# PDF to Markdown: native files and local display checks

Checked 14 October 2026. We reused the completed Doc2X result of the public,
self-authored two-page motion-notes PDF from the earlier LaTeX test. We selected
Markdown and clicked Start export. No new upload, OCR or recognition edit was
needed. The source still says “PDF to LaTeX demo” and discusses TeX; this wording
has not been relabelled inside the actual PDF or exported Markdown.

Observed export options: Markdown table, local image package, backslash math
delimiters, retain formulas, keep tables by original page, image enhancement off.
The former Word + preserved-layout preference was restored by another export,
then reopened and checked. That extra Word download is not this Markdown proof.

The native ZIP is unchanged: one 1,916-byte MD and one 27,456-byte JPG under
images/1_279_471_1061_604_0.jpg. Extract the complete ZIP. Downloading the MD alone
does not supply its image. Three data rows A/B/C retain 6.00/12.00/15.00 distances,
2.00/4.00/5.00 times and 3.00 speeds. There are two display formulas and eight
inline math spans, using \[...\] and \(...\). All six headings use ##.

The chart image bytes match the earlier native LaTeX export. Its top label,
“Distance s (m) | Constant-speed model: s = 3t”, is absent from this image and the
MD. The caption is present. No repair was silently applied. This is a simple
digital sample, not a scan, multicolumn or merged-cell accuracy benchmark.

Local controls use Pandoc 3.4, a separately authored stylesheet/notice and Chrome:
- complete: pipe tables + tex_math_single_backslash, MathML output; 10 math
  nodes, 3 data rows, one loaded image;
- missing-image: identical MD and renderer, deliberately omitted image folder;
  one broken image, formulas/table still present;
- math-disabled: identical MD and image, math extensions disabled; zero math
  nodes, three data rows and image remain. In this Pandoc configuration, raw TeX
  commands can be omitted in HTML, so the displayed equations are incomplete.

The HTML previews are local demonstrations, not product HTML exports, Typora,
Obsidian or GitHub imports. Renderer behavior differs. Native MD bytes are the
same in all three cases. MathML display or successful image loading does not
prove source completeness. Keep the original PDF alongside your edited notes.

The public proof ZIP contains the unchanged native export, a copy of the source
PDF, the labelled local controls and aggregate checks. Only the input and native
export are product workflow evidence; local HTML/CSS and checks were authored
afterward. Current sign-in, quota and export conditions apply.
