Limitations

Markdown writes less than HTML. The conversion keeps the structure and the text of the converted elements, and loses the rest:

• Attributes and styles. Only href and title of links, src, alt and title of images, the start of ordered lists, the checked state of checkboxes, the alignment of table cells and the language of code blocks are read. The class, id, style and data-* attributes are lost, except on unknown elements written as raw HTML.
• Fragments. Convert takes the content of a body. A complete page gives its head elements as raw HTML: see Convert a complete document.
• Whitespace. The whitespace of the text is collapsed, as a browser displays it. Only code blocks keep their line breaks and their indentation. CSS is not read: an element styled with white-space: pre is collapsed too.
• Code. A code block or a code span keeps the text of its content, without its elements.
• Tables. A pipe table has one header row, and one line per row. colspan and rowspan are ignored: a cell takes one column, and the shorter rows are completed with empty cells. Lists, code blocks and other blocks of a cell are flattened onto one line, separated by <br>.
• Forms and media. input elements other than checkboxes are removed. video, audio, iframe, sub, sup, u, kbd and the other elements without a Markdown syntax are unknown elements: raw HTML, removed or unwrapped, depending on UnknownElementHandling.
• Security. The converter is not a sanitizer: see Convert untrusted HTML.

The escapes are conservative: the Markdown can hold a backslash that is not needed, such as 1\.20 at the start of a table cell. It renders the same text.