A Transcription Quality Assurance (QA) task is a QA task in which a configured percentage of transcribed fields or table cells is reviewed by a human. It helps measure and improve the accuracy of both machine and human transcriptions.
Before this task: The document’s field and table-cell values have already been transcribed, any required Supervision has been completed, and the submission has finished processing.
Why this task was generated: Transcription QA is enabled for the flow, and the QA sampling settings selected completed field or table-cell transcriptions for review. The task is an accuracy check; it is not generated because the system is uncertain about a value.
How the result is used: Your independent transcription creates a QA record. The system compares it with other transcriptions to measure accuracy and, when necessary, reach consensus. The task does not correct or reprocess the completed submission’s output.
The task opens as either Field Transcription Quality Assurance or Table Transcription Quality Assurance and uses the same editing controls as Transcription. For Semi-structured fields, Field Identification QA can affect which fields reach this task and how reliable Field Output Accuracy is.
Configure Transcription QA
Configure Transcription QA in the flow's Document Processing Subflow settings. Field and table sampling are controlled separately.
Setting | What it controls |
|---|---|
Field Transcription Quality Assurance | Enables QA sampling for transcribed fields. |
Automatic QA Sample Rate | Lets the system calculate the sample rates for supported field types instead of using the configured field rates. |
Structured Text Transcription QA Sample Rate | The percentage of Structured text fields sampled for QA. |
Structured Checkbox Transcription QA Sample Rate | The percentage of Structured checkbox fields sampled for QA. |
Structured Signature Transcription QA Sample Rate | The percentage of Structured signature fields sampled for QA. |
Semi-structured Transcription QA Sample Rate | The percentage of Semi-structured fields sampled for QA. |
Table Transcription Quality Assurance | Enables QA sampling for transcribed table cells. |
Table Transcription QA Sample Rate | The percentage of table cells sampled for QA. |
For the full setting descriptions and defaults, see Document Processing Subflow Settings.
Start a Transcription QA task
Go to Tasks > Perform Tasks.
In the Transcription QA row, click Perform Tasks.
The task opens the field or table experience required by the sampled content.
Field Transcription Quality Assurance
Field Transcription QA samples individual fields from Structured and Semi-structured documents. The resulting work is grouped into a document-level task, so one task can contain multiple sampled fields from the same document.
Complete a Field Transcription QA task
Review the highlighted area in the document panel.
Enter the value shown on the page. For checkbox and signature fields, select the state shown in the task.
If the content cannot be read, select Mark Illegible. If the document explicitly contains an N/A or Not Available value, select Mark N/A.
Click Submit to move to the next sampled location or field.
On the confirmation screen, review the number of fields and submit the task.
If a field has multiple identified locations, submitting a value moves through each included location before advancing to the next field. The field's data type can also enforce formatting rules; use the validation-override action when the document value is correct but does not match the configured format.
Table Transcription Quality Assurance
Table Transcription QA samples individual table cells. A document-level task includes the sampled cells for its independent, parent, and child tables. Child tables are presented before their related parent tables.
Complete a Table Transcription QA task
Select a cell that requires review. Cells requiring review are marked in orange.
Compare the current value with the highlighted content in the document image. Blue highlights show approximate, machine-predicted locations; they are not review-status indicators.
Edit the value when needed. Data-type validations apply to table cells.
Continue until all sampled cells in the current table have been reviewed.
Click Submit. If the task contains another table, it opens next. Submitting the final table completes the task.
The right panel provides DOCUMENT DETAILS and ALL COLUMNS. You can also reorder columns by dragging their headers and use the row actions available for the current independent, parent, or child table.
How Identification QA affects Transcription QA
For Semi-structured documents, Field Identification QA and Field Transcription QA are configured independently. Field Identification QA is not required for Transcription QA, but enabling or disabling it changes when fields are sampled and how reliable the resulting Field Output Accuracy measurement is.
Configuration | Transcription QA behavior | Accuracy implication |
|---|---|---|
Field ID QA and Transcription QA enabled | Field ID QA runs first for documents selected by its sample rate. After Field ID QA is complete, fields whose original locations agree with the Field ID QA result can be sampled for Transcription QA. The effective Transcription QA sampling rate is adjusted relative to the Field ID QA sample rate. | Both identification and transcription can be verified, producing reliable Field Output Accuracy measurements. |
Field ID QA disabled; Transcription QA enabled | Semi-structured fields are sampled directly at the configured Transcription QA rate, using the machine-identified locations. | Field Output Accuracy can be measured, but it is not accurate because field identification was not independently verified. |
Field ID QA enabled; Transcription QA disabled | Field locations can be verified, but no transcription values are sampled for QA. | System Field Output Accuracy cannot be measured. |
Fields changed during Field Identification QA
If Field Identification QA determines that a field's original location is wrong, the field does not appear in a subsequent Transcription QA task. This includes moving an existing field location or adding a location that was missing from the original identification. The identification output is scored as incorrect, and the field's Transcription QA record is abandoned rather than sent to a data keyer.
