@@ -29,6 +29,15 @@ jobs:
         with:
           python-version: ${{ matrix.python-version }}
 
+      - name: Install media tools
+        if: runner.os == 'Linux'
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y --no-install-recommends libimage-exiftool-perl ffmpeg
+          exiftool -ver
+          ffmpeg -version
+          ffprobe -version
+
       - name: Install Hatch
         run: pipx install hatch
       - name: Run tests
 
@@ -14,7 +14,6 @@
 import sys
 from pathlib import Path
 from typing import Any
-from unittest.mock import MagicMock, patch
 
 import pytest
 
@@ -207,28 +206,30 @@ def test_pdf_scanned_report(svc: MockOCRService) -> None:
 # ---------------------------------------------------------------------------
 
 
-def test_pdf_scanned_fallback_format(svc: MockOCRService) -> None:
-    """_ocr_full_pages emits *[Image OCR]...[End OCR]* for each page."""
+def test_pdf_scanned_fallback_format() -> None:
+    """Render a real PDF, replacing only the remote OCR service."""
+    from PIL import Image
+
     path = TEST_DATA_DIR / "pdf_image_start.pdf"
-    if not path.exists():
-        pytest.skip(f"Test file not found: {path}")
+    seen = []
+
+    class RecordingOCRService:
+        def extract_text(self, image_stream: Any) -> OCRResult:
+            assert image_stream.tell() == 0
+            with Image.open(image_stream) as image:
+                assert image.format == "PNG"
+                assert image.width > 0 and image.height > 0
+                image.load()
+                seen.append(image.size)
+            return OCRResult(text=_MOCK_TEXT, backend_used="mock")
 
     converter = PdfConverterWithOCR()
-    with patch("pdfplumber.open") as mock_plumber:
-        mock_pdf = MagicMock()
-        mock_page = MagicMock()
-        mock_page.page_number = 1
-        mock_pdf.pages = [mock_page]
-        mock_pdf.__enter__.return_value = mock_pdf
-        mock_plumber.return_value = mock_pdf
-
-        with open(path, "rb") as f:
-            md = converter._ocr_full_pages(io.BytesIO(f.read()), svc)
-
-    expected = "## Page 1\n\n\n" "*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*"
-    assert (
-        md == expected
-    ), f"_ocr_full_pages must produce:\n{expected!r}\nActual:\n{md!r}"
+    markdown = converter._ocr_full_pages(
+        io.BytesIO(path.read_bytes()), RecordingOCRService()
+    )
+
+    assert len(seen) == 1
+    assert markdown == "## Page 1\n\n\n" + _OCR_BLOCK
 
 
 # ---------------------------------------------------------------------------
 
@@ -12,11 +12,16 @@
 Slide text, notes, and layout come from the core PPTX converter.
 """
 
+import io
 import sys
 from pathlib import Path
 from typing import Any
 
 import pytest
+from pptx import Presentation
+from pptx.chart.data import CategoryChartData
+from pptx.enum.chart import XL_CHART_TYPE
+from pptx.util import Inches
 
 sys.path.insert(0, str(Path(__file__).parent.parent / "src"))
 
@@ -138,54 +143,45 @@ def test_pptx_no_ocr_service_no_tags() -> None:
     assert "[End OCR]*" not in md
 
