> ## Documentation Index > Fetch the complete documentation index at: https://docs.textin.ai/llms.txt > Use this file to discover all available pages before exploring further. # Response Format > This page describes the **Extract** response format. The Extract response returns top-level request metadata and a `result` object. The `result` object contains the extracted data in `extracted_schema` and can also include citations, page-level fields, stamp results, and usage metadata. ## Top-level response The `entity_extraction` endpoint returns the following top-level structure: ```json theme={null} { "code": 200, "message": "Success", "status": "finished", "version": "v3.x.x", "duration": 28450, "part_durations": { "parse_duration": 0, "doc_cvt_duration": 48, "retrieve_duration": 0, "prompt_duration": 0, "llm_duration": 26950, "format_duration": 12 }, "result": { "extracted_schema": {}, "pages": [ { "height": 1584, "width": 1224, "angle": 0, "status": "Success", "durations": 0, "page_number": 1, "image_id": "" } ], "stamps": [], "success_count": 1, "usage": { "acgpt_request_number": 3, "prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0, "request_number": 0, "acgpt_prompt_tokens": 4200, "acgpt_completion_tokens": 120, "acgpt_total_tokens": 4320 } }, "x_request_id": "req_example_id" } ``` ### Top-level fields | Field | Type | Description | | ---------------- | ------- | -------------------------------------------------------------- | | `code` | integer | Business-level response code returned by the service | | `message` | string | Response message | | `status` | string | Request status, such as `finished` | | `version` | string | Extract service version | | `duration` | integer | Total processing time reported by the service, in milliseconds | | `part_durations` | object | Timing breakdown for internal processing stages | | `result` | object | Extraction result object | | `x_request_id` | string | Request identifier returned with every response | ### part\_durations fields Observed fields in `part_durations` include: | Field | Type | Description | | ------------------- | ------- | ---------------------------- | | `parse_duration` | integer | Parse stage duration | | `doc_cvt_duration` | integer | Document conversion duration | | `retrieve_duration` | integer | Retrieval stage duration | | `prompt_duration` | integer | Prompt preparation duration | | `llm_duration` | integer | LLM inference duration | | `format_duration` | integer | Output formatting duration | All duration values are reported in milliseconds. ## The result object The result object can include the following fields: | Field | Type | Description | | ------------------ | ------- | ---------------------------------------------------------------------- | | `extracted_schema` | object | Structured data returned according to your schema | | `citations` | object | Field-level source citations, returned when citation output is enabled | | `pages` | array | Page-level processing metadata | | `stamps` | array | Stamp results, returned when stamp output is enabled | | `success_count` | integer | Number of pages processed successfully | | `usage` | object | Usage metadata returned by the service | ## The extracted\_schema object The `result.extracted_schema` object contains the extracted data in the structure defined by your schema. ### Field behavior * String fields that cannot be extracted may be returned as empty strings. * Array fields that cannot be extracted may be returned as empty arrays. * The current response format does not force missing fields to return `null`. * Number and integer fields are returned according to the schema type when a value is extracted successfully. * Enum fields return one of the predefined values defined in the schema when a value is extracted successfully. ### Example: full response with extracted\_schema ```json theme={null} { "code": 200, "message": "Success", "status": "finished", "version": "v3.x.x", "duration": 28450, "part_durations": { "parse_duration": 0, "doc_cvt_duration": 48, "retrieve_duration": 0, "prompt_duration": 0, "llm_duration": 26950, "format_duration": 12 }, "result": { "extracted_schema": { "purchase_order": "PO-100045", "purchase_date": "2026-02-10", "total": "$1,250.00", "vendor_name": "Example Supplier Inc.", "ship_to_name": "Example Distribution Center" }, "pages": [ { "height": 1584, "width": 1224, "angle": 0, "status": "Success", "durations": 0, "page_number": 1, "image_id": "" } ], "stamps": [], "success_count": 1, "usage": { "acgpt_request_number": 3, "prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0, "request_number": 0, "acgpt_prompt_tokens": 4200, "acgpt_completion_tokens": 120, "acgpt_total_tokens": 4320 } }, "x_request_id": "req_example_id" } ``` ## Citations The `result.citations` object identifies the source location of extracted values. When citation output is enabled, the response can include a `citations` object keyed by field name. ### Example: full response with citations ```json theme={null} { "code": 200, "message": "Success", "status": "finished", "version": "v3.x.x", "duration": 28450, "part_durations": { "parse_duration": 0, "doc_cvt_duration": 