diff --git a/tests/decommutator/test_ccsds_reader_pipeline.py b/tests/decommutator/test_ccsds_reader_pipeline.py
new file mode 100644
index 0000000..65657aa
--- /dev/null
+++ b/tests/decommutator/test_ccsds_reader_pipeline.py
@@ -0,0 +1,146 @@
+"""
+Tests of the private helper functions in ccsds_reader_pipeline.
+
+The end-to-end test in ``test_decommutator.py`` checks that a whole L0
+file is converted to the expected L05 CDF files, but when it fails it
+does not show which step went wrong. The tests here check individual
+functions so that a failure points to the function responsible.
+
+These tests were generated by Claude.  These tests should be treated
+as regression tests to catch unexpected changes in behavior, rather
+than checks to make sure that functions are behaving correctly.
+"""
+
+from __future__ import annotations
+
+import struct
+
+import pytest
+
+from pyfaradaycup.decommutator.ccsds_reader_pipeline import _parse_ccsds_head
+
+_FIELD_WIDTHS = {
+    "CCSDS_Version": 3,
+    "CCSDS_PacketType": 1,
+    "CCSDS_SecHdrFlag": 1,
+    "CCSDS_ApID": 11,
+    "CCSDS_GroupFlags": 2,
+    "CCSDS_SeqCnt": 14,
+    "CCSDS_PacketLen": 16,
+    "CCSDS_MET": 32,
+}
+
+
+def _make_ccsds_head(fields: dict[str, int]) -> bytes:
+    """
+    Encode CCSDS header fields into a 10-byte header.
+
+    This is the inverse of `_parse_ccsds_head`. It is written with
+    `struct.pack` and bit shifts rather than by reusing any code from
+    the package, so that a mistake in the decoder is not repeated in
+    the encoder and hidden from the tests.
+
+    Parameters
+    ----------
+    fields : dict of str to int
+        The header fields, with the same keys that `_parse_ccsds_head`
+        returns.
+
+    Returns
+    -------
+    bytes
+        The 6-byte CCSDS primary header followed by the 4-byte MET.
+    """
+    word0 = (
+        fields["CCSDS_Version"] << 13
+        | fields["CCSDS_PacketType"] << 12
+        | fields["CCSDS_SecHdrFlag"] << 11
+        | fields["CCSDS_ApID"]
+    )
+    word1 = fields["CCSDS_GroupFlags"] << 14 | fields["CCSDS_SeqCnt"]
+    return struct.pack(
+        ">HHHI", word0, word1, fields["CCSDS_PacketLen"], fields["CCSDS_MET"]
+    )
+
+
+_typical_fields = {
+    "CCSDS_Version": 0,
+    "CCSDS_PacketType": 0,
+    "CCSDS_SecHdrFlag": 1,
+    "CCSDS_ApID": 0x352,
+    "CCSDS_GroupFlags": 3,
+    "CCSDS_SeqCnt": 0x1234,
+    "CCSDS_PacketLen": 0x098D,
+    "CCSDS_MET": 523462910,
+}
+
+
+@pytest.mark.parametrize(
+    "fields",
+    [
+        _typical_fields,
+        {**_typical_fields, "CCSDS_ApID": 0x081},
+        {**_typical_fields, "CCSDS_ApID": 0x35E},
+        {**_typical_fields, "CCSDS_ApID": 0x7FF},
+        dict.fromkeys(_FIELD_WIDTHS, 0),
+        {name: 2**width - 1 for name, width in _FIELD_WIDTHS.items()},
+    ],
+)
+def test_parse_ccsds_head_round_trip(fields: dict[str, int]) -> None:
+    """
+    Test that decoding an encoded header recovers the original fields.
+
+    Every packet that the decommutator reads is identified and sized
+    from its header, so a wrong field here means packets are skipped,
+    assigned to the wrong APID, or given wrong times. The cases include
+    the SPC and spacecraft APIDs that the decommutator looks for, the
+    largest 11-bit APID (to check that the top 3 APID bits, which share
+    a byte with the version and flags, are decoded), and headers with
+    every field at its minimum or maximum value.
+    """
+    assert _parse_ccsds_head(_make_ccsds_head(fields)) == fields
+
+
+@pytest.mark.parametrize("name", _FIELD_WIDTHS)
+def test_parse_ccsds_head_field_isolation(name: str) -> None:
+    """
+    Test that setting every bit of one field does not leak into the others.
+
+    Several fields share a byte, so a mask or shift that is off by one
+    bit would move a bit from one field into its neighbor. Round-trip
+    tests with typical values can miss this, because the bits next to a
+    field boundary are often zero. Setting one field to its maximum and
+    the rest to zero makes any such leak show up as a nonzero value in
+    a field that should be zero, or a smaller value in the field
+    being tested.
+    """
+    fields = dict.fromkeys(_FIELD_WIDTHS, 0)
+    fields[name] = 2 ** _FIELD_WIDTHS[name] - 1
+    assert _parse_ccsds_head(_make_ccsds_head(fields)) == fields
+
+
+def test_parse_ccsds_head_ignores_extra_bytes() -> None:
+    """
+    Test that only the first 10 bytes are decoded.
+
+    Callers do not always pass exactly 10 bytes: ``read_file_sc`` passes
+    the contents of the whole file to read the first header. The bytes
+    after the header must not change the result.
+    """
+    header = _make_ccsds_head(_typical_fields)
+    assert _parse_ccsds_head(header + b"\xff" * 20) == _typical_fields
+
+
+@pytest.mark.parametrize("length", range(10))
+def test_parse_ccsds_head_too_short(length: int) -> None:
+    """
+    Test that a header shorter than 10 bytes raises a ValueError.
+
+    A packet can be cut off at the end of an L0 file. ``_read_bytestr``
+    and ``read_file_sc`` catch `ValueError` to skip such a packet, so a
+    short header must raise `ValueError` rather than another exception
+    or a header with missing fields.
+    """
+    header = _make_ccsds_head(_typical_fields)[:length]
+    with pytest.raises(ValueError, match="not as long as expected"):
+        _parse_ccsds_head(header)