Table Identification Quality Assurance

Prev Next

Table Identification Quality Assurance (QA) is the process of reviewing and validating the system’s identification of tables within Semi-structured documents. The task allows you to measure the system’s accuracy in table identification.

Before this task: The document was matched to a Semi-structured layout. During table identification, the system or a human created the table structure, including its rows, columns, and cell locations. The document then completed processing and became eligible for QA sampling.

Why this task was generated: Table Identification QA is enabled for the flow, and the Table Identification QA Sample Rate selected this completed document for review. The task is a sampled accuracy check; it is not generated because a specific table or cell is below a confidence threshold.

How the result is used: Your corrected table annotations create a QA record used to measure table-identification accuracy and determine consensus. Table Identification QA is independent from Table Transcription QA, so these changes do not update a Table Transcription QA task that was sampled separately.

In the task, correct rows, columns, cell bounding boxes, and parent-child relationships so the table structure matches the source page. Table Identification QA and Field Identification QA are both available from the Identification QA task queue.

Configure Table Identification QA

Configure Table Identification QA separately for each flow in the flow’s Identification settings:

  1. Enable Table Identification Quality Assurance.

  2. Set Table Identification QA Sample Rate.

  3. Save the flow.

The sample rate is the percentage of eligible documents that the system randomly selects for Table Identification QA. For example, a 10% rate means that approximately 10 out of every 100 eligible documents are selected over time.

For the complete setting descriptions, see Document Processing Subflow Settings.

Start a Table Identification QA task

  1. Go to Tasks > Perform Tasks.

  2. In Quality Assurance Tasks, find Identification QA, and then click Perform Tasks.

  3. When a Table Identification QA task opens, review the table name, columns, and document pages before changing any annotations.

The Identification QA queue can also open Field Identification QA tasks. For that workflow, see Field Identification Quality Assurance.

Table Identification QA interface

The task interface shows document pages and table annotations in the main area. The right panel shows the current table, its columns, progress through multiple tables, Document details, and the action for continuing or submitting.

Dashed yellow borders indicate cells that may be incorrect. Column labels can also identify a Missing column, Misplaced column, or other anomalies. These indicators help prioritize review, but you must check every cell on every page.

Review table-cell locations

For each row and column, confirm that the bounding box contains the complete cell value.

Prediction state

Action

All cells are correct.

Review every page, and then continue or submit the table.

Some cells are missing or incorrect.

Resize or redraw incorrect boxes and identify missing cells.

No cells are predicted.

Identify one representative template row, click Continue to Review, and then check the generated rows and cells.

You can resize cells, draw bounding boxes, split rows, insert or delete rows, find missing cells, and select or delete a column’s cells. For detailed instructions, see Review table cells and Manage rows and cells.

If the table needs to be regenerated from another template row, click Manually Re-identify Table.

Review multiple and nested tables

For documents with multiple tables, review each table in the order shown by the progress indicator. Click Continue to next table until you reach the final table.

For a nested table, review the child table first and then the parent table. Make sure every child row has a parent row. After all related tables reach consensus, the document reaches Table Identification consensus.

Resolve completion warnings

Before submitting, resolve any Duplicate cells, Empty Row, or Orphan Rows message. If Review All Pages appears, return to the pages you have not reviewed or confirm that every page has been checked.

Effect on Table Transcription QA

Table Identification QA and Table Transcription QA are sampled independently. Table Transcription QA does not wait for Table Identification QA to finish; both tasks can be created in parallel when the submission completes.

Changes made during Table Identification QA are therefore not reflected in the Table Transcription QA task. The Table Transcription QA task uses the table cells and locations that existed when it was sampled.

For more information, see Transcription Quality Assurance.

Handle an incorrect layout

If the document was matched to the wrong layout:

  1. Expand Document details in the right panel.

  2. Click Mark Layout Variation Incorrect.

  3. In Confirm Mark Layout Variation Incorrect, click Yes, Mark Incorrect.

The QA task ends and the next available Identification QA task opens. The document is abandoned from Table Identification QA: its table cells are not scored for Table Identification accuracy and are not retained as ground truth for model training. QA does not send the document for reprocessing.

Complete the task

After reviewing the current table, click Continue to next table if another table remains. On the final table, click Submit Document. Resolve any completion warning before submitting.

Keyboard shortcuts

Press F2 to open the keyboard shortcuts panel. The panel shows the shortcuts for the current QA step and operating system.

The updated shortcuts are optional in v43.2 and can be enabled or disabled from the keyboard shortcuts panel. Use the panel as the source of truth instead of relying on a static shortcut list.