If a document is not selected by the Field ID QA sample rate, its Semi-structured fields are not subsequently sampled for Transcription QA when Field ID QA is enabled.
Table Identification QA and Table Transcription QA
Table Identification QA does not gate Table Transcription QA. Table-cell Transcription QA is sampled when the submission completes and can be created in parallel with Table Identification QA. As a result, changes made during Table Identification QA are not reflected in the Table Transcription QA task. The table-cell task uses the cells and locations that existed when it was sampled.
For the scoring details, see Scoring Field Output Accuracy.
Handle an incorrect layout
If the assigned layout variation does not match the document, open DOCUMENT ACTIONS and select Mark Layout Incorrect. Confirming this action ends the QA task and stops submission processing without extracting data. The sampled field and cell QA records are abandoned, and the document is removed from the QA ground-truth set.
What happens after you submit
Your response is added to the QA record and compared with the other available transcriptions. If the responses do not yet establish consensus, the system can select the unresolved field or table cell for another QA response.
Transcription QA creates accuracy evidence. It does not replace the value in the completed submission, reprocess the document, or correct the document output.
How Transcription QA reaches consensus
Transcription QA reaches consensus when two valid responses have matching normalized transcriptions. Normalization applies the configured formatting rules before values are compared, so the displayed text can differ while the normalized values still match.
The responses available for consensus can include an eligible machine transcription, a transcription entered during Supervision, and one or more Transcription QA responses. If two existing responses already match, the system can establish consensus without generating another QA task.
Transcription Supervision Consensus is a different process. It is configured for a field and requires matching transcriptions during document processing. Transcription QA consensus is established after the completed submission has been sampled for an accuracy check.
How transcription accuracy is scored
After consensus is reached, the system compares each eligible machine or human transcription with the consensus value. Accuracy is scored for the complete normalized field or table-cell value, not for individual characters.
For example, if the source value is 123-45-6789 and a response is 123-45-678, the response is inaccurate even though most characters match.
Entry | Scoring result |
|---|---|
An eligible machine or human transcription matches consensus. | Accurate |
An eligible machine or human transcription does not match consensus. | Inaccurate |
A machine transcription required Supervision. | Not measured for Machine Accuracy. The human responses can still help establish consensus and measure Manual Accuracy. |
Abandoned fields and table cells
A sampled field or table cell is abandoned when the system cannot establish consensus within the allowed responses. This can also happen when repeated responses are invalid or when two reviewers mark the value as illegible.
Entries for an abandoned item are not included in accuracy reporting. Completed QA responses can still be included in data-keyer productivity reporting.
How QA records affect accuracy and automation
Transcription QA records provide the verified examples used to measure transcription performance:
Accuracy is the proportion of eligible transcriptions that match the consensus value.
Automation is the proportion of values accepted without human Supervision.
A higher target accuracy generally results in more values being sent for human review. A lower target accuracy generally allows more automation.
How Table Transcription Automation is calculated
Table Transcription Automation is calculated from the completed Transcription task, not from the later Table Transcription QA review. It is the percentage of completed table cells that were transcribed without human involvement.
Table Transcription Automation = machine-transcribed table cells ÷ total transcribed table cells × 100
For this calculation, a completed table cell is counted as:
Automated when the machine-transcribed value is accepted without keyer review.
Not automated when the cell is sent to a keyer for review, or when a keyer changes a cell that was not initially selected for review.
The report uses the stored transcription outcome—machine-only or human-involved—rather than the cell’s visual border. Table Transcription QA happens after processing and does not change how the completed Transcription task is classified for automation reporting. For reporting details, see Automation.
QA data does not change a live Transcription model immediately. The records become eligible for a later training cycle, and model behavior changes only after the applicable data requirements and training conditions are met.
For a complete explanation of confidence scores, static thresholds, target accuracy, training requirements, and how a deployed model decides whether to accept a value or send it for review, see Transcription Models Overview.
Keyboard shortcuts
Press F2 in a Field Transcription QA or Table Transcription QA task to open the shortcuts available for the current screen. If updated shortcuts are enabled for your account, the panel lets you switch between the current and legacy shortcut sets. Use the panel as the source of truth because available actions vary by task type and screen.