48, "retrieve_duration": 0, "prompt_duration": 0, "llm_duration": 26950, "format_duration": 12 }, "result": { "extracted_schema": { "purchase_order": "PO-100045" }, "citations": { "purchase_order": { "value": "PO-100045", "bounding_regions": [ { "page_number": 1, "position": [326, 478, 638, 478, 638, 521, 326, 521], "text": "PO-100045" } ], "llm_confidence": 0.97, "llm_confidence_level": "high" } }, "pages": [ { "height": 1584, "width": 1224, "angle": 0, "status": "Success", "durations": 0, "page_number": 1, "image_id": "" } ], "stamps": [], "success_count": 1, "usage": { "acgpt_request_number": 3, "prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0, "request_number": 0, "acgpt_prompt_tokens": 4200, "acgpt_completion_tokens": 120, "acgpt_total_tokens": 4320 } }, "x_request_id": "req_example_id" } ``` ### Citation fields | Field | Type | Description | | ---------------------- | ------ | ------------------------------------------- | | `value` | string | Extracted value for the field | | `llm_confidence` | number | LLM confidence score, when available | | `llm_confidence_level` | string | Confidence label, when available | | `bounding_regions` | array | Source regions used for the extracted value | ### Bounding region fields | Field | Type | Description | | ------------- | ------- | -------------------------------------------- | | `page_number` | integer | Page number, starting at 1 | | `text` | string | Source text used for the citation | | `position` | array | Four-point bounding box in pixel coordinates | ### Coordinate format Citation positions use four-point bounding boxes: ```text theme={null} [x1, y1, x2, y2, x3, y3, x4, y4] ``` The four points are ordered as: ```text theme={null} (x1,y1) -------- (x2,y2) | | | Field Area | | | (x4,y4) -------- (x3,y3) ``` ## Pages The `result.pages` field contains page-level processing metadata. Observed page fields include: | Field | Type | Description | | ------------- | ------- | ----------------------------------------------------------------- | | `page_number` | integer | Page number, starting at 1 | | `width` | integer | Page width in pixels | | `height` | integer | Page height in pixels | | `angle` | integer | Page rotation angle reported by the service | | `status` | string | Page processing status, such as `Success` | | `image_id` | string | Internal page image identifier; may be an empty string | | `durations` | number | Page processing duration reported by the service, in milliseconds | ### Example: pages metadata ```json theme={null} { "pages": [ { "height": 1584, "width": 1224, "angle": 0, "status": "Success", "durations": 0, "page_number": 1, "image_id": "" } ] } ``` ## Stamps The `result.stamps` field is returned as an array when stamp output is enabled. If no stamps are detected, the field may be returned as an empty array: ```json theme={null} { "stamps": [] } ``` When stamp results are returned, each item can include the following fields: | Field | Type | Description | | ------------- | ------ | ------------------------------------- | | `color` | string | Stamp color | | `position` | array | Four-point bounding box for the stamp | | `stamp_shape` | string | Stamp shape | | `type` | string | Stamp type | | `value` | string | Recognized stamp text | ## Usage The `result.usage` field contains usage metadata returned by the service. Observed fields include: | Field | Type | Description | | ------------------------- | ------- | -------------------------------------------- | | `acgpt_request_number` | integer | Number of internal ACGPT requests | | `prompt_tokens` | integer | Prompt token count for the standard path | | `completion_tokens` | integer | Completion token count for the standard path | | `total_tokens` | integer | Total token count for the standard path | | `request_number` | integer | Number of requests for the standard path | | `acgpt_prompt_tokens` | integer | Prompt token count for the ACGPT path | | `acgpt_completion_tokens` | integer | Completion token count for the ACGPT path | | `acgpt_total_tokens` | integer | Total token count for the ACGPT path | ### Example: usage metadata ```json theme={null} { "usage": { "acgpt_request_number": 3, "prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0, "request_number": 0, "acgpt_prompt_tokens": 4200, "acgpt_completion_tokens": 120, "acgpt_total_tokens": 4320 } } ``` ## Extract-specific errors Most authentication, file, and request validation errors are shared across Parse and Extract. See [Errors](/xparse/errors) for the full common error reference. The following errors are specific to Extract: | Code | Description | Solution | | ------- | --------------------- | ----------------------------------------------------------- | | `40306` | QPS limit exceeded | Reduce request rate or request a higher limit. | | `50011` | LLM connection failed | Retry the request. Contact support if the problem persists. | | `50012` | LLM engine failed | Retry the request. Contact support if the problem persists. | ## Related resources Extract your first document. Create extraction schemas. Improve extraction accuracy. Review common error codes. Review supported formats and limits.