Field Identification Quality Assurance (QA) is a manual quality check for Semi-structured documents in which you review and correct the system’s predicted field locations or a human’s input. The review helps improve how accurately the model finds and extracts information from future documents, and the human input is used to measure human performance for reporting.
Before this task: The document was matched to a Semi-structured layout. During identification, the system or a human placed bounding boxes over the values for the layout’s fields. The document then completed processing and became eligible for QA sampling.
Why this task was generated: Field Identification QA is enabled for the flow, and the Field Identification QA Sample Rate selected this completed document for review. The task is a sampled accuracy check; it is not generated because a specific field is below a confidence threshold.
How the result is used: Your corrected field locations create a QA record used to measure identification machine and human performance. If the flow also uses Transcription QA, changes made here will determine which fields proceed to that task and will affect Transcription Field Output Accuracy.
In the task, correct, add, or remove bounding boxes and occurrences so every reviewed field covers the intended content without missing or extra text.
Start a Field Identification QA task
Go to Tasks > Perform Tasks.
In the Quality Assurance Tasks section, find Identification QA, and then click Perform Tasks.
Review the task guidance, and then continue to the task. You can choose not to show the guidance again.
Before you begin, review Best Practices for Field ID Supervision and QA.
Field Identification QA interface
The task interface has three main areas:
Page thumbnails on the left let you move between document pages.
Document view in the center shows the active page and the field bounding boxes.
All Fields on the right lists the fields that must be reviewed. A warning icon identifies an unreviewed field; a checkmark identifies a reviewed field. Document details and the completion action also appear in this panel.
Review field locations
Select a field in the right panel. Its bounding box is highlighted in the document.
Confirm that the box captures the complete field value without including its label.
If the location is incorrect, draw a box around the correct value. If the field is missing, clear its bounding box.
Continue until every field shows as reviewed.
If... | Then... |
|---|---|
The field location is correct. | Continue to the next field. |
The field location is incorrect. | Draw a bounding box around the correct value. |
The field was identified as not present, but it appears in the document. | Draw a bounding box around the field value. |
The field was identified as present, but it does not appear in the document. | Clear the field’s bounding box. |
Available corrections
You can add or remove field occurrences and add, remove, redraw, or resize bounding boxes during Field Identification QA.
Review and add Multiple Occurrences
When a field contains multiple occurrences, expand its occurrence group in the right panel, and select each occurrence to review its bounding box.
Remove an occurrence
Use the Delete icon beside an additional occurrence to remove it. By default, the next occurrence becomes active and the document may move to the page containing it.
To remove an occurrence and remain on the current page:
Hold Shift while clicking the Delete icon.
When the updated keyboard shortcuts are enabled, press Option + Shift + X on Mac or Alt + Shift + X on Windows.
Effect on Transcription QA
For Semi-structured documents, Field Identification QA and Transcription QA can be enabled independently. When both are enabled, Field Identification QA runs first for documents selected by its sample rate. After Field Identification QA is complete, the fields can be sampled for Transcription QA.
If you move a field location or add a location that was missing from the original identification, that field does not appear in a subsequent Transcription QA task. Its identification output is scored as incorrect, and its Transcription QA record is abandoned.
If Field Identification QA is disabled, Semi-structured fields can still be sampled for Transcription QA. However, Field Output Accuracy is not accurate because the field locations were not independently verified.
If Field Identification QA is enabled but a document is not selected by its sample rate, fields in that document are not subsequently sampled for Transcription QA.
For the full dependency and sampling behavior, see Transcription Quality Assurance and Scoring Field Output Accuracy.
Handle an incorrect layout
If the document was matched to the wrong layout:
Expand Document details in the right panel.
Click Mark Layout Variation Incorrect.
Confirm the action.
The QA task ends, and no fields in the document are scored for accuracy. The document is not automatically marked as No Layout Found, and this QA action does not send it for reprocessing.
Complete the task
After you have reviewed every field, click Confirm All and Continue. If any fields remain unreviewed, the Fields Need Review message returns you to the task.
Keyboard shortcuts
Press F2 in the task to open the keyboard shortcuts panel. The panel shows the shortcuts that apply to Field Identification QA and your operating system.
The updated shortcuts are optional in v43.2 and can be enabled or disabled from the keyboard shortcuts panel.