@@ -1,3 +1,9 @@
+## 0.27.5-dev0
+
+### Fixes
+
+- **Attachment elements keep their own filetype.** When partitioning an email with attachments, `auto.partition()` re-stamped `metadata.filetype` on every element with the containing message's type, so content extracted from an attached PDF was labelled `message/rfc822`. Attachment elements now retain the filetype assigned by the nested `partition()` call that produced them, including when the containing document is a file-like object with no known file-name.
+
 ## 0.27.4
 
 ### Fixes
@@ -8,7 +14,7 @@
 
 ### Fixes
 
-- **Updated README:** Added Sign up Link for Transform MCP in README. 
+- **Updated README:** Added Sign up Link for Transform MCP in README.
 
 ## 0.27.2
 
 
@@ -48,6 +48,7 @@
 from unstructured.file_utils.model import FileType, create_file_type, register_partitioner
 from unstructured.partition.auto import _PartitionerLoader, partition
 from unstructured.partition.common import UnsupportedFileFormatError
+from unstructured.partition.common.metadata import is_attachment_element
 from unstructured.partition.utils.constants import PartitionStrategy
 from unstructured.staging.base import elements_from_json, elements_to_dicts, elements_to_json
 
@@ -1417,6 +1418,46 @@ def test_auto_partition_applies_the_correct_filetype_for_all_filetypes(
     )
 
 
+@pytest.mark.parametrize(
+    "file_name",
+    ["eml/fake-email-attachment.eml", "fake-email-attachment.msg"],
+)
+def test_auto_partition_preserves_the_filetype_of_attachment_elements(file_name: str):
+    """Attachment elements keep their own filetype, not the containing message's.
+
+    Their filetype was assigned by the nested `partition()` call that produced them, so the
+    outer call must not re-stamp them with e.g. `message/rfc822`.
+    """
+    elements = partition(example_doc_path(file_name), process_attachments=True)
+
+    attachment_elements = [e for e in elements if is_attachment_element(e)]
+    assert attachment_elements
+    assert all(e.metadata.filetype == FileType.TXT.mime_type for e in attachment_elements)
+
+
+@pytest.mark.parametrize(
+    "file_name",
+    ["eml/fake-email-attachment.eml", "fake-email-attachment.msg"],
+)
+def test_auto_partition_preserves_attachment_filetype_when_container_filename_is_unknown(
+    file_name: str,
+):
+    """The attachment guard must not depend on `.metadata.attached_to_filename`.
+
+    That field is `None` when the containing document's file-name is unknown -- partitioning a
+    file-like object with no `metadata_filename` -- so keying the guard on it would let the
+    containing document's filetype overwrite the attachment's in exactly that case.
+    """
+    with open(example_doc_path(file_name), "rb") as f:
+        elements = partition(file=f, process_attachments=True)
+
+    attachment_elements = [e for e in elements if is_attachment_element(e)]
+    assert attachment_elements
+    # -- precondition: this is the case the `attached_to_filename` guard cannot see --
+    assert all(e.metadata.attached_to_filename is None for e in attachment_elements)
+    assert all(e.metadata.filetype == FileType.TXT.mime_type for e in attachment_elements)
+
+
 def test_detect_filetype_maps_file_to_bytes_io_when_spooled_temp_file_used(mocker):
     detect_filetype_mock = MagicMock(return_value=FileType.JSON)
     mocker.patch("unstructured.file_utils.filetype._FileTypeDetector", detect_filetype_mock)
 
