diff --git a/README.rst b/README.rst
index 059a68f..744abbd 100644
--- a/README.rst
+++ b/README.rst
@@ -181,6 +181,18 @@ nested ``Options`` class in ``MarkdownConverter`` subclasses.
 Converting BeautifulSoup objects
 ================================
 
+``convert_soup`` accepts a BeautifulSoup document or a Tag subtree. For Tag
+roots, ``strip_document`` applies to the completed Markdown after the tag's
+normal conversion. Use ``strip_document=None`` to retain boundary newlines.
+Excluding ``[document]`` through ``strip`` or ``convert``, or resolving its
+conversion function to ``None``, disables this final formatting.
+Tag roots use the built-in ``strip_document`` policy. This final formatting
+step does not invoke the resolved document conversion function; normal tag
+dispatch and the original tree are retained. The document-function lookup is
+cached, including ``None``, so Tag conversion can affect later document
+resolution on the same converter. ``strip_document=None`` retains boundary
+text but still performs this lookup.
+
 .. code:: python
 
     from markdownify import MarkdownConverter
diff --git a/markdownify/__init__.py b/markdownify/__init__.py
index 28cdaf6..242b344 100644
--- a/markdownify/__init__.py
+++ b/markdownify/__init__.py
@@ -79,6 +79,20 @@ def strip_pre(text):
     return text
 
 
+def _strip_document(text, strip_mode):
+    """Remove only the selected document separation newlines."""
+    if strip_mode == LSTRIP:
+        return text.lstrip('\n')
+    elif strip_mode == RSTRIP:
+        return text.rstrip('\n')
+    elif strip_mode == STRIP:
+        return text.strip('\n')
+    elif strip_mode is None:
+        return text
+    else:
+        raise ValueError('Invalid value for strip_document: %s' % strip_mode)
+
+
 def chomp(text):
     """
     If the text in an inline tag like b, a, or em contains a leading or trailing
@@ -223,7 +237,12 @@ def convert(self, html):
         return self.convert_soup(soup)
 
     def convert_soup(self, soup):
-        return self.process_tag(soup, parent_tags=set())
+        text = self.process_tag(soup, parent_tags=set())
+        if (isinstance(soup, Tag)
+                and soup.name.lower() != '[document]'
+                and self.get_conv_fn_cached('[document]') is not None):
+            text = _strip_document(text, self.options['strip_document'])
+        return text
 
     def process_element(self, node, parent_tags=None):
         if isinstance(node, NavigableString):
@@ -327,18 +346,7 @@ def _can_ignore(el):
 
     def convert__document_(self, el, text, parent_tags):
         """Final document-level formatting for BeautifulSoup object (node.name == "[document]")"""
-        if self.options['strip_document'] == LSTRIP:
-            text = text.lstrip('\n')  # remove leading separation newlines
-        elif self.options['strip_document'] == RSTRIP:
-            text = text.rstrip('\n')  # remove trailing separation newlines
-        elif self.options['strip_document'] == STRIP:
-            text = text.strip('\n')  # remove leading and trailing separation newlines
-        elif self.options['strip_document'] is None:
-            pass  # leave leading and trailing separation newlines as-is
-        else:
-            raise ValueError('Invalid value for strip_document: %s' % self.options['strip_document'])
-
-        return text
+        return _strip_document(text, self.options['strip_document'])
 
