/**
 * @copyright Sister Software
 * @license AGPL-3.0
 * @author Teffen Ellis, et al.
 *
 *   Convert a JSONL of LabeledRow objects to a Parquet shard matching the v0.5.0 schema.
 *
 *   Ported faithfully from scripts/jsonl-to-parquet.py. The Python original wrote Parquet through
 *   PyArrow; this writes it through DuckDB (`@duckdb/node-api`) — `read_json` with an EXPLICIT
 *   `columns` type map projects the validated rows to the v0.5.0 schema, then `COPY … TO … (FORMAT
 *   PARQUET, COMPRESSION SNAPPY, ROW_GROUP_SIZE …)` emits the shard. DuckDB reproduces the exact
 *   logical schema PyArrow did — `VARCHAR` (UTF8) scalars, `VARCHAR[]` (LIST<UTF8>) for the string
 *   arrays, and `INTEGER[]` (LIST<INT32>) for the span offsets — in the exact column order below.
 *   Verified field-for-field against the PyArrow original (same column order, same logical types,
 *   same `list<element: …>` child naming, same values), so a PyArrow reader sees an identical
 *   table. The trainer in any case reads shards by column name (`pq.read_table(...).to_pylist()`),
 *   which is blind to physical layout. INT32 matches the corpus's native TS writer
 *   (`@mailwoman/corpus` `LABELED_ROW_SCHEMA`), which already writes the base shards this overlay
 *   rides alongside.
 *
 *   Schema: raw, tokens, labels, span_starts, span_ends, span_tags, country, locale, source,
 *   source_id, corpus_version, license, synth_method, synth_base_id.
 *
 *   The span triple (#519, v0.5.0 char-offset labels) is REQUIRED on every row: `alignRow` emits it
 *   on every labeled row, so a row arriving without it came from a producer that hasn't migrated —
 *   writing it would silently drop the v0.5.0 labels from the shard. Loud failure, naming the row
 *   number, instead.
 *
 *   Usage: mailwoman dev jsonl-to-parquet --input /tmp/po-box-labeled.jsonl --output
 *   /tmp/part-po-box.parquet
 */
/**
 * Options for {@linkcode jsonlToParquet}.
 */
export interface JSONLToParquetOptions {
    /**
     * The labeled-row JSONL to convert.
     */
    input: string;
    /**
     * The parquet shard to write.
     */
    output: string;
    /**
     * Parquet row-group size. Default 50000.
     */
    rowGroupSize?: number;
}
/**
 * Summary returned by {@linkcode jsonlToParquet}.
 */
export interface JSONLToParquetSummary {
    read: number;
    written: number;
    outPath: string;
}
/**
 * Convert a labeled-row JSONL to a v0.5.0-schema Parquet shard.
 */
export declare function jsonlToParquet(options: JSONLToParquetOptions, report?: (line: string) => void): Promise<JSONLToParquetSummary>;
//# sourceMappingURL=jsonl-to-parquet.d.ts.map