@@ -1 +1 @@
-__version__ = "0.27.4"  # pragma: no cover
+__version__ = "0.27.5-dev0"  # pragma: no cover
@@ -16,6 +16,7 @@
 from unstructured.partition.common import UnsupportedFileFormatError
 from unstructured.partition.common.common import exactly_one
 from unstructured.partition.common.lang import check_language_args
+from unstructured.partition.common.metadata import is_attachment_element
 from unstructured.partition.utils.constants import PartitionStrategy
 from unstructured.safe_http import safe_get
 from unstructured.telemetry import partition_runtime_telemetry, set_partition_document_type
@@ -201,6 +202,13 @@ def augment_metadata(elements: list[Element]) -> list[Element]:
         for element in elements:
             element.metadata.url = url
             element.metadata.data_source = data_source_metadata
+            # -- an attachment's elements were assigned their own (correct) filetype by the
+            # -- nested `partition()` call that produced them; don't re-stamp them with the
+            # -- containing document's type (e.g. an attached PDF is not `message/rfc822`).
+            # -- Note this cannot key on `.metadata.attached_to_filename`, which is `None`
+            # -- whenever the containing document's file-name is unknown. --
+            if is_attachment_element(element):
+                continue
             if content_type is not None:
                 out_filetype = FileType.from_mime_type(content_type)
                 element.metadata.filetype = out_filetype.mime_type if out_filetype else None
 
@@ -17,6 +17,28 @@
 
 _P = ParamSpec("_P")
 
+# -- Name of a transient, in-process attribute used to identify elements produced by partitioning
+# -- an attachment. It is deliberately NOT an `ElementMetadata` field: it must not appear in
+# -- dict/JSON output, and it must work when the containing document's file-name is unknown --
+# -- which is exactly when `.metadata.attached_to_filename` is `None` and so cannot serve as the
+# -- marker (e.g. `partition(file=f)` with no `metadata_filename`).
+_ATTACHMENT_ELEMENT_ATTR = "_produced_by_attachment_partitioning"
+
+
+def mark_as_attachment_element(element: Element) -> None:
+    """Record that `element` was produced by partitioning an attachment.
+
+    Attachment elements are partitioned by a nested `partition()` call, which assigns them their
+    own (correct) source-document metadata. Marking them lets the containing document's partitioner
+    avoid overwriting that metadata with its own.
+    """
+    setattr(element, _ATTACHMENT_ELEMENT_ATTR, True)
+
+
+def is_attachment_element(element: Element) -> bool:
+    """True when `element` was produced by partitioning an attachment."""
+    return getattr(element, _ATTACHMENT_ELEMENT_ATTR, False) is True
+
 
 def get_last_modified_date(filename: str) -> str | None:
     """Modification time of file at path `filename`, if it exists.
 
@@ -21,7 +21,10 @@
 from unstructured.file_utils.model import FileType
 from unstructured.logger import logger
 from unstructured.partition.common import EXPECTED_ATTACHMENT_ERRORS
-from unstructured.partition.common.metadata import get_last_modified_date
+from unstructured.partition.common.metadata import (
+    get_last_modified_date,
+    mark_as_attachment_element,
+)
 from unstructured.partition.html import partition_html
 from unstructured.partition.text import partition_text
 from unstructured.telemetry import partition_runtime_telemetry
@@ -416,6 +419,7 @@ def _iter_elements(self) -> Iterator[Element]:
 
         for e in elements:
             e.metadata.attached_to_filename = self._attached_to_filename
+            mark_as_attachment_element(e)
             yield e
 
     @cached_property
 
@@ -13,7 +13,10 @@
 from unstructured.file_utils.model import FileType
 from unstructured.logger import logger
 from unstructured.partition.common import EXPECTED_ATTACHMENT_ERRORS
-from unstructured.partition.common.metadata import get_last_modified_date
+from unstructured.partition.common.metadata import (
+    get_last_modified_date,
+    mark_as_attachment_element,
+)
 from unstructured.partition.html import partition_html
 from unstructured.partition.text import partition_text
 from unstructured.telemetry import partition_runtime_telemetry
@@ -282,6 +285,7 @@ def _iter_elements(self) -> Iterator[Element]:
 
             for e in elements:
                 e.metadata.attached_to_filename = self._opts.metadata_file_path
+                mark_as_attachment_element(e)
                 yield e
 
     @cached_property
Original file line number	Diff line number	Diff line change
`@@ -1 +1 @@`
`1`		`-__version__ = "0.27.4" # pragma: no cover`
	`1`	`+__version__ = "0.27.5-dev0" # pragma: no cover`