fix(core): register Excel misto numerico+stringa come VARCHAR

tests/test_clean_duckdb_read.py

@@ -392,6 +392,65 @@ def test_read_excel_compact_columns_rename_ods(tmp_path: Path):
         assert rows == [(2022, "ROMA", "1.600"), (2022, "TORINO", "987")]
 
 
+@pytest.mark.policy
+def test_read_excel_mixed_numeric_string_column_registers_as_varchar(tmp_path: Path):
+    """policy: colonne object miste (numeri + stringhe) si registrano come VARCHAR.
+
+    Regression: l'export xlsx Terna Download Center termina con una riga footer
+    "Applied filters: ..." nella prima colonna numerica. DuckDB deduceva INT32
+    dai valori dominanti e falliva sul cast della stringa al register — anche
+    quando il clean.sql filtrava gia' la riga con TRY_CAST.
+    """
+    input_file = tmp_path / "footer.xlsx"
+    pd.DataFrame(
+        [
+            {"Anno": 2022, "Regione": "PIEMONTE", "Valore": 1.5},
+            {"Anno": 2022, "Regione": "LAZIO", "Valore": 2.5},
+            {"Anno": "Applied filters:  Year is 2022;", "Regione": None, "Valore": None},
+        ]
+    ).to_excel(input_file, index=False, engine="openpyxl")
+
+    with safe_connect() as con:
+        logger = logging.getLogger("tests.clean.duckdb_read.xlsx_mixed_footer")
+
+        info = duckdb_read.read_raw_to_relation(
+            con, [input_file], {"header": True}, "fallback", logger
+        )
+
+        assert info.source == "excel"
+        rows = con.execute(
+            'SELECT "Anno", "Regione", "Valore" FROM raw_input ORDER BY "Valore" NULLS LAST'
+        ).fetchall()
+        assert rows == [
+            ("2022", "PIEMONTE", 1.5),
+            ("2022", "LAZIO", 2.5),
+            ("Applied filters:  Year is 2022;", None, None),
+        ]
+        types = {r[0]: r[1] for r in con.execute("DESCRIBE raw_input").fetchall()}
+        assert types["Anno"] == "VARCHAR"
+        assert types["Valore"] == "DOUBLE"
+
+
+@pytest.mark.policy
+def test_normalize_mixed_excel_columns_is_pure(tmp_path: Path):
+    """policy: _normalize_mixed_excel_columns non muta il DataFrame di input."""
+    from toolkit.core.read_excel import _normalize_mixed_excel_columns
+
+    original = pd.DataFrame(
+        [
+            {"Anno": 2022, "Valore": 1.5},
+            {"Anno": "Applied filters", "Valore": None},
+        ]
+    )
+    snapshot = original.copy(deep=True)
+
+    result = _normalize_mixed_excel_columns(original)
+
+    pd.testing.assert_frame_equal(original, snapshot)
+    assert result is not original
+    assert list(result["Anno"]) == ["2022", "Applied filters"]
+
+
 @pytest.mark.policy
 def test_read_excel_source_label_uses_all_batch_suffixes(tmp_path: Path):
     """policy: source_label considera tutti i suffix del batch, non solo il primo."""

toolkit/core/read_excel.py

@@ -34,6 +34,47 @@ def _trim_excel_dataframe(df: pd.DataFrame) -> pd.DataFrame:
     )
 
 
+def _normalize_mixed_excel_columns(df: pd.DataFrame) -> pd.DataFrame:
+    """Return a copy of ``df`` with mixed number+string object columns stringified.
+
+    DuckDB's pandas converter infers a numeric type from the dominant values
+    of an object column and hard-fails on stray strings (e.g. the
+    "Applied filters" footer row emitted by the Terna Download Center xlsx
+    export). Stringifying mixed columns keeps registration total; SQL
+    consumers cast with the toolkit macros (cast_int/cast_double).
+    Pure: the input frame is never mutated (copy-on-write).
+    """
+    import numbers
+
+    import pandas as pd
+
+    result: pd.DataFrame | None = None
+    for col in df.columns:
+        series = df[col]
+        if series.dtype != object:
+            continue
+        has_str = False
+        has_num = False
+        for value in series:
+            if value is None or (isinstance(value, float) and pd.isna(value)):
+                continue
+            if isinstance(value, str):
+                has_str = True
+            elif isinstance(value, numbers.Number):
+                has_num = True
+        if has_str and has_num:
+            if result is None:
+                result = df.copy()
+            result[col] = series.map(
+                lambda value: (
+                    None
+                    if value is None or (isinstance(value, float) and pd.isna(value))
+                    else (value if isinstance(value, str) else str(value))
+                )
+            )
+    return df if result is None else result
+
+
 def _load_excel_frame(
     input_file: Path,
     read_cfg: dict[str, Any],
@@ -122,6 +163,7 @@ def _execute_excel_read(
             params_used = frame_params
 
     combined = pd.concat(frames, ignore_index=True) if len(frames) > 1 else frames[0]
+    combined = _normalize_mixed_excel_columns(combined)
     con.register(RAW_INPUT_DF_VIEW, combined)
     con.execute(f"CREATE OR REPLACE VIEW {RAW_INPUT_VIEW} AS SELECT * FROM {RAW_INPUT_DF_VIEW};")