HTML to Markdown, losses declared

Because a Markdown pipe table has no way to hold another table inside a cell — the syntax is one line per row, so anything containing a line break or a block element cannot go in. The same limit catches colspan, rowspan, a list inside a cell and a heading inside a cell, and this converter checks for all five before it commits to a pipe table. When one of them appears you get whatever the policy control says: a bullet list of “**Header:** value” pairs under Approximate, the original <table> element under Keep, or the cell text separated by em dashes under Drop.

The element and its structure, minus every attribute that is not load-bearing: href, src, alt, title, colspan, rowspan, align, style, start and type survive, and class, id, data-*, aria-* and inline event handlers are stripped. That distinction is what makes the option usable — pasting a fragment of a modern site with Keep on would otherwise drop 400 characters of utility classes into your Markdown for every div. Raw HTML inside Markdown is legal and GitHub renders a filtered subset of it, so kept fragments usually still display where the file ends up.

Because those editors express formatting as inline styles on spans rather than as semantic elements: a heading is often a span with font-size:20pt, not an h2, and bold is font-weight:700 on a span rather than a strong. This converter reads font-weight, font-style and text-decoration and maps them to bold, italic and strikethrough under the Approximate policy, which recovers the emphasis, but a size-only heading has no equivalent it can trust and stays a paragraph. Pasting from a rendered web page usually goes much better, because browsers ship real h1-h6 and strong elements.

They stop text from being re-read as syntax: a literal asterisk, backtick, square bracket or angle bracket in the source page would otherwise turn into emphasis, code or a broken link when the Markdown is rendered, so each one is escaped. Line-leading characters get the same treatment — a paragraph that happens to start with “1. ” or “- ” would become a list item without it. Underscores are deliberately left alone, because inside a word they never open emphasis and escaping them would mangle every snake_case identifier on the page.

No, and it is not an oversight — this page has no server, so there is nothing that could fetch a URL on your behalf, and a browser cannot read another site's HTML directly because of the same-origin policy. The two routes that do work are copying the rendered selection, which brings the markup with it through the clipboard, and saving the page from your browser and opening the .html file here.

None of those four exist in Markdown, so they follow the policy control like everything else. Under Approximate, mark becomes bold and kbd becomes inline code, while u, sub and sup keep their text and lose the presentation; under Keep, all four stay as their original tags; under Drop, each becomes plain text. Underline is the honest casualty here: Markdown has no underline at all, and mapping it to emphasis would be a lie about what the source said.

Because a div carries no meaning that Markdown can represent — it is layout, and layout is exactly what Markdown throws away. Divs, sections, articles, headers, footers and navs are treated as transparent: their children are converted and the wrapper is discarded, with a count of how many were removed shown under the output. If a div's only content was inline text, it becomes a paragraph so the text does not run into its neighbours.