_indices.extend(first_p.index.tolist())
+        else:
+            kept_indices.extend(group.index.tolist())
 
-    return (
-        df.groupby(["transaction_number", "folio_number"], sort=False, group_keys=False)
-        .apply(_pick, include_groups=False)
-        .reset_index(drop=True)
-    )
+    return df.loc[kept_indices].reset_index(drop=True)

tests/unit/test_cleaner.py

@@ -3,6 +3,7 @@
 import pandas as pd
 
 from openreversefeed.adapters.cams import CamsAdapter
+from openreversefeed.adapters.kfintech import KFintechFormat1Adapter
 from openreversefeed.core.cleaner import Cleaner
 
 
@@ -32,3 +33,57 @@ def test_cleaner_runs_all_steps_on_minimal_cams_df():
     assert "composite_key" in out.columns
     assert "action" in out.columns
     assert "action_tag" in out.columns
+
+
+def test_cleaner_kfintech_preserves_group_columns_through_conflict_resolution():
+    """Regression: the conflict resolver must not strip transaction_number or
+    folio_number from the cleaned DataFrame, otherwise the composite-key
+    builder fails with KeyError downstream."""
+    cleaner = Cleaner()
+    adapter = KFintechFormat1Adapter()
+    df = pd.DataFrame(
+        [
+            {
+                "transaction_id": "KF001",
+                "transaction_number": "T1",
+                "parent_transaction_number": "0",
+                "folio_number": "F1",
+                "product_code": "P1",
+                "scheme_code": "S1",
+                "units": 100.0,
+                "amount": 10000.0,
+                "transaction_date": date(2025, 1, 1),
+                "transaction_mode": "N",
+                "transaction_purred": "P",
+                "transaction_flag": "",
+                "transaction_type": "",
+                "registrar_row_index": 0,
+                "__source_meta": [{}],
+            },
+            {
+                "transaction_id": "KF002",
+                "transaction_number": "T2",
+                "parent_transaction_number": "0",
+                "folio_number": "F1",
+                "product_code": "P1",
+                "scheme_code": "S1",
+                "units": 50.0,
+                "amount": 5000.0,
+                "transaction_date": date(2025, 1, 2),
+                "transaction_mode": "N",
+                "transaction_purred": "R",
+                "transaction_flag": "TO",
+                "transaction_type": "",
+                "registrar_row_index": 1,
+                "__source_meta": [{}],
+            },
+        ]
+    )
+    out = cleaner.run(df, adapter)
+    assert "transaction_number" in out.columns, (
+        "transaction_number must survive the cleaner pipeline — "
+        "otherwise build_kfintech_key raises KeyError"
+    )
+    assert "folio_number" in out.columns
+    assert "composite_key" in out.columns
+    assert not out["composite_key"].isna().any()