/**
 * Markdown → the words, for a caption someone READS.
 *
 * The live voice caption is the one surface where the model's raw Markdown was
 * printed literally: bold stars, hash headings, table pipes, fence backticks.
 * The audio path has been clean for a while (`toSpeakableText`), but that
 * function is the wrong tool here and reusing it would be worse than the bug.
 *
 * ★ `toSpeakableText` is a SCRIPT FOR A VOICE, not a Markdown cleaner. It
 * replaces a table with "Table with 40 rows and 2 columns", a fenced block with
 * "Code block", a link with the word "link", and rewrites `x@y.com` as "x at y
 * dot com" — sensible in an earpiece, destructive on a screen where the reader
 * can simply read the thing. Worse, those replacements come from an English
 * default list only localised inside the speech provider, so reusing it would
 * print English chrome inside a Tamil or Hindi caption: the exact
 * mixed-language defect that code exists to prevent.
 *
 * So this strips MARKERS and keeps CONTENT. Every cell value, all code text,
 * bare URLs, email addresses, emoji and punctuation survive verbatim. Nothing
 * is summarised, truncated or annotated: a caption that quietly drops the
 * user's numbers is a worse failure than one showing a stray asterisk. Where
 * this function is unsure, it leaves the text ALONE.
 *
 * ★ Newlines are PRESERVED, which only helps where the element renders them.
 * `VoiceCaption` carries `whitespace-pre-wrap` for exactly that. The mini
 * controller deliberately does NOT: it is `line-clamp-3`, and real newlines
 * each burn one of the three lines — measured at the strip's true width, that
 * cut a 116-character reply to 41 visible characters. Markers still go; the
 * breaks are allowed to collapse on the clamped surface.
 *
 * ★ NO LOOKBEHIND ANYWHERE IN CODE. Safari < 16.4 throws on that assertion
 * class at PARSE time, not match time — one literal takes down the whole chunk
 * on those devices, and this ships to phones. Lookahead is fine; where a
 * leading boundary is needed it is CAPTURED and put back. Same rule as
 * `speakableText.ts`, and `captionText.test.ts` asserts it on the source.
 */
/**
 * Strip Markdown markers for display, keeping all of the content.
 *
 * Safe on plain text: a reply with no Markdown comes back unchanged.
 */
export declare function toCaptionText(markdown: string): string;
//# sourceMappingURL=captionText.d.ts.map