/**
 * Heading-aware document chunker.
 *
 * `chunk()` is the size-based sliding-window primitive — kept as the
 * low-level splitter that operates within a single heading section.
 *
 * `chunkDocument()` is the real entry point: it walks the document,
 * splits on heading boundaries first, preserves fenced code blocks
 * intact, classifies each segment (frontmatter / heading / paragraph /
 * code / list), attaches the heading_path breadcrumb, and only then
 * size-splits oversized segments. This gives retrieval real metadata
 * to rank on (heading_path match, chunk_type filter, title boost).
 *
 * Boundaries are preserved in absolute char offsets against the input
 * text so callers can slice back into the source if needed. The
 * (char_start, char_end) range is byte-faithful to the original input —
 * text.slice(char_start, char_end) returns the exact source bytes
 * including any CR, original line separators, and surrounding whitespace
 * that the chunker trimmed when building chunk.text. The chunk.text
 * itself is the trimmed/normalized content; rely on the slice for
 * byte-exact recovery.
 */
export type ChunkType = "heading" | "paragraph" | "code" | "list" | "frontmatter";
export interface Chunk {
    index: number;
    char_start: number;
    char_end: number;
    text: string;
    heading_path: string[];
    chunk_type: ChunkType;
}
export interface ChunkOptions {
    chunk_chars: number;
    chunk_overlap: number;
}
export declare const DEFAULT_CHUNK: ChunkOptions;
export interface ChunkedDocument {
    title: string | null;
    chunks: Chunk[];
}
interface RawChunk {
    index: number;
    char_start: number;
    char_end: number;
    text: string;
}
/**
 * Primitive size-based sliding-window splitter. Used internally by
 * chunkDocument() to size-split oversized heading sections; also exposed
 * so existing tests can exercise window math directly.
 */
export declare function chunk(text: string, opts?: ChunkOptions): RawChunk[];
/**
 * Heading-aware document chunker.
 *
 * Behavior:
 *  - YAML frontmatter at the very top becomes a single "frontmatter" chunk.
 *  - Every markdown heading (# through ######) updates a depth-indexed
 *    stack; chunks inherit the heading_path of the section they sit in.
 *  - Fenced code blocks (``` or ~~~) are preserved intact — never split.
 *  - Non-code sections are classified as "list" if majority of non-empty
 *    lines are list items, else "paragraph".
 *  - Sections that exceed chunk_chars are size-split via chunk() with
 *    overlap; sub-chunks inherit the parent's heading_path + type.
 *  - Title = first H1 heading after frontmatter, else null.
 */
export declare function chunkDocument(text: string, opts?: ChunkOptions): ChunkedDocument;
export {};
//# sourceMappingURL=chunker.d.ts.map