User Experience
Tooltips when moving the pointer from one text box to another – When a user hovers over a text box, a tooltip is shown, but when the user hovers over another text box, a tooltip for that text box is not shown. This issue affects various portions of the application, but it primarily affects the Layout Editor for Structured layouts.
Using calendars in date filters – Using the drop-down menus in calendars to select different months sometimes causes the calendars to automatically close.
Workaround: To select different months, use the Previous (
) and Next (
) buttons that are located next to the drop-down menus.
Layouts and Layout Variations
Multiple live versions of a layout variation – The system allows you to deploy multiple versions of a layout variation across different flows, as it does not perform any validation in this area when assigning a release to a flow. However, since we only train a model on one of the multiple live versions, it's possible that some fields or columns that exist in one of the layouts but not the others could be skipped in the trained model.
Workaround: When creating and assigning a release, make sure that other versions of the release's layout variations have not already been assigned to other live flows.
Focus at the top of Add Layout dialog – The system places the focus at the top of the Add Layout dialog rather than on the Layout Name field in that dialog.
"Multiline" option for checkboxes – In the Layout Editor for Structured layouts, the Multiline option is shown as being available for checkboxes. However, this option does not apply to checkbox fields.
Notifications when importing multiple layouts – When a user uploads multiple layouts, the system does not include the layout names in the notifications ("Import of undefined in progress," "Import of has finished").
Autoscrolling to the first page's fields in the Layout Editor – After a user clicks on the last page of a document in the Layout Editor's left-hand sidebar, if they click on the document's first page, the list of fields in the right-hand sidebar does not automatically scroll to show the first page's fields.
Setting Submission Processing Deadlines for layout variations – When setting a Custom Submission Processing Deadline for a Structured layout variation, the actual deadline for the variation is not updated.
[Fixed in 34.0.1] Text of "Read only" label in the Layout Editor – The "Read only" label in the Layout Editor has white text on a white background, making the text illegible.
Adding identical layout variations to the same layout – The system displays a 500 error page if a user attempts to add identical layout variations to the same layout.
Workaround: Make sure the name of each layout variation is unique to that layout.
Committing changes to layout variations – When a user edits a layout variation, the system sometimes creates empty fields that are shared across the layout's variations. The addition of these fields prevents the user from committing their changes.
Updating layout variation names after import – If a layout is reimported after the name of one of its variations is changed in the layout's original instance, the name of the variation is not updated after the import is complete.
Data Types
Creating a data type from a list of values with no lines – If a user attempts to create a data type from a list of values that has no lines of data, the system shows a "Bad request" error message. The message should explain that a data type can't be created without at least one row of values.
Normalization
Normalized values in Oracle databases – In instances with Oracle databases, the system stores normalized values as empty strings instead of as null values. This behavior differs from the behavior in instances with PostgreSQL or MSSQL databases.
Normalization of fields with multiple bounding boxes – The system does not always normalize fields properly if their values are captured in multiple bounding boxes and Manual Transcription is disabled.
Releases
Focus at the top of Add Release dialog – The system places the focus at the top of the Add Release dialog rather than on the Release Name field in that dialog.
Copying a layout with new variation to a new release – If a user creates a release with a layout, adds a variation to the layout, and copies the layout's release to a new release, the new variation is not immediately shown in the Release Editor for the new release. The system only shows the new variation after deselecting the layout's original variation in the Release Editor.
[Fixed in 34.0.1] Release validation error when no release is available – The release validation error shown when no release is available shows (missing) null as the selected release.
Flows
Outdated URLs for tasks created in a previous version – After completing Manual Supervision tasks in an upgraded instance, submissions may halt because the system attempts to connect to outdated URLs.
Settings for unarchived flows – After a user unarchives a flow, the system does not show all of the available settings for that flow. If a user reloads the flow's page in Flow Studio, all of the flow's settings appear.
Archiving or unarchiving a flow while adding a connection – If a user attempts to archive or unarchive a flow while adding a connection to a flow, an "Unexpected error" message appears.
