Inconsistent Markdown Output from export_to_markdown() vs Per-Page Export

[shubham-777]

Issue: Markdown Export from PDF Misaligns Text and Tables, and Skips Text Items

I'm encountering issues when exporting PDF content to Markdown. Specifically:

1. Inconsistent Output Between Full Document Export and Per-Page Export

When I use:

conv_result.document.export_to_markdown()

the output includes all expected text content. However, when I export each page separately using:

conv_result.document.export_to_markdown(page_no=2)

some TextItems are randomly skipped. Even if I export each page’s markdown to a separate file, the output differs from the full export, and some text content is missing.

2. Content Order Changes – Table Position is Wrong

In the PDF, the logical content sequence is:

2.2 Header  
2.2.1 Text content of first point  
      Table based on first point
2.2.2 Text content of second point  
2.2.3 Text content of third point  

But the Markdown output from export_to_markdown() (without specifying a page) is:

2.2 Header  
2.2.1 Text content of first point  
2.2.2 Text content of second point  
2.2.3 Text content of third point  
      Table based on first point

As you can see, the table is misplaced at the end.

3. generate_multimodal_pages() Produces Inconsistent or Missing Outputs

When using generate_multimodal_pages(), I receive:

• Two different outputs for the same page number, or
• None for certain page numbers

Pipeline Configuration

pipeline_options = PdfPipelineOptions()
pipeline_options.do_ocr = True
pipeline_options.ocr_options.force_full_page_ocr = True
pipeline_options.do_table_structure = True
pipeline_options.images_scale = IMAGE_RESOLUTION_SCALE
pipeline_options.table_structure_options.do_cell_matching = False
pipeline_options.table_structure_options.mode = TableFormerMode.FAST
pipeline_options.generate_picture_images = True
pipeline_options.do_picture_description = False
pipeline_options.do_formula_enrichment = False

doc_converter = DocumentConverter(
    format_options={
        InputFormat.PDF: PdfFormatOption(
            pipeline_options=pipeline_options,
            backend=DoclingParseV4DocumentBackend
        )
    }
)

Package Versions

Note: Have also tried new version 2.31.0 but didnt work

docling==2.28.4  
docling-core==2.28.1  
docling-ibm-models==3.4.2  
docling-parse==4.0.1  

Has anyone else faced similar issues or found a workaround for:

• Ensuring per-page exports don't skip text items?
• Preserving correct order between text and tables?
• Consistent outputs from generate_multimodal_pages()?

Any help or insights would be appreciated!