From scan to Markdown: clean up the meaning as well as the text
Extracted text is a starting point. Before you rely on it, compare the parts that matter with the original page.
A document can contain recognizable words in the wrong order. A date can look plausible and still be wrong. A heading can be flattened into the middle of a paragraph.
A useful cleanup pass checks accuracy first, restores the structure, and preserves anything you cannot confidently read.
Keep the original available
Save the scan beside the reconstructed document when your workflow allows it. Give the files related names so you can tell which source belongs to which text.
For example, workshop-notes.md and workshop-notes-scan.pdf are easier to match than two unrelated names generated during import.
A link in the note can help you return to the source:
Source: [Original workshop scan](workshop-notes-scan.pdf)Whether that link opens depends on the viewer and the file’s location. Check it after moving the files.
The point is to keep a way to resolve questions. Do not make yourself repeat the capture process because the extracted text looked complete.
Check the details that change an action
Read names, dates, amounts, addresses, and instructions against the original. Give particular attention to anything someone might use to pay, schedule, contact, or make a decision.
That does not mean every other word is unimportant. It means some errors are easier to carry into an action before anyone notices.
When a detail is unclear, mark it explicitly:
Delivery date: [Unclear in scan; confirm against original]Use your judgment about the wording. The marker should tell the next reader that the uncertainty comes from the source, rather than from an unfinished sentence.
Rebuild the reading order
Look at the page as a document. Where does the reader begin? What comes next? Does a sidebar interrupt the main explanation, or sit beside it?
A two-column page may need each column reconstructed separately before the text becomes one sequence. Repeated page headers and footers may need to be removed from the body. Keep information from them elsewhere if it matters, such as a document title or page reference.
Read the reconstructed passage aloud. An abrupt change of subject can reveal a column boundary or a misplaced caption.
Restore the relationships
Once the text is accurate enough to structure, use headings and lists to show how it fits together.
# Workshop notes
## Materials
- Paper
- Pens
- Printed examples
## Preparation
1. Read the examples.
2. Write down two questions.
3. Bring the questions to the discussion.A numbered procedure should retain its order. A list of materials should remain a list. A caption should stay near the image or be labeled clearly if the image is not included.
For a table, check that values still belong to the correct row and column. If you cannot reconstruct those associations confidently, leave a source reference and mark the unresolved area.
Keep transcription separate from your interpretation
You may want to summarize the page after cleaning it up. Put that summary in a separate section.
For example, use ## Reconstructed text for the source content and ## My notes for your observations. If you turn an instruction into an action item, make clear that you made that transformation.
That keeps a helpful interpretation from quietly replacing what the source actually said.
Do one final comparison
Check the beginning, the end, and every place you changed the structure. Make sure you did not lose a final paragraph or duplicate a section while cleaning it up.
Then test the source link. The finished note should be easier to use than the raw extraction while remaining honest about where the information came from.
Clean formatting helps. A reliable path back to the original helps even more when a question comes up later.
Keep exploring
Read the Everyday Markdown notes for more examples.