ChatGPT deep-research exports are markdown with quirks: citations are invisible private-use Unicode sequences (U+E200 + cite + U+E202-separated turnview / turnsearch tags) that point at sources the file itself never lists, math arrives as raw LaTeX (inline) and [display], and flowcharts arrive as mermaid code. This converter handles all three: citation clusters become numbered superscripts [1][2] linked to a references section; KaTeX (bundled locally, CSS and fonts embedded as data URIs) renders the math so the file works fully offline; and a hand-rolled SVG layer draws the mermaid flowchart subset — rectangles, decision diamonds, edge labels and even cycles routed around the side. The layout follows a clean paper-card article style with a clickable table of contents, styled tables with horizontal scroll, blockquotes and a print stylesheet, in your choice of accent colour. The markdown file contains no source URLs, so citations reference their ChatGPT source tags by default; paste the report’s source list (one per line — bare URLs fill references in order of first citation, "turn2view0 https://…" maps a tag explicitly, or "Title https://…" gives a named source) and the references become real links. Content is preserved verbatim — nothing is summarised or dropped.