> ## Documentation Index
> Fetch the complete documentation index at: https://docs.trueparser.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Parquet

> Public result contract for Parquet outputs.

# Parquet

Use this contract for Parquet dataset results. It is designed for data engineering and analytics workflows where teams need schema inspection, dataset profiling, lakehouse ingestion, feature extraction, and validation over columnar data.

## Top-level envelope

```json theme={null}
{
  "schema_version": "1.0",
  "document": {},
  "warnings": [],
  "content": []
}
```

## Document fields

| Field                                                                                                                                                                                                           | Notes                                 |
| --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------- |
| `source_file`, `document_name`                                                                                                                                                                                  | Source file and display name.         |
| `format`                                                                                                                                                                                                        | Always `parquet`.                     |
| `format_family`                                                                                                                                                                                                 | Always `parquet_file`.                |
| `document_id`, `content_hash`                                                                                                                                                                                   | Optional stable identifiers.          |
| `is_partial`                                                                                                                                                                                                    | Public completeness flag.             |
| `metadata`                                                                                                                                                                                                      | Additive document metadata.           |
| `record_count`, `schema_node_count`, `leaf_column_count`, `row_group_count`, `row_record_count`, `column_chunk_count`, `statistics_count`, `key_value_metadata_count`, `semantic_tag_count`, `diagnostic_count` | Summary counts when available.        |
| `row_count`, `compressed_size`, `uncompressed_size`, `confidence`, `status`, `completeness`                                                                                                                     | Public summary values when available. |

## Universal content/block shape

Every public content record uses the same base shape.

| Field          | Notes                             |
| -------------- | --------------------------------- |
| `id`           | Stable record id.                 |
| `type`         | Public record type.               |
| `order`        | Deterministic order.              |
| `path`         | Public structural path.           |
| `parent_id`    | Parent record id.                 |
| `depth`        | Structural depth.                 |
| `source_ref`   | Provenance object.                |
| `is_inferred`  | Inference marker.                 |
| `warnings`     | Record-local notes.               |
| `content_hash` | Optional hash.                    |
| `text`         | Searchable text projection.       |
| `attributes`   | Parquet-specific structured data. |

Common record types include `schema_node`, `column`, `row_group`, `row`, `column_chunk`, `statistics`, `key_value_metadata`, `semantic_tag`, and `diagnostic`.

## Warnings

* `warnings` is always present.
* Use plain strings.
* Keep warnings human-readable.
* Surface recoverable issues without changing the public shape.

## What clients can rely on

* Schema order stays stable.
* Row-group and row order stay stable.
* Raw metadata and summary values remain separate.
* Additive fields do not break older clients.
* The public contract does not expose internal transport or worker details.

## Example

```json theme={null}
{
  "schema_version": "1.0",
  "document": {
    "format": "parquet",
    "format_family": "parquet_file",
    "source_file": "orders.parquet",
    "row_group_count": 1,
    "record_count": 2,
    "metadata": {}
  },
  "warnings": [],
  "content": [
    {
      "id": "schema_0001",
      "type": "schema_node",
      "order": 1,
      "path": ["schema_node", "schema-order-id"],
      "parent_id": null,
      "depth": 0,
      "source_ref": {
        "record_id": "schema-order-id",
        "record_type": "schema_node",
        "source_file": "orders.parquet"
      },
      "is_inferred": false,
      "warnings": [],
      "text": "order.id",
      "attributes": {}
    }
  ]
}
```