A Transcription Quality Assurance (QA) task is a QA task in which a configured percentage of transcribed fields or table cells is reviewed by a human. It helps measure and improve the accuracy of both machine and human transcriptions.
Before this task: The document’s field and table-cell values have already been transcribed, any required Supervision has been completed, and the submission has finished processing.
Why this task was generated: Transcription QA is enabled for the flow, and the QA sampling settings selected completed field or table-cell transcriptions for review. The task is an accuracy check; it is not generated because the system is uncertain about a value.
How the result is used: Your independent transcription creates a QA record. The system compares it with other transcriptions to measure accuracy and, when necessary, reach consensus. The task does not correct or reprocess the completed submission’s output.
The task opens as either Field Transcription Quality Assurance or Table Transcription Quality Assurance and uses the same editing controls as Transcription. For Semi-structured fields, Field Identification QA can affect which fields reach this task and how reliable Field Output Accuracy is.
Configure Transcription QA
Configure Transcription QA in the flow's Document Processing Subflow settings. Field and table sampling are controlled separately.
Setting | What it controls |
|---|---|
Field Transcription Quality Assurance | Enables QA sampling for transcribed fields. |
Automatic QA Sample Rate | Lets the system calculate the sample rates for supported field types instead of using the configured field rates. |
Structured Text Transcription QA Sample Rate | The percentage of Structured text fields sampled for QA. |
Structured Checkbox Transcription QA Sample Rate | The percentage of Structured checkbox fields sampled for QA. |
Structured Signature Transcription QA Sample Rate | The percentage of Structured signature fields sampled for QA. |
Semi-structured Transcription QA Sample Rate | The percentage of Semi-structured fields sampled for QA. |
Table Transcription Quality Assurance | Enables QA sampling for transcribed table cells. |
Table Transcription QA Sample Rate | The percentage of table cells sampled for QA. |
For the full setting descriptions and defaults, see Document Processing Subflow Settings.
Start a Transcription QA task
Go to Tasks > Perform Tasks.
In the Transcription QA row, click Perform Tasks.
The task opens the field or table experience required by the sampled content.
Field Transcription Quality Assurance
Field Transcription QA samples individual fields from Structured and Semi-structured documents. The resulting work is grouped into a document-level task, so one task can contain multiple sampled fields from the same document.
Complete a Field Transcription QA task
Review the highlighted area in the document panel.
Enter the value shown on the page. For checkbox and signature fields, select the state shown in the task.
If the content cannot be read, select Mark Illegible. If the document explicitly contains an N/A or Not Available value, select Mark N/A.
Click Submit to move to the next sampled location or field.
On the confirmation screen, review the number of fields and submit the task.
If a field has multiple identified locations, submitting a value moves through each included location before advancing to the next field. The field's data type can also enforce formatting rules; use the validation-override action when the document value is correct but does not match the configured format.
Table Transcription Quality Assurance
Table Transcription QA samples individual table cells. A document-level task includes the sampled cells for its independent, parent, and child tables. Child tables are presented before their related parent tables.
Complete a Table Transcription QA task
Select a cell that requires review. Cells requiring review are marked in orange.
Compare the current value with the highlighted content in the document image. Blue highlights show approximate, machine-predicted locations; they are not review-status indicators.
Edit the value when needed. Data-type validations apply to table cells.
Continue until all sampled cells in the current table have been reviewed.
Click Submit. If the task contains another table, it opens next. Submitting the final table completes the task.
The right panel provides DOCUMENT DETAILS and ALL COLUMNS. You can also reorder columns by dragging their headers and use the row actions available for the current independent, parent, or child table.
How Identification QA affects Transcription QA
For Semi-structured documents, Field Identification QA and Field Transcription QA are configured independently. Field Identification QA is not required for Transcription QA, but enabling or disabling it changes when fields are sampled and how reliable the resulting Field Output Accuracy measurement is.
Configuration | Transcription QA behavior | Accuracy implication |
|---|---|---|
Field ID QA and Transcription QA enabled | Field ID QA runs first for documents selected by its sample rate. After Field ID QA is complete, fields whose original locations agree with the Field ID QA result can be sampled for Transcription QA. The effective Transcription QA sampling rate is adjusted relative to the Field ID QA sample rate. | Both identification and transcription can be verified, producing reliable Field Output Accuracy measurements. |
Field ID QA disabled; Transcription QA enabled | Semi-structured fields are sampled directly at the configured Transcription QA rate, using the machine-identified locations. | Field Output Accuracy can be measured, but it is not accurate because field identification was not independently verified. |
Field ID QA enabled; Transcription QA disabled | Field locations can be verified, but no transcription values are sampled for QA. | System Field Output Accuracy cannot be measured. |
Fields changed during Field Identification QA
If Field Identification QA determines that a field's original location is wrong, the field does not appear in a subsequent Transcription QA task. This includes moving an existing field location or adding a location that was missing from the original identification. The identification output is scored as incorrect, and the field's Transcription QA record is abandoned rather than sent to a data keyer.
