/**
 * ChatgptUiScrapeParser — PURE, network-free transform from raw scraped
 * ChatGPT-UI answer-page markdown into a normalized `ParsedAnswer` (in-house-
 * ai-visibility, Sprint 10; arch-20260717-in-house-oss-ai-visibility-
 * architecture.md:139-141,250-251,281).
 *
 * This module has ZERO dependencies on damcrawler, egress, or any network
 * primitive — it is a synchronous, deterministic markdown -> data mapping,
 * mirroring `DeterministicMentionCitationExtractor`'s "PURE, synchronous,
 * network-free transform" contract (`./mention-citation-extractor.ts:8`).
 * `ScrapeArmEngineProvider` is the ONLY caller, and it invokes `.parse()`
 * strictly AFTER the raw scraped markdown has already passed through the
 * fail-closed `ContentSanitizer` at the network->in-process boundary
 * (sc-10-2, Sprint-9 F1 lesson) — this parser never sees unsanitized text
 * in production, but it does not (and must not) assume that; it never
 * throws regardless of input.
 *
 * ChatGPT's exported/rendered answer-page markdown is NOT a stable, fully
 * documented format — this parser degrades gracefully on drift: "parser
 * drift => empty answerText => downstream observation counts as not-
 * mentioned" (arch:281) is the explicit contract, not a bug to fix later.
 * Empty/whitespace/malformed input always yields `{ answerText: "",
 * citations: [] }` — this class NEVER fabricates a positive observation.
 *
 * Extraction strategy (deterministic, markdown-shape heuristics only):
 *   1. `answerText` — the raw markdown with any trailing "Sources"/
 *      "Citations" section (and any markdown link syntax) stripped, then
 *      trimmed. A ChatGPT-UI answer page renders the assistant's prose
 *      followed by an optional sources block; this parser treats everything
 *      before that block (or the whole trimmed text, when no block is
 *      found) as the answer.
 *   2. `citations` — markdown links (`[title](url)`) found ANYWHERE in the
 *      raw markdown, deduplicated by URL, mapped to `GroundedCitation`
 *      `{ url, title }`. A bare autolink (`<https://...>`) or a bare URL is
 *      also captured with its url as its own title (no better title is
 *      derivable). Malformed/empty href or link text never crashes the
 *      parser — such candidates are simply skipped.
 */
import type { GroundedCitation } from "../../providers/grounded-search.js";
/** Raw scraped page handed to the parser — already sanitized by the caller (sc-10-2). */
export type RawScrape = {
    url: string;
    markdown: string;
    links?: string[];
};
/** Normalized parse result. `citations` reuses the locked `GroundedCitation` shape. */
export type ParsedAnswer = {
    answerText: string;
    citations: GroundedCitation[];
};
export interface EngineScrapeParser {
    parse(raw: RawScrape): ParsedAnswer;
}
/**
 * `EngineScrapeParser` for the `"chatgpt-ui"` scrape engine. PURE,
 * synchronous, dependency-free; never throws; empty/malformed input yields
 * `{ answerText: "", citations: [] }`.
 */
export declare class ChatgptUiScrapeParser implements EngineScrapeParser {
    parse(raw: RawScrape): ParsedAnswer;
}
//# sourceMappingURL=engine-scrape-parser-chatgpt.d.ts.map