     def process_text(self, el, parent_tags=None):
         # For the top-level element, initialize the parent context with an empty set.
diff --git a/tests/test_soup.py b/tests/test_soup.py
new file mode 100644
index 0000000..98e3fbc
--- /dev/null
+++ b/tests/test_soup.py
@@ -0,0 +1,242 @@
+import pytest  # type: ignore[import-not-found]
+from bs4 import BeautifulSoup
+from markdownify import MarkdownConverter, LSTRIP, RSTRIP, STRIP, STRIP_ONE
+
+
+@pytest.mark.parametrize('mode,expected', [
+    (STRIP, 'Hello'),
+    (LSTRIP, 'Hello\n\n'),
+    (RSTRIP, '\n\n\nHello'),
+    (None, '\n\n\nHello\n\n'),
+])
+def test_tag_strip_document(mode, expected):
+    soup = BeautifulSoup('<html>\n<body><div><p>Hello</p></div></body></html>', 'html.parser')
+    assert MarkdownConverter(strip_document=mode).convert_soup(soup.html) == expected
+
+
+@pytest.mark.parametrize('tag', ['html', 'body', 'div', 'p'])
+def test_tag_default_strip_document(tag):
+    soup = BeautifulSoup('<html>\n<body><div><p>Hello</p></div></body></html>', 'html.parser')
+    assert MarkdownConverter().convert_soup(soup.find(tag)) == 'Hello'
+
+
+@pytest.mark.parametrize('options,expected', [
+    ({'convert': ['p']}, '\n\nHello\n\n'),
+    ({'strip': ['[document]']}, '\n\nHello\n\n'),
+    ({'convert': []}, 'Hello'),
+])
+@pytest.mark.parametrize('mode', [STRIP, LSTRIP, RSTRIP, None, 'invalid'])
+def test_excluded_document_policy(options, expected, mode):
+    soup = BeautifulSoup('<p>Hello</p>', 'html.parser')
+    converter = MarkdownConverter(strip_document=mode, **options)
+    assert converter.convert_soup(soup.p) == expected
+    assert converter.convert_soup(soup) == expected
+
+
+@pytest.mark.parametrize('options', [
+    {'strip': []}, {'strip': ['div']}, {'convert': ['p', '[document]']},
+])
+def test_document_policy_enabled_independently_of_root_converter(options):
+    soup = BeautifulSoup('<div><p>Hello</p></div>', 'html.parser')
+    assert MarkdownConverter(**options).convert_soup(soup.div) == 'Hello'
+
+
+@pytest.mark.parametrize('mode', ['invalid', STRIP_ONE])
+@pytest.mark.parametrize('html', ['<p></p>', '<p>Hello</p>'])
+def test_invalid_tag_document_mode(html, mode):
+    soup = BeautifulSoup(html, 'html.parser')
+    with pytest.raises(ValueError, match='^Invalid value for strip_document: %s$' % mode):
+        MarkdownConverter(strip_document=mode).convert_soup(soup.p)
+
+
+@pytest.mark.parametrize('mode,expected', [
+    (STRIP, ' \tA\r\n\nB\t '),
+    (LSTRIP, ' \tA\r\n\nB\t \n\n'),
+    (RSTRIP, '\n\n \tA\r\n\nB\t '),
+    (None, '\n\n \tA\r\n\nB\t \n\n'),
+])
+def test_custom_root_preserves_significant_whitespace(mode, expected):
+    soup = BeautifulSoup('<root><b>child</b></root>', 'html.parser')
+    calls = []
+
+    class Converter(MarkdownConverter):
+        def convert_root(self, el, text, parent_tags):
+            assert el is soup.root
+            assert text == '**child**'
+            assert parent_tags == set()
+            calls.append(el)
+            return '\n\n \tA\r\n\nB\t \n\n'
+
+    assert Converter(strip_document=mode).convert_soup(soup.root) == expected
+    assert len(calls) == 1
+
+
+@pytest.mark.parametrize('mode', [STRIP, None, 'invalid'])
+def test_replacing_document_callback_remains_document_only(mode):
+    soup = BeautifulSoup('<p>Hello</p>', 'html.parser')
+    calls = []
+
+    class Converter(MarkdownConverter):
+        def convert__document_(self, el, text, parent_tags):
+            assert el is soup
+            assert text == '\n\nHello\n\n'
+            assert parent_tags == set()
+            calls.append(el)
+            return '\n\nDOCUMENT\n\n'
+
+    converter = Converter(strip_document=mode)
+    assert converter.convert_soup(soup) == '\n\nDOCUMENT\n\n'
+    if mode == 'invalid':
+        with pytest.raises(ValueError):
+            converter.convert_soup(soup.p)
+    else:
+        expected = 'Hello' if mode == STRIP else '\n\nHello\n\n'
+        assert converter.convert_soup(soup.p) == expected
+    assert len(calls) == 1
+
+
+def test_process_tag_override_preserves_document_output():
+    soup = BeautifulSoup('<p>Hello</p>', 'html.parser')
+    calls = []
+
+    class Converter(MarkdownConverter):
+        def process_tag(self, node, parent_tags=None):
+            assert parent_tags == set()
+            calls.append(node)
+            return '\n\nCUSTOM\n\n'
+
+    converter = Converter()
+    assert converter.convert_soup(soup) == '\n\nCUSTOM\n\n'
+    assert converter.convert_soup(soup.p) == 'CUSTOM'
+    assert len(calls) == 2
+    assert calls[0] is soup and calls[1] is soup.p
+    soup.name = '[DOCUMENT]'
+    assert converter.convert_soup(soup) == '\n\nCUSTOM\n\n'
+
+
+@pytest.mark.parametrize('available', [True, False])
+def test_document_lookup_cache_and_original_tag_dispatch(available):
+    soup = BeautifulSoup('<p>Hello</p>', 'html.parser')
+    resolved = []
+    converted = []
+    processed = []
