What we did
We converted one Word document with MarkItDown 0.1.8, using Python 3.12.14 on Linux. The file includes headings, a short list, a two-row table, a currency amount, dates, a link, and separate Word header and footer text.
We used the local-file converter with third-party plugins disabled and no language-model client or cloud extraction configured. Python network connection and DNS events were blocked during conversion. This does not test every possible native-library or plugin network behavior.
What the output retained
- Main heading survives as a Markdown heading: verified in this sample.
- Next-actions subheading survives: verified in this sample.
- Both action items survive as list items: verified in this sample.
- The two-row task table and named columns survive: verified in this sample.
- The sample budget and review date survive unchanged: verified in this sample.
- The hyperlink label and URL survive: verified in this sample.
What it missed
The original file contains “SYNTHETIC HEADER: check this separately” and “SYNTHETIC FOOTER: demonstration only” in its Word page header and footer. Neither appears in the converted Markdown. A clean-looking output file can therefore still be incomplete.
Check the original before relying on the conversion, particularly if document identifiers, restrictions or other important information are stored outside the main body.
Read the actual output
# Project Cedar meeting notes
Synthetic example only. No real customer, meeting or budget is represented.
Budget approved: MYR 1,200.00. Review date: 18 October 2026.
## Next actions
* Prepare the project brief
* Check the figures against the original document
| | | |
| --- | --- | --- |
| Task | Owner | Due |
| Prepare brief | Alex | 18 October 2026 |
| Review budget | Morgan | 19 October 2026 |
Reference: [Artificially Nerdy](https://artificiallynerdy.com/)Inspect or repeat the experiment
- Original synthetic Word document
- Converted Markdown
- Results, versions and file checksums
- Reproduction script
- Exact Python dependency versions
For readers comfortable with Python: inspect the script, download it and the dependency list, then run it in a separate virtual environment. Installation needs internet access; the conversion itself uses the generated local file. The script refuses to overwrite existing results.
python -m venv .venv
# Activate the virtual environment for your operating system.
python -m pip install -r requirements.txt
python run.py --output experiment-outputWhat this experiment establishes
It verifies these results for this sample and configuration. An immediate second run produced identical input and output file checksums. We did not test PDFs, scans, images, tracked changes or a broad set of real-world documents. We did not measure better AI answers, time savings or conversion speed.
Preparing Markdown is a separate step from sending it to an AI assistant. Review confidentiality and the destination service before uploading any real document.