From e73435f7ffc27966fddb04a0ec48cbbdb156d3aa Mon Sep 17 00:00:00 2001
From: HarshitR2004 <harshithranjan6971@gmail.com>
Date: Wed, 30 Sep 2026 23:15:04 +0530
Subject: [PATCH] fix(converters): preserve literal NA strings in
 XLSXToDocument (#12945)

---
 haystack/components/converters/xlsx.py        |  4 ++
 ...ps-literal-na-values-b18412ebd88214fe.yaml |  7 +++
 .../converters/test_xlsx_to_document.py       | 57 +++++++++++++++++++
 3 files changed, 68 insertions(+)
 create mode 100644 releasenotes/notes/fix-xlsx-drops-literal-na-values-b18412ebd88214fe.yaml

diff --git a/haystack/components/converters/xlsx.py b/haystack/components/converters/xlsx.py
index 86aeaa0d76..3589e12040 100644
--- a/haystack/components/converters/xlsx.py
+++ b/haystack/components/converters/xlsx.py
@@ -63,6 +63,8 @@ def __init__(
         :param table_format: The format to convert the Excel file to.
         :param sheet_name: The name of the sheet to read. If None, all sheets are read.
         :param read_excel_kwargs: Additional arguments to pass to `pandas.read_excel`.
+            By default, `keep_default_na` is set to False and `na_values` to `[""]` so literal text
+            values like "NA" and "N/A" are preserved while empty cells remain empty.
             See https://pandas.pydata.org/docs/reference/api/pandas.read_excel.html#pandas-read-excel
         :param table_format_kwargs: Additional keyword arguments to pass to the table format function.
             - If `table_format` is "csv", these arguments are passed to `pandas.DataFrame.to_csv`.
@@ -164,6 +166,8 @@ def _extract_tables(self, bytestream: ByteStream) -> tuple[list[str], list[dict]
         """
         file_bytes = io.BytesIO(bytestream.data)
         resolved_read_excel_kwargs = {
+            "keep_default_na": False,
+            "na_values": [""],
             **self.read_excel_kwargs,
             "sheet_name": self.sheet_name,
             "header": None,  # Don't assign any pandas column labels
diff --git a/releasenotes/notes/fix-xlsx-drops-literal-na-values-b18412ebd88214fe.yaml b/releasenotes/notes/fix-xlsx-drops-literal-na-values-b18412ebd88214fe.yaml
new file mode 100644
index 0000000000..408e4c3d34
--- /dev/null
+++ b/releasenotes/notes/fix-xlsx-drops-literal-na-values-b18412ebd88214fe.yaml
@@ -0,0 +1,7 @@
+---
+fixes:
+  - |
+    Fixed an issue where ``XLSXToDocument`` dropped literal text values such as ``"NA"`` and ``"N/A"``,
+    converting them into empty cells. ``keep_default_na=False`` and ``na_values=[""]`` are now passed to
+    ``pandas.read_excel`` by default while preserving truly empty cells and allowing callers to override them
+    via ``read_excel_kwargs``.
diff --git a/test/components/converters/test_xlsx_to_document.py b/test/components/converters/test_xlsx_to_document.py
index 458d564568..19c9a4d866 100644
--- a/test/components/converters/test_xlsx_to_document.py
+++ b/test/components/converters/test_xlsx_to_document.py
@@ -254,3 +254,60 @@ def test_link_extraction_from_typed_columns(
         ]
         assert rows == [linked_values + ["1.23"], ["43", "4.5", "False", "2.35"]]
         assert documents[0].meta["xlsx"] == {"sheet_name": "Sheet"}
+
+    @pytest.mark.parametrize("table_format", ["csv", "markdown"])
+    def test_run_preserves_literal_na_strings(self, tmp_path: Path, table_format: Literal["csv", "markdown"]) -> None:
+        workbook = Workbook()
+        sheet = workbook.active
+        assert sheet is not None
+        sheet.append(["Code", "Status", "Blank"])
+        sheet.append(["NA", "OK", None])
+        sheet.append(["N/A", "pending", ""])
+        sheet.append(["North America", "active", ""])
+        path = tmp_path / "literal_na.xlsx"
+        workbook.save(path)
+        workbook.close()
+
+        converter = XLSXToDocument(table_format=table_format)
+        documents = converter.run(sources=[path])["documents"]
+
+        assert len(documents) == 1
+        content = documents[0].content
+        assert content is not None
+
+        if table_format == "csv":
+            rows = [row[1:] for row in list(csv.reader(io.StringIO(content)))[1:]]
+            assert rows == [
+                ["Code", "Status", "Blank"],
+                ["NA", "OK", ""],
+                ["N/A", "pending", ""],
+                ["North America", "active", ""],
+            ]
+        else:
+            rows = [[cell.strip() for cell in row.split("|")[2:-1]] for row in content.splitlines()[2:]]
+            assert rows == [
+                ["Code", "Status", "Blank"],
+                ["NA", "OK", ""],
+                ["N/A", "pending", ""],
+                ["North America", "active", ""],
+            ]
+
+    def test_run_keep_default_na_override(self, tmp_path: Path) -> None:
+        workbook = Workbook()
+        sheet = workbook.active
+        assert sheet is not None
+        sheet.append(["Code", "Status"])
+        sheet.append(["NA", "OK"])
+        sheet.append(["N/A", "pending"])
+        path = tmp_path / "override_na.xlsx"
+        workbook.save(path)
+        workbook.close()
+
+        converter = XLSXToDocument(read_excel_kwargs={"keep_default_na": True})
+        documents = converter.run(sources=[path])["documents"]
+
+        assert len(documents) == 1
+        content = documents[0].content
+        assert content is not None
+        rows = [row[1:] for row in list(csv.reader(io.StringIO(content)))[1:]]
+        assert rows == [["Code", "Status"], ["", "OK"], ["", "pending"]]