If a document is not selected by the Field ID QA sample rate, its Semi-structured fields are not subsequently sampled for Transcription QA when Field ID QA is enabled.
Table Identification QA and Table Transcription QA
Table Identification QA does not gate Table Transcription QA. Table-cell Transcription QA is sampled when the submission completes and can be created in parallel with Table Identification QA. As a result, changes made during Table Identification QA are not reflected in the Table Transcription QA task. The table-cell task uses the cells and locations that existed when it was sampled.
For the scoring details, see Scoring Field Output Accuracy.
Handle an incorrect layout
If the assigned layout variation does not match the document, open DOCUMENT ACTIONS and select Mark Layout Incorrect. Confirming this action ends the QA task and stops submission processing without extracting data. The sampled field and cell QA records are abandoned, and the document is removed from the QA ground-truth set.
What happens after you submit
Your response is added to the QA record and compared with the other available transcriptions. If the responses do not yet establish consensus, the system can select the unresolved field or table cell for another QA response.
Transcription QA creates accuracy evidence. It does not replace the value in the completed submission, reprocess the document, or correct the document output.
How Transcription QA reaches consensus
Transcription QA reaches consensus when two valid responses have matching normalized transcriptions. Normalization applies the configured formatting rules before values are compared, so the displayed text can differ while the normalized values still match.
The responses available for consensus can include an eligible machine transcription, a transcription entered during Supervision, and one or more Transcription QA responses. If two existing responses already match, the system can establish consensus without generating another QA task.
Transcription Supervision Consensus is a different process. It is configured for a field and requires matching transcriptions during document processing. Transcription QA consensus is established after the completed submission has been sampled for an accuracy check.
How transcription accuracy is scored
After consensus is reached, the system compares each eligible machine or human transcription with the consensus value. Accuracy is scored for the complete normalized field or table-cell value, not for individual characters.
For example, if the source value is 123-45-6789 and a response is 123-45-678, the response is inaccurate even though most characters match.
Entry | Scoring result |
|---|---|
An eligible machine or human transcription matches consensus. | Accurate |
An eligible machine or human transcription does not match consensus. | Inaccurate |
A machine transcription required Supervision. | Not measured for Machine Accuracy. The human responses can still help establish consensus and measure Manual Accuracy. |
Abandoned fields and table cells
A sampled field or table cell is abandoned when the system cannot establish consensus within the allowed responses. This can also happen when repeated responses are invalid or when two reviewers mark the value as illegible.
Entries for an abandoned item are not included in accuracy reporting. Completed QA responses can still be included in data-keyer productivity reporting.
How QA records affect accuracy and automation
Transcription QA records provide the verified examples used to measure transcription performance:
Accuracy is the proportion of eligible transcriptions that match the consensus value.
Automation is the proportion of values accepted without human Supervision.
A higher target accuracy generally results in more values being sent for human review. A lower target accuracy generally allows more automation.
How Table Transcription Automation is calculated
Table Transcription Automation is calculated from the completed Transcription task, not from the later Table Transcription QA review. It is the percentage of completed table cells that were transcribed without human involvement.
Table Transcription Automation = machine-transcribed table cells ÷ total transcribed table cells × 100
For this calculation, a completed table cell is counted as:
Automated when the machine-transcribed value is accepted without keyer review.
Not automated when the cell is sent to a keyer for review, or when a keyer changes a cell that was not initially selected for review.
The report uses the stored transcription outcome—machine-only or human-involved—rather than the cell’s visual border. Table Transcription QA happens after processing and does not change how the completed Transcription task is classified for automation reporting. For reporting details, see Automation.
QA data does not change a live Transcription model immediately. The records become eligible for a later training cycle, and model behavior changes only after the applicable data requirements and training conditions are met.
For a complete explanation of confidence scores, static thresholds, target accuracy, training requirements, and how a deployed model decides whether to accept a value or send it for review, see Transcription Models Overview.
Keyboard shortcuts
Press F2 in a Field Transcription QA or Table Transcription QA task to open the shortcuts available for the current screen. If updated shortcuts are enabled for your account, the panel lets you switch between the current and legacy shortcut sets. Use the panel as the source of truth because available actions vary by task type and screen.