+
+    class Converter(MarkdownConverter):
+        def get_conv_fn(self, name):
+            resolved.append(name)
+            if name == '[document]' and not available:
+                return None
+            return super().get_conv_fn(name)
+
+        def convert_p(self, el, text, parent_tags):
+            converted.append((el, set(parent_tags)))
+            return super().convert_p(el, text, parent_tags)
+
+        def process_element(self, node, parent_tags=None):
+            processed.append(node)
+            return super().process_element(node, parent_tags)
+
+    converter = Converter()
+    expected = 'Hello' if available else '\n\nHello\n\n'
+    for root in (soup.p, soup.p, soup, soup.p):
+        assert converter.convert_soup(root) == expected
+    assert resolved.count('[document]') == 1
+    assert resolved.count('p') == 1
+    assert len(converted) == 4
+    assert all(el is soup.p for el, _ in converted)
+    assert [parents for _, parents in converted] == [set(), set(), {'[document]'}, set()]
+    assert len(processed) == 5
+    assert sum(node is soup.p for node in processed) == 1
+
+
+@pytest.mark.parametrize('mode', [STRIP, LSTRIP, RSTRIP, None, 'invalid'])
+def test_attached_tree_relationships_survive_conversion(mode):
+    soup = BeautifulSoup('<section><i>before</i><p>Hello</p><b>after</b></section>', 'html.parser')
+    serialised = str(soup)
+    nodes = [soup] + list(soup.descendants)
+    links = [(node.parent, node.previous_sibling, node.next_sibling,
+              node.previous_element, node.next_element) for node in nodes]
+    converter = MarkdownConverter(strip_document=mode)
+    for _ in range(2):
+        if mode == 'invalid':
+            with pytest.raises(ValueError):
+                converter.convert_soup(soup.p)
+        else:
+            converter.convert_soup(soup.p)
+        current_nodes = [soup] + list(soup.descendants)
+        assert len(current_nodes) == len(nodes)
+        assert all(current is original for current, original in zip(current_nodes, nodes))
+        for node, original in zip(nodes, links):
+            current = (node.parent, node.previous_sibling, node.next_sibling,
+                       node.previous_element, node.next_element)
+            assert all(a is b for a, b in zip(current, original))
+        assert str(soup) == serialised
+
+
+@pytest.mark.parametrize('mode,prefix,suffix', [
+    (STRIP, '', ''), (LSTRIP, '', '\n\n'),
+    (RSTRIP, '\n\n', ''), (None, '\n\n', '\n\n'),
+])
+@pytest.mark.parametrize('pre_mode,code', [
+    (STRIP, '```\n  Hello\n```'),
+    (STRIP_ONE, '```\n  \n  Hello  \n  \n```'),
+    (None, '```\n  \n  \n  Hello  \n  \n  \n```'),
+])
+@pytest.mark.parametrize('tag', ['section', 'pre'])
+def test_pre_root_and_descendant_keep_code_whitespace(mode, prefix, suffix, pre_mode, code, tag):
+    soup = BeautifulSoup('<section><pre>  \n  \n  Hello  \n  \n  </pre></section>', 'html.parser')
+    callbacks = []
+
+    def language(el):
+        assert el is soup.pre
+        callbacks.append(el)
+
+    converter = MarkdownConverter(strip_document=mode, strip_pre=pre_mode, code_language_callback=language)
+    assert converter.convert_soup(soup.find(tag)) == prefix + code + suffix
+    assert len(callbacks) == 1
+
+
+@pytest.mark.parametrize('mode,expected', [(STRIP, 'ALIAS'), (None, '\n\nALIAS\n\n')])
+def test_document_method_alias_keeps_original_root_dispatch(mode, expected):
+    soup = BeautifulSoup('<p>Hello</p>', 'html.parser')
+    root = soup.p
+    root.name = '_document_'
+    calls = []
+
+    class Converter(MarkdownConverter):
+        def convert__document_(self, el, text, parent_tags):
+            assert el is root
+            assert text == 'Hello'
+            assert parent_tags == set()
+            calls.append(el)
+            return '\n\nALIAS\n\n'
+
+    assert Converter(strip_document=mode).convert_soup(root) == expected
+    assert len(calls) == 1
+
+
+@pytest.mark.parametrize('mode', [STRIP, None])
+def test_tag_lookup_caches_document_unavailability_before_state_changes(mode):
+    soup = BeautifulSoup('<p>Hello</p>', 'html.parser')
+    resolved = []
+    document_calls = []
+
+    class Converter(MarkdownConverter):
+        available = False
+
+        def get_conv_fn(self, name):
+            resolved.append((name, self.available))
+            if name == '[document]' and not self.available:
+                return None
+            return super().get_conv_fn(name)
+
+        def convert__document_(self, el, text, parent_tags):
+            document_calls.append(el)
+            return super().convert__document_(el, text, parent_tags)
+
+    converter = Converter(strip_document=mode)
+    assert converter.convert_soup(soup.p) == '\n\nHello\n\n'
+    converter.available = True
+    assert converter.convert_soup(soup) == '\n\nHello\n\n'
+    assert resolved == [('p', False), ('[document]', False)]
+    assert document_calls == []