diff --git a/src/bls/oews.py b/src/bls/oews.py
index 5a560d3..1c2e87a 100644
--- a/src/bls/oews.py
+++ b/src/bls/oews.py
@@ -129,7 +129,8 @@ def read_national(zip_path: Path, year: int) -> list[dict]:
         if ext == "xlsx"
         else pl.read_excel(io.BytesIO(raw), infer_schema_length=0, engine="calamine")
     )
-    return df.to_dicts()
+    # The 2019 release shipped lower-case headers; every other year is upper.
+    return df.rename({c: c.upper() for c in df.columns}).to_dicts()
 
 
 # ── rows ──────────────────────────────────────────────────────────────
diff --git a/tests/bls/test_oews.py b/tests/bls/test_oews.py
index de40d2f..8f9dd2e 100644
--- a/tests/bls/test_oews.py
+++ b/tests/bls/test_oews.py
@@ -271,6 +271,14 @@ class TestReadNational:
         recs = read_national(p, 2023)
         assert len(recs) == 2 and recs[0]["O_GROUP"] == "detailed"
 
+    def test_lower_case_headers_are_normalised(self, tmp_path):
+        # the May 2019 file is the one release with lower-case column names
+        p = tmp_path / "oesm19nat.zip"
+        p.write_bytes(_zip(2019, [c.lower() for c in NEW_COLS], NEW_ROWS))
+        recs = read_national(p, 2019)
+        assert recs[0]["OCC_CODE"] == "31-9092"
+        assert len(to_rows(2019, recs, item_key="K")) == 2
+
     def test_missing_member(self, tmp_path):
         p = tmp_path / "bad.zip"
         with zipfile.ZipFile(p, "w") as z: