/**
 * Paper and page text: PDF extraction, sections, quote matching and short
 * passages.
 *
 * Quote matching follows the W3C TextQuoteSelector idea: the matched text plus
 * a little text before and after it, so a reader can find the place again. It
 * tries an exact match first, then a match after normalising case, spacing,
 * quotes, ligatures and hyphens between letters, then the closest passage
 * with a similarity score.
 */
export type Section = {
    heading: string;
    level: number;
    text: string;
    page?: number;
};
/** Pages up to this size come back whole; longer ones come back by section. */
export declare const WHOLE_TEXT_CHARS = 40000;
/** Split markdown into sections at its headings. Text before the first heading is "(start)". */
export declare function splitSections(markdown: string): Section[];
/** Markdown to plain text: link text without its URL, no emphasis marks or escapes. */
export declare function stripMarkdown(text: string): string;
/** Cut a section longer than max at paragraph breaks into parts, so no one reply is unbounded. */
export declare function splitLongSection(section: Section, max: number): Section[];
/** A line that reads as a section heading in PDF text: "2 SPADL: A LANGUAGE ...", "3.1 Data", "ABSTRACT". */
export declare function headingLevel(line: string): number;
/**
 * Sections of a PDF from its page texts. Headings are found by the rules in
 * headingLevel; a heading in capitals that wraps onto a second line in
 * capitals is joined. With fewer than three headings, the sections are pages.
 */
export declare function pdfSections(pages: string[]): Section[];
export type PdfText = {
    pages: string[];
    title?: string;
    author?: string;
    doi?: string;
};
/** Text of every page of a PDF, with the title, author and first DOI it states. */
export declare function readPdf(data: Uint8Array): Promise<PdfText>;
export declare function isPdf(data: Uint8Array): boolean;
export declare function outline(sections: Section[]): string;
export declare function sectionText(section: Section): string;
/** Whole text when short, else the outline and the first sections that fit. */
export declare function renderFull(sections: Section[], requested?: number): {
    lines: string[];
    isError?: boolean;
};
/** Normalised text plus, for each of its characters, the index in the original. */
export type Normalised = {
    text: string;
    map: number[];
};
/**
 * Lower case, one space for any run of white space, plain quotes, ligatures
 * expanded, soft hyphens and hyphens between letters removed, markdown
 * emphasis and link brackets dropped.
 */
export declare function normalise(input: string): Normalised;
export type QuoteMatch = {
    kind: "exact" | "normalised" | "close" | "none";
    /** 1 for exact and normalised matches; the similarity of the closest passage otherwise. */
    score: number;
    /** Offsets in the original text. */
    start: number;
    end: number;
};
/** The score a closest passage needs to count as a close match. */
export declare const CLOSE_MATCH_SCORE = 0.85;
/** Find a quote in a text. Offsets refer to the text as given. */
export declare function findQuote(source: string, quote: string): QuoteMatch;
/** Text before and after a span, cut at word boundaries. */
export declare function context(source: string, start: number, end: number, chars: number): {
    prefix: string;
    suffix: string;
};
/** Up to `limit` passages of at most `cap` characters around the places a query's terms occur together. */
export declare function passages(source: string, query: string, cap: number, limit?: number): string[];