Input connection names in Flow Execution pages in IE 11 – The name of an input connection may overlap with the connection icon on the Flow Execution page in IE 11.
Loading of Flows page when many flows are present – The Flows page is slow to load in instances with more than 80 flows.
Importing flows with malformed JSON files – The error message shown when a user imports a flow with a malformed JSON states that the system could not open the flow's ZIP file. Because it does not indicate that the system could not parse the JSON file, the user does not know the exact cause of the error.
Metadata changes in imported flows – When a user imports a new version of a flow, the system may sometimes indicate that the metadata has changed in the imported version, even if no changes were made to the metadata.
Spacing between validation flow errors in Flow Studio – There is no spacing between errors in the list of validation errors shown in the Flow Studio, making the list difficult to read.
[Fixed in 34.0.1] Blocks for future tasks in Flow Execution diagrams – The background for future tasks in Flow Execution diagrams was gray in previous versions, but it is transparent in v34.
Saving invalid or incomplete connection settings – If a user attempts to add a connection to a disabled flow but clicks Save before entering complete and valid settings for that connection, the system does not save the connection settings. Instead, the system should save the settings and show validation errors.
Clicking on HTTP REST Blocks in Flow Studio – If a user clicks on an HTTP REST Block in the Flow Studio, the Save button is enabled, even though no changes were made.
Flow Execution diagrams for halted submissions – In some situations, if a submission is halted, all of its blocks appear as "Completed" in its Flow Execution diagram.
Expired sessions or session timeouts when importing flows – If a user's session expires or times out while they are importing a flow, the system shows a generic error message. Instead, it should notify the user of the expired session or redirect them to the login page.
Submissions Table and Documents Table
Displaying flow titles and restrictions – If a user does not have access to certain flows, the system does not display the names of those flows in the Submissions table or Documents table, nor does it indicate that those flows are restricted.
Task Queue
“Layout” filter on the Task Queue page – The Layout filter on the Task Queue page shows all layouts in the instance, even if they have been archived or are not live.
Classification
"Mark Layout Variation Incorrect" for Semi-structured documents – If a keyer clicks Mark Layout Variation Incorrect during Identification Supervision, the system shows a message saying that the submission's processing will end, followed by a 404 error page after the keyer completes the tasks. However, processing only ends for the document the task was generated for, and the keyer should be redirected to the Submission Details page after task completion.
[Fixed in 34.0.1] Memory usage when downloading Classification models and training data – When downloading a Classification model and its training data, the system may run out of available memory.
Identification
Annotating a document for a long time – If a user annotates a document for a long time and then submits the document, the user might be redirected to the Submissions table instead of to the next available task for this document.
Field Identification
Overlapping bounding boxes – During Field ID QA, if you select a bounding box and unintentionally drag your mouse over a different bounding box, these two bounding boxes will overlap.
Table Identification
One-click bounding boxes for the first row’s cell in a column – If a Table ID model predicts a few cells in a column but does not predict the first row’s cell in this column, you will not be able to properly select and create this cell’s bounding box. Hovering over this cell with your mouse does not allow creating one-click bounding boxes from all angles.
Row that is overlapping the “Extend row [row’s number] to the next page” button – If the last row on a page overlaps the Extend row [row’s number] to the next page button, the Extend row [row’s number] to the next page button is unclickable.
Clicking the “Delete Selected Row” link when a table column is selected – When you select a table column and right-click a row, clicking the Delete Selected Row link deletes only the column’s cell in this row.
Two rounds of Table ID QA to reach consensus – If you:
add an extra row to your table during Table Identification, and
submit the document, and
complete Table ID QA without making any changes,
then, consensus is not reached. To reach consensus, you need to complete a second round of Table ID QA.
Thresholding for rare columns – Due to thresholding issues, when a table column appears infrequently in a set of Semi-structured documents, the system predicts the wrong locations for all its rows.
