adm_my_form_extractions_v

One row per form extraction result of a document the ADM user established in the current session may see. Needs the AI Pack.

Overview

Group: Application integration

Category: Form extraction - Data read from forms by the AI Pack form extraction

Use Cases

A review page for extraction results in your own application
Importing accepted invoice values into another system
Listing the documents that wait for review

Columns

The view contains 20 columns providing comprehensive information:

Column Name	Description
`extraction_id`	Primary key of the extraction result. The unique key of this view.
`profile_name`	Name of the form extraction profile that extracted the document.
`document_id`	The extracted document. Joins to adm_my_documents_v.document_id.
`document_name`	Name of the document.
`version_id`	The version that was extracted. Joins to adm_my_document_versions_v.version_id.
`form_code`	Code of the form that was recognised. NULL when no form was recognised, or when the form was deleted later.
`doc_type`	Code of the document type of the form, for example invoice. NULL when form_code is NULL.
`status`	OK: every check passed. REVIEW: a person must look at the document. FAILED: an error stopped the extraction. APPROVED: a person approved the result.
`method`	TEXT: extracted from the text of the document. IMAGE: extracted from the image by a vision model, which also picked the form. FILE: the markers found the form in the text, and the vision model read the values from the file. NONE: not extracted.
`text_source`	Where the text that the markers ran on came from; for method TEXT also the text that the model read. FILTER: the built-in text extraction. OCR: the OCR text that ADM stored on the version. NULL when no text was read: an image, or a failure that the job recorded without values.
`fields`	The extracted fields, as a JSON object. Read a field with json_value(fields, ‘’$.total_amount’’).
`line_items`	The extracted line items, as a JSON array. Read them with json_table(line_items, ‘’$[*]’’ columns (…)).
`issues`	The checks that failed, one line for each issue, or the reason of a failure. NULL when every check passed.
`model`	The model that extracted the data. NULL when no model was called.
`prompt_tokens`	Input tokens of all model calls of the extraction. 0 when no model was called, NULL when the job recorded a failure without values.
`completion_tokens`	Output tokens of all model calls of the extraction. 0 when no model was called, NULL when the job recorded a failure without values.
`seconds`	Run time of the model calls of the extraction in seconds. For a failed extraction, the time of the whole extraction. NULL when the job recorded a failure without values.
`extracted_at`	Timestamp of the extraction.
`reviewed_by`	User who approved the result. NULL unless status is APPROVED.
`reviewed_at`	Timestamp when the result was approved. NULL unless status is APPROVED.

Key Features

Unique by extraction_id, and by (profile_name, document_id)
fields and line_items are JSON: read them with json_value and json_table
status OK or APPROVED marks a result that may be used
Token counts and run time of the extraction, and text_source: the text that the extraction read (FILTER or OCR)

Other views in the Form extraction category: