/**
 * Visual audit helpers — open a page, take viewport-sized screenshots,
 * and dump the DOM text inventory so the agent can compare scraped data
 * to what's visible to a human.
 *
 * Used during Phase 5e of site reconnaissance ("visual ground truth").
 * Stays out of the scraper hot path — recon-time only.
 *
 * Designed to be SITE-AGNOSTIC: we tried several real sites (99acres,
 * MagicBricks, IndiaMART) and the common shapes for label:value data are:
 *
 *   - "Label: Value" in a single text node (99acres, generic SSR)
 *   - <table><tr><th>...</th><td>...</td></tr> (IndiaMART spec tables)
 *   - <dl><dt>Label</dt><dd>Value</dd></dl> (definition lists, common)
 *   - sibling spans/divs with "label" or "value" class hints (MagicBricks)
 *
 * We emit each as a separate field so consumers know provenance.
 */
import type { Page } from "playwright";
export interface LabelValue {
    readonly label: string;
    readonly value: string;
}
export interface AuditResult {
    readonly screenshotPaths: readonly string[];
    /** All headings (h1/h2/h3) — gives you section structure. */
    readonly headings: readonly string[];
    /** "Label: Value" extracted from single text nodes (the old colon regex). */
    readonly colonPairs: readonly LabelValue[];
    /** <table><tr><th>label</th><td>value</td></tr> rows. */
    readonly tableRows: readonly LabelValue[];
    /** <dl><dt>term</dt><dd>def</dd></dl> entries. */
    readonly definitionList: readonly LabelValue[];
    /** Label/value pairs found via sibling-class heuristic (e.g. MagicBricks). */
    readonly siblingPairs: readonly LabelValue[];
    /**
     * Stacked layout pairs — two text children inside a flex/grid cell where
     * the first is short (≤40 chars) and visually sits above the second.
     * Catches the CommonFloor "label-above-value" pattern.
     */
    readonly stackedPairs: readonly LabelValue[];
    /**
     * Pairs derived from data-* attribute hints (`data-testid`, `data-aut-id`,
     * `data-qa`, `data-q`). Modern React stacks (OLX, Airbnb, Booking, …) tag
     * semantic fields here instead of via meaningful class names.
     */
    readonly attrPairs: readonly LabelValue[];
    /**
     * Merged + deduped union of all six sources. Use this for casual scans;
     * use the individual sources when you need provenance (e.g. table data is
     * usually more reliable than colon-regex hits).
     */
    readonly labelValuePairs: readonly LabelValue[];
    /** Status / badge text nodes. Conservatively filtered to avoid nav matches. */
    readonly statusBadges: readonly string[];
    /** Anchors that look like detail-page links. */
    readonly detailLinks: readonly {
        href: string;
        text: string;
    }[];
    /** Currency tokens visible in the rendered DOM. */
    readonly currencyMentions: readonly string[];
}
export interface AuditOptions {
    /** Output prefix for screenshots, e.g. "/tmp/99acres-srp-". */
    readonly screenshotPrefix: string;
    /** Scroll fractions of viewport height for each capture. Default [0, 1.2, 2.5]. */
    readonly scrollPoints?: readonly number[];
    /**
     * Extra regex patterns to recognise as detail-page links. Merged with the
     * defaults (which catch `/spid-`, `/pdpid-`, `/proddetail/`, `/dp/`, `/p/N`,
     * and any path segment with 4+ consecutive digits).
     */
    readonly detailLinkPatterns?: readonly RegExp[];
}
/**
 * Audit a page that's already been navigated to. Caller is responsible for
 * launching the browser, navigating, and any anti-bot warmup.
 */
export declare function auditPage(page: Page, options: AuditOptions): Promise<AuditResult>;
/**
 * Diff a scraped record against an audit. Returns labels visible on the page
 * whose values don't appear anywhere in the scraped record.
 *
 * Filters out URL/phone "values" — those are 100% noise based on real-site
 * testing (IndiaMART contact numbers, partial URL fragments).
 */
export declare function findMissingFields(audit: AuditResult, scraped: Record<string, unknown>): readonly LabelValue[];