Table Identification for nested tables
Child and parent cross-page rows in nested tables – If a child row spans over multiple pages and is also part of a parent row that spans over multiple pages, the machine identifies the child row twice. The child row is both:
correctly identified as part of its parent row, and
incorrectly identified as part of the next parent row in the table.
Tables with more than 99 child rows – When a table has more than 99 child rows, the table’s scrollbar hides a part of the three-digit relationship numbers that are located on the right-hand side of the table.
Automatically identifying new cells for parent columns – The .png?sv=2022-11-02&spr=https&st=2025-05-12T03%3A35%3A29Z&se=2025-05-12T04%3A03%3A29Z&sr=c&sp=r&sig=zdTgmiJ6wUWBjqgNkqtIs%2FqcBe7yKKdqjIyrzNUIIWE%3D)
button that automatically identifies new cells for a column can sometimes identify incorrect cells for parent columns.
Automatically identifying child rows – The machine cannot automatically identify child rows that are located too far from the last automatically identified child row.
Annotated cells outside of a parent row and row’s height – A parent row’s height does not change based on the annotated cells in the row. For example, if you annotate cells that are outside of a parent row, the parent row won’t automatically adjust its height.
Identifying a template row for a parent table – Once you’ve identified a template row for a parent table, you can’t re-identify a new template row.
Deleting parent rows – You can delete a parent row only if no column is selected.
Deleting a cell during Table ID Supervision and then transcribing this cell during Table Transcription – If you:
delete the first cell of a child table’s first column, and
delete the first cell of a parent table’s first column, and
edit the transcription value of the first cell of a child table’s first column during Transcription Supervision, and
edit the transcription value of the first cell of a parent table’s first column during Transcription Supervision,
then, these transcription values will shift from the first column to the second column in both tables.
“Next column” keyboard shortcut during template-row identification – For nested tables, pressing the E key during template-row identification navigates you to the previous column instead of to the next one.
Annotating parent rows with different height – If the height of the parent rows is different, the machine might predict some of the bounding boxes’ locations incorrectly.
Table Identification training with nested tables – Table Identification training jobs fail if they contain nested tables with any of the following characteristics:
overlapping child columns
orphan child cell
orphan parent cell
Transcription
[Fixed in 34.0.1] “None” normalized transcription values and the Semi-structured Transcription Confidence Boost setting – The Semi-structured Transcription Confidence Boost setting cannot handle “None” normalized transcription values.
Scheduling recalibration, auto-thresholding, and finetuning – Scheduling recalibration, auto-thresholding, and finetuning at /admin/forms_qa/recalibrationrecord/run-recalibration-at-ft/ results in server errors for:
disabled flows, or
flows that have their Transcription Automation Training setting disabled.
Transcribing fields with multiple bounding boxes – When transcribing fields with multiple bounding boxes, pressing the Shift + Backspace keyboard shortcut navigates you to the previous field and not to the previous bounding box of the field you transcribe.
Pasting text when only numeric values are accepted – If you click a table cell in the transcription work area, you can paste text even if only numeric values are accepted for the cell’s table column. The text is pasted without validation.
Copying and pasting a cell’s transcription value – If you copy a cell’s transcription value and press Enter on another cell before pasting the copied text, you will actually paste a different text from the one you’ve copied.
Workaround: Paste the copied text without pressing Enter.
Transcription of table cells with lower confidence than Table Minimum Legibility Threshold’s value – Table cells with transcription confidence that is lower than the Table Minimum Legibility Threshold’s value do not have any exceptions. Instead, such table cells should have Illegible Field exceptions.
Deleting a row during Table Transcription when no columns are selected – When no columns are selected during Table Transcription, the Delete a Row link is still active instead of being inactive. Clicking the Delete a Row link results in an “Unexpected error” message.
“Submit” button during Table Transcription QA – During Table Transcription QA, if a table cell is located on the right-hand side of a page, the Submit button for this cell might be fully or partially hidden.
Line splitting in multiline fields – Due to line-splitting errors, the system does not always transcribe multiline fields correctly, and it may not recognize certain characters in those fields.
