diff --git a/unittests/test_stability.py b/unittests/test_stability.py
new file mode 100644
index 0000000..952f9d3
--- /dev/null
+++ b/unittests/test_stability.py
@@ -0,0 +1,64 @@
+"""
+tests that running migmose over multiple files does not change the results
+"""
+
+import csv
+import random
+import uuid
+from pathlib import Path
+from typing import TypeAlias
+from uuid import UUID
+
+from efoli import EdifactFormat
+
+from migmose.mig.nachrichtenstrukturtabelle import NachrichtenstrukturTabelle
+from migmose.mig.nestednachrichtenstruktur import NestedNachrichtenstruktur
+from migmose.mig.reducednestednachrichtenstruktur import ReducedNestedNachrichtenstruktur
+from migmose.parsing import parse_raw_nachrichtenstrukturzeile
+from unittests import path_to_test_FV2310
+
+_CsvFileContent: TypeAlias = list[list[str]]
+
+
+def _read_csv_file(file_path: Path) -> _CsvFileContent:
+    with open(file_path, encoding="utf-8", mode="r", newline="") as file:
+        reader = csv.reader(file)
+        return list(reader)
+
+
+def test_stability(tmp_path) -> None:
+    input_files: list[tuple[UUID, Path, EdifactFormat]] = [
+        (
+            uuid.uuid4(),
+            path_to_test_FV2310 / "ORDCHGMIG-informatorischeLesefassung1.1_99991231_20231001.docx",
+            EdifactFormat.ORDCHG,
+        ),
+        (
+            uuid.uuid4(),
+            path_to_test_FV2310
+            / "IFTSTAMIG-informatorischeLesefassung2.0e-AußerordentlicheVeröffentlichung_99991231_20231001.docx",
+            EdifactFormat.IFTSTA,
+        ),
+        (
+            uuid.uuid4(),
+            path_to_test_FV2310
+            # pylint:disable=line-too-long
+            / "UTILMDMIGStrom-informatorischeLesefassungS1.1KonsolidierteLesefassungmitFehlerkorrekturenStand12.12.2023_20240402_20231212.docx",
+            EdifactFormat.UTILMDS,
+        ),
+    ]
+    results: dict[UUID, list[_CsvFileContent]] = {}
+    for i in range(10):
+        for file_id, input_file, message_format in random.sample(input_files, len(input_files)):
+            output_dir = tmp_path / f"output-{message_format}-{i}"
+            raw_lines = parse_raw_nachrichtenstrukturzeile(input_file)
+            nachrichtenstrukturtabelle = NachrichtenstrukturTabelle.create_nachrichtenstruktur_tabelle(raw_lines)
+            nested_nachrichtenstruktur, _ = NestedNachrichtenstruktur.create_nested_nachrichtenstruktur(
+                nachrichtenstrukturtabelle
+            )
+            _ = ReducedNestedNachrichtenstruktur.create_reduced_nested_nachrichtenstruktur(nested_nachrichtenstruktur)
+            # just repeating these steps with the nachrichtenstruktur, because they're also done in __main__.py
+            nachrichtenstrukturtabelle.to_csv(message_format, output_dir)
+            results[file_id] = results.get(file_id, []) + [_read_csv_file(output_dir / "nachrichtenstruktur.csv")]
+    for file_id, csv_files in results.items():
+        assert all(csv_file == csv_files[0] for csv_file in csv_files)