 
-def test_pptx_ocr_chart_no_title_text_frame() -> None:
-    from unittest.mock import MagicMock
-    from markitdown_ocr._pptx_converter_with_ocr import PptxConverterWithOCR
-
-    # python-pptx's ChartTitle.text_frame is destructive -- it creates a text
-    # frame if one isn't already present, so it never returns None.
-    # has_text_frame is the property that actually reflects presence/absence.
-    mock_chart = MagicMock()
-    mock_chart.has_title = True
-    mock_chart.chart_title.has_text_frame = False
-
-    mock_category = MagicMock()
-    mock_category.label = "Cat 1"
-    mock_chart.plots = [MagicMock(categories=[mock_category])]
-
-    mock_series = MagicMock()
-    mock_series.name = "Series 1"
-    mock_series.values = [10.0]
-    mock_chart.series = [mock_series]
-
+def _chart_presentation(title: str | None) -> io.BytesIO:
+    # Modify an in-memory copy of an existing deck; keep its fixture intact.
+    presentation = Presentation(TEST_DATA_DIR / "pptx_image_middle.pptx")
+    slide = presentation.slides.add_slide(presentation.slide_layouts[6])
+    data = CategoryChartData()
+    data.categories = ["Cat 1"]
+    data.add_series("Series 1", [10.0])
+    chart = slide.shapes.add_chart(
+        XL_CHART_TYPE.COLUMN_CLUSTERED,
+        Inches(1),
+        Inches(1),
+        Inches(5),
+        Inches(3),
+        data,
+    ).chart
+    chart.has_title = True
+    if title is not None:
+        chart.chart_title.text_frame.text = title
+    assert chart.chart_title.has_text_frame is (title is not None)
+    stream = io.BytesIO()
+    presentation.save(stream)
+    stream.seek(0)
+    return stream
+
+
+@pytest.mark.parametrize("title", [None, "Revenue"])
+def test_pptx_ocr_chart_title_text_frame(title: str | None) -> None:
+    stream = _chart_presentation(title)
+    # Reload the serialized package so the real parser reads title XML.
+    chart = Presentation(stream).slides[-1].shapes[-1].chart
     converter = PptxConverterWithOCR()
-    result = converter._convert_chart_to_markdown(mock_chart)
-
-    assert "### Chart" in result
+    result = converter._convert_chart_to_markdown(chart)
+    heading = "### Chart" if title is None else "### Chart: Revenue"
+    assert result.strip().splitlines()[0] == heading
     assert "Cat 1" in result
     assert "Series 1" in result
-    assert ":" not in result
-
-
-def test_pptx_ocr_chart_with_title_text_frame() -> None:
-    from unittest.mock import MagicMock
-    from markitdown_ocr._pptx_converter_with_ocr import PptxConverterWithOCR
-
-    mock_chart = MagicMock()
-    mock_chart.has_title = True
-    mock_chart.chart_title.has_text_frame = True
-    mock_chart.chart_title.text_frame.text = "Revenue"
-
-    mock_category = MagicMock()
-    mock_category.label = "Cat 1"
-    mock_chart.plots = [MagicMock(categories=[mock_category])]
-
-    mock_series = MagicMock()
-    mock_series.name = "Series 1"
-    mock_series.values = [10.0]
-    mock_chart.series = [mock_series]
-
-    converter = PptxConverterWithOCR()
-    result = converter._convert_chart_to_markdown(mock_chart)
-
-    assert "### Chart: Revenue" in result
+    assert "10.0" in result
+    # Accessing a title without a text frame must not create one.
+    assert chart.chart_title.has_text_frame is (title is not None)
+    stream.seek(0)
+    markdown = converter.convert(stream, StreamInfo(extension=".pptx")).markdown
+    assert result.strip() in markdown
@@ -52,3 +52,10 @@ The existing remote, credential, platform, and optional-dependency skip conditio
 apply. A consolidation should preserve test bodies, parametrization, fixtures,
 and skip conditions; compare collected cases as well as test results to catch
 accidentally dropped or overwritten tests.
+
+The `test_files/pdf_cleanup_{plain,form,mixed}.pdf` fixtures contain copied pages
+from `test.pdf` and `REPAIR-2022-INV-001_multipage.pdf`. They exercise real PDF
+parsing, fallback, and cleanup without changing either source fixture. Rebuild
+them with `python tests/test_files/generate_pdf_cleanup_fixtures.py` in an
+environment with PyMuPDF installed; PyMuPDF is needed only to rebuild the files,
+not to run the core tests.
@@ -97,6 +97,45 @@ class FileTestVector(object):
         ],
         must_not_include=[],
     ),
+    FileTestVector(
+        filename="pdf_cleanup_plain.pdf",
+        mimetype="application/pdf",
+        charset=None,
+        url=None,
+        must_include=[
+            "While there is contemporaneous exploration of multi-agent approaches",
+        ],
+        must_not_include=["ZAVA AUTO REPAIR", "| Customer Information |"],
+    ),
+    FileTestVector(
+        filename="pdf_cleanup_form.pdf",
+        mimetype="application/pdf",
+        charset=None,
+        url=None,
+        must_include=[
+            "ZAVA AUTO REPAIR",
+            "| Customer Information |",
+            "Gabriel Diaz",
+            "| GRAND TOTAL",
+            "5,738",
+        ],
+        must_not_include=["While there is contemporaneous exploration"],
+    ),
+    FileTestVector(
+        filename="pdf_cleanup_mixed.pdf",
+        mimetype="application/pdf",
+        charset=None,
+        url=None,
+        must_include=[
+            "ZAVA AUTO REPAIR",
+            "| Customer Information |",
+            "Gabriel Diaz",
+            "| GRAND TOTAL",
+            "5,738",
+            "While there is contemporaneous exploration of multi-agent approaches",
+        ],
+        must_not_include=[],
+    ),
     FileTestVector(
         filename="test_blog.html",
         mimetype="text/html",
 
@@ -0,0 +1,31 @@
+"""Rebuild copied PDF fixtures with PyMuPDF (already used by the OCR package).
+
+Run from packages/markitdown with: python tests/test_files/generate_pdf_cleanup_fixtures.py
+The existing source PDFs are opened read-only; only pdf_cleanup_*.pdf is written.
+"""
+
+from pathlib import Path
+
+import pymupdf
+
+
+FIXTURES = Path(__file__).resolve().parent
+
+
+def main() -> None:
+    with pymupdf.open(FIXTURES / "test.pdf") as plain, pymupdf.open(
+        FIXTURES / "REPAIR-2022-INV-001_multipage.pdf"
+    ) as form:
+        for name, pages in (
+            ("plain", [plain, plain, plain]),
+            ("form", [form, form, form]),
+            ("mixed", [form, plain, form]),
+        ):
+            with pymupdf.open() as copied:
+                for source in pages:
+                    copied.insert_pdf(source, from_page=0, to_page=0)
+                copied.save(FIXTURES / f"pdf_cleanup_{name}.pdf", deflate=True)
+
+
+if __name__ == "__main__":
+    main()
@@ -173,7 +173,11 @@ def test_image_without_llm_is_unchanged(
     skip_exiftool,
     reason="do not run if exiftool is not installed",
 )
-def test_markitdown_exiftool() -> None:
+def test_markitdown_exiftool(monkeypatch: pytest.MonkeyPatch) -> None:
+    import speech_recognition as sr
+
+    # Decode audio locally, but avoid the remote speech service in metadata tests.
+    monkeypatch.setattr(sr.Recognizer, "recognize_google", lambda *args, **kwargs: "")
     which_exiftool = shutil.which("exiftool")
     assert which_exiftool is not None
 
@@ -185,7 +189,7 @@ def test_markitdown_exiftool() -> None:
         assert target in result.text_content
 
     # Test setting the exiftool path through an environment variable
-    os.environ["EXIFTOOL_PATH"] = which_exiftool
+    monkeypatch.setenv("EXIFTOOL_PATH", which_exiftool)
     markitdown = MarkItDown()
     result = markitdown.convert(os.path.join(TEST_FILES_DIR, "test.jpg"))
     for key in JPG_TEST_EXIFTOOL: