diff --git a/projects/python-multipart/fuzz_form.patch b/projects/python-multipart/fuzz_form.patch
index 6d0b6421a81a..fc1fd3b43c5e 100644
--- a/projects/python-multipart/fuzz_form.patch
+++ b/projects/python-multipart/fuzz_form.patch
@@ -1,13 +1,106 @@
 diff --git a/fuzz/fuzz_form.py b/fuzz/fuzz_form.py
-index 9a3d854..fbc6ad9 100644
+index 9a3d854..2d519ce 100644
 --- a/fuzz/fuzz_form.py
 +++ b/fuzz/fuzz_form.py
-@@ -29,7 +29,7 @@ def parse_form_urlencoded(fdp: EnhancedDataProvider) -> None:
-
-
- def parse_multipart_form_data(fdp: EnhancedDataProvider) -> None:
+@@ -1,6 +1,5 @@
+ import io
+ import sys
+-from unittest.mock import Mock
+ 
+ import atheris
+ from helpers import EnhancedDataProvider
+@@ -9,40 +8,78 @@ with atheris.instrument_imports():
+     from python_multipart.exceptions import FormParserError
+     from python_multipart.multipart import parse_form
+ 
+-on_field = Mock()
+-on_file = Mock()
++
++# Simple no-op callbacks, stateless and safe to share across iterations.
++def _on_field(field) -> None:
++    pass
++
++
++def _on_file(file) -> None:
++    file.close()
+ 
+ 
+ def parse_octet_stream(fdp: EnhancedDataProvider) -> None:
+     header = {"Content-Type": "application/octet-stream"}
+-    parse_form(header, io.BytesIO(fdp.ConsumeRandomBytes()), on_field, on_file)
++    parse_form(header, io.BytesIO(fdp.ConsumeRandomBytes()), _on_field, _on_file)
+ 
+ 
+ def parse_url_encoded(fdp: EnhancedDataProvider) -> None:
+     header = {"Content-Type": "application/x-url-encoded"}
+-    parse_form(header, io.BytesIO(fdp.ConsumeRandomBytes()), on_field, on_file)
++    parse_form(header, io.BytesIO(fdp.ConsumeRandomBytes()), _on_field, _on_file)
+ 
+ 
+ def parse_form_urlencoded(fdp: EnhancedDataProvider) -> None:
+     header = {"Content-Type": "application/x-www-form-urlencoded"}
+-    parse_form(header, io.BytesIO(fdp.ConsumeRandomBytes()), on_field, on_file)
++    parse_form(header, io.BytesIO(fdp.ConsumeRandomBytes()), _on_field, _on_file)
++
++
++def parse_multipart_raw(fdp: EnhancedDataProvider) -> None:
++    """Fuzz both the boundary value and the raw body to stress-test boundary matching."""
++    # Boundary: 1-70 bytes, no CR/LF (RFC 2046 constraint kept to avoid ValueError).
++    boundary_len = fdp.ConsumeIntInRange(1, max(1, min(70, fdp.remaining_bytes() // 2)))
++    boundary = fdp.ConsumeBytes(boundary_len)
++    boundary = boundary.replace(b"\r", b"-").replace(b"\n", b"-").rstrip(b" \t") or b"B"
++    header = {"Content-Type": "multipart/form-data; boundary=" + boundary.decode("latin-1")}
++    body = fdp.ConsumeRandomBytes()
++    parse_form(header, io.BytesIO(body), _on_field, _on_file)
+ 
+ 
+-def parse_multipart_form_data(fdp: EnhancedDataProvider) -> None:
 -    boundary = "boundary"
-+    boundary = fdp.ConsumeRandomStringOfSize(16) or "boundary"
-     header = {"Content-Type": f"multipart/form-data; boundary={boundary}"}
-     body = (
-         f"--{boundary}\r\n"
+-    header = {"Content-Type": f"multipart/form-data; boundary={boundary}"}
+-    body = (
+-        f"--{boundary}\r\n"
+-        f"Content-Type: multipart/form-data; boundary={boundary}\r\n\r\n"
+-        f"{fdp.ConsumeRandomString()}\r\n"
+-        f"--{boundary}--\r\n"
+-    )
+-    parse_form(header, io.BytesIO(body.encode("latin1", errors="ignore")), on_field, on_file)
++def parse_multipart_with_content_length(fdp: EnhancedDataProvider) -> None:
++    """Fuzz Content-Length handling together with multipart parsing."""
++    boundary = b"boundary"
++    content_length = fdp.ConsumeIntInRange(0, 1024)
++    header = {
++        "Content-Type": "multipart/form-data; boundary=boundary",
++        "Content-Length": str(content_length),
++    }
++    body = fdp.ConsumeRandomBytes()
++    parse_form(header, io.BytesIO(body), _on_field, _on_file)
++
++
++def parse_form_urlencoded_chunked(fdp: EnhancedDataProvider) -> None:
++    """Feed URL-encoded body in small chunks to exercise streaming state machine."""
++    from python_multipart.multipart import create_form_parser
++
++    num_chunks = fdp.ConsumeIntInRange(1, 8)
++    header = {"Content-Type": "application/x-www-form-urlencoded"}
++    parser = create_form_parser(header, _on_field, _on_file)
++    body = fdp.ConsumeRandomBytes()
++    chunk_size = max(1, (len(body) + num_chunks - 1) // num_chunks)
++    for i in range(0, len(body), chunk_size):
++        parser.write(body[i : i + chunk_size])
++    parser.finalize()
+ 
+ 
+ def TestOneInput(data: bytes) -> None:
+     fdp = EnhancedDataProvider(data)
+-    targets = [parse_octet_stream, parse_url_encoded, parse_form_urlencoded, parse_multipart_form_data]
++    targets = [
++        parse_octet_stream,
++        parse_url_encoded,
++        parse_form_urlencoded,
++        parse_multipart_raw,
++        parse_multipart_with_content_length,
++        parse_form_urlencoded_chunked,
++    ]
+     target = fdp.PickValueInList(targets)
+ 
+     try: