API Reference

Extract

Copy Markdown

Open in Claude

Open in ChatGPT

Open in Cursor

---

Copy Markdown

View as Markdown

Review

Create Extract Review

POST/api/v2/extract/{job_id}/review

Get Extract Review

GET/api/v2/extract/{job_id}/review

Change Extract Review Fields

POST/api/v2/extract/{job_id}/review/fields

Approve Extract Review

POST/api/v2/extract/{job_id}/review/approve

Reject Extract Review

POST/api/v2/extract/{job_id}/review/reject

Reopen Extract Review

POST/api/v2/extract/{job_id}/review/reopen

List Extract Review Events

GET/api/v2/extract/{job_id}/review/events

ModelsExpand Collapse

ExtractReview object { id, confidence_threshold, created_at, 16 more }

An extract job’s review: its settings, flags, field actions and decision, with the result as read and as reviewed.

id: string

Unique identifier of the review.

confidence_threshold: number

Fields read below this confidence are flagged.

created_at: string

When review was turned on.

formatdate-time

decided_at: string

When it was approved or rejected.

formatdate-time

decided_by: string

ID of the user who approved or rejected it.

decided_reason: string

Why it was rejected.

field_actions: map[object { action, actor_id, at, value } ]

The latest action per field, keyed by JSON Pointer.

action: "clear" or "confirm" or "correct"

confirm, correct or clear.

One of the following:

"clear"

"confirm"

"correct"

actor_id: string

ID of the user who acted.

at: string

When they acted.

formatdate-time

value: optional unknown

The corrected value, for correct.

fields_in_scope: array of string

JSON Pointers of the fields the threshold applies to; empty means all.

flags: array of object { handled, path, reason, 2 more }

Fields the review asks a person to check.

handled: boolean

Whether someone confirmed, corrected or cleared it.

path: string

JSON Pointer of the field.

reason: "below_threshold" or "required_empty"

below_threshold, or required_empty for a required field left empty.

One of the following:

"below_threshold"

"required_empty"

cell: optional object { row, column }

Where in a list the least confident value is.

row: number

Index of the row, from 0.

column: optional string

Property of the row, when rows are objects.

confidence: optional number

The field’s confidence; for a list, its least confident value.

job_id: string

The extract job reviewed.

open_flags: array of string

JSON Pointers of flagged fields nobody checked.

original_result: map[unknown]

The result as the model read it.

required_fields: array of string

JSON Pointers of fields flagged when empty.

retain_until: string

When the job and its document may be deleted.

formatdate-time

reviewed_result: map[unknown]

The result with the field actions.

revision: number

Bumped by every change; actions name the one they saw.

source_file_id: string

The file the job read, or the file behind the parse job it read; null for a URL.

status: "approved" or "pending" or "rejected"

pending until approved or rejected.

One of the following:

"approved"

"pending"

"rejected"

updated_at: string

When the review last changed.

formatdate-time

ExtractReviewEvent object { id, actor_id, created_at, 5 more }

One entry in a review’s history.

id: string

Unique identifier of the entry.

actor_id: string

ID of the user who acted.

created_at: string

When it happened.

formatdate-time

kind: "approved" or "created" or "field_cleared" or 7 more

What happened.

One of the following:

"approved"

"created"

"field_cleared"

"field_confirmed"

"field_corrected"

"field_restored"

"field_undone"

"flags_signed_off"

"rejected"

"reopened"

detail: optional object { confidence_threshold, fields_in_scope, origin, required_fields } or object { reason } or object { paths }

The settings for created, the reason for rejected, and the fields for flags_signed_off; null for the rest.

One of the following:

ExtractReviewCreatedDetail object { confidence_threshold, fields_in_scope, origin, required_fields }

What a created entry records: where review was turned on, and its settings.

confidence_threshold: number

Fields read below this confidence are flagged.

fields_in_scope: array of string

JSON Pointers of the fields the threshold applies to; empty means all.

origin: "api" or "edit" or "history" or "run"

Turned on as the job ran, from History, by a first edit, or by the API.

One of the following:

"api"

"edit"

"history"

"run"

required_fields: array of string

JSON Pointers of fields flagged when empty.

ExtractReviewRejectedDetail object { reason }

What a rejected entry records.

reason: string

Why the result was rejected.

ExtractReviewSignedOffDetail object { paths }

What a flags_signed_off entry records.

paths: array of string

JSON Pointers of the flagged fields confirmed as part of approving.

path: optional string

JSON Pointer of the field it changed.

value_after: optional unknown

The field’s value after the change.

value_before: optional unknown

The field’s value before the change.