Conversion checks and known limits

These notes describe AI Output Delivery Kit, including the examples linked from its guides. They do not certify the output in every word processor.

What we check

Our automated tests convert synthetic Markdown and inspect the DOCX package: paragraph text, real heading styles, nested list numbering, table cells, hyperlink targets, literal citation markers, Chinese text, and long identifiers. Browser tests exercise pasted input, selected files, duplicate filenames, partial failures, repeated downloads and unsafe raw HTML.

The editorial examples were generated on 30 October 2026 using the project's pinned Node.js dependencies, including docx 9.15.0 and markdown-it 15.0.3. Screenshots show the actual converter running in desktop Brave with synthetic input. A 390px browser viewport checks narrow layout; it is not an iPhone test. Inspecting OOXML proves document structure, not how a recipient's application will lay out each page.

Pagination and editing

Exports request repeated table headers and rows kept together. An independent LibreOffice inspection on 29 October checked 17 pages across five diagnostic DOCX files: the previously split action row stayed together, headers repeated, and a deliberately oversized row still spanned pages. That earlier check does not prove every newly edited document has the same layout.

Actual Microsoft Word and real iOS Safari have not been tested. Open the output in the recipient's application, edit a cell, and inspect every page after making changes. A row taller than a page may split. The browser preview does not show final DOCX page breaks.

Unsupported content

Images or detected math cause the entire document to export as literal Markdown source with a warning. The preview shows that source. Ordinary prices such as $19 and $1,250.00 retain formatting; paired math delimiters remain unsupported. Detection is a heuristic, not a math parser. Raw HTML is preserved as text and is never executed.

Tables with inconsistent column counts are rejected before export. Fix the source or keep it in a fenced code block if literal text is what you need. The table checklist includes a broken example you can reproduce.

Documents, analytics and prepared templates

Document conversion runs in the browser. The hosted site loads optional Google Analytics only after acceptance; the privacy page explains the metadata Google receives and how to withdraw consent. Rejecting does not restrict conversion.

All converter features, including batch ZIP export, are free. The optional $19 report pack contains separately authored DOCX templates. The converter does not import those templates or reproduce their prepared layout. Conversion does not verify facts or whether a citation supports a claim.