Dropout text included in multiline field transcriptions – If the Dropout option is enabled for a multiline field in a Structured layout, the system may still include surrounding text in the field's raw transcriptions.
Custom Supervision
Selecting a new cell in the table editor during Table Transcription in Custom Supervision – During Table Transcription in Custom Supervision, the whole Transcription table re-renders when you select a new cell in the table editor.
Rendering of the right panel – The whole right panel re-renders when you:
click any field to make it active.
edit a field, select a decision, add a case, etc.
select cells in the table editor.
Styling of empty cells with decisions in IE 11 – In the table editor, empty cells with decisions do not have appropriate styling as opposed to cells with values which are underlined.
Selecting table cells in the table editor – A single click with the mouse over a table cell in the table editor highlights the table cell, while a double click with the mouse allows you to edit the cell’s transcription value.
Editing and saving table cell transcription values – If your Custom Supervision Block’s Custom Supervision transcription masking setting is enabled, to save a change to a cell’s transcription value, you need to press Enter or click anywhere outside of the cell’s bounding box. If you only edit a cell’s transcription value and immediately after that click the Complete Task button at the bottom of the right panel, your edit won’t be saved.
Focusing on pages in tall browser resolutions – When completing Custom Supervision tasks in browsers whose resolutions are taller than they are wide, clicking on a crop in a different page does not always shift the focus to that page.
[Fixed in 34.0.1] Table columns that are not identified during Machine or Manual Identification tasks – If the Custom Supervision’s template references a table column that was not identified during Machine or Manual Identification tasks, this table column will still be displayed during Custom Supervision. When a user selects a cell in this column, the table editor automatically closes.
Quality Assurance
Field ID QA sampling for resubmitted submissions – If submissions are originally submitted to a flow with a Field Identification QA Sample Rate of 0%, and the submissions are resubmitted to a flow with a Field Identification QA Sample Rate greater than 0%, the system still does not generate Field ID QA tasks for the submissions.
Clearing QA tasks with the All Flows filter selected – If the All Flows filter is selected on the Tasks page when QA tasks are cleared, the system clears all unassigned QA tasks from the queue. However, the system should only clear the tasks for the flows that the user has access to.
Classification QA tasks for pages marked for deletion – When the system generates a Classification QA task, it does not check if its associated image is marked for deletion. As a result, some tasks may be associated with images that no longer exist.
Keyer Data Management
Deleting a parent row that contains a child row – Deleting a parent row that contains at least one child row and saving your changes in the Review Table Column Annotations section results in an error message.
Long segments of text in the right-hand sidebar – If you add long segments of text to a table column’s Notes (optional) setting in the Layout Editor, the right-hand sidebar’s size in the Review Table Column Annotations section changes depending on the text’s length.
Deleting a row – Deleting a row in the Review Table Column Annotations section is only possible if no column is selected.
Paging through available documents – Pressing Option + ➡ for Mac or Alt + ➡ for Windows does not page through all available documents.
Model Validation Tasks
“Delete Selected Row” link – Right-clicking a table row and then pressing the Esc key changes the displayed link’s name from “Delete Selected Row” to “Delete Selected Rows”.
Extra Model Validation Tasks (MVTs) for nested tables – The system does not validate if all table columns in a nested table should go through MVTs. As a result, a few extra MVTs might be created for nested tables, even though columns from these MVTs have already been validated.
Submission Output
[Fixed in 34.0.1] Rejected documents in the Submission Output page – Documents that are rejected during Transcription Supervision are accessible in the Submission Output page.
Document Output
Cells without a value and table output – If you click a cell without a value in the table output, the table output closes automatically.
Scrolling the table output – Scrolling the table output with the mouse is not possible.
Workaround: Scroll the table output with keyboard shortcuts.
Connections
Recording last-processed times in Eastern Standard Time in Salesforce – If the Salesforce Listener Block runs on a machine configured to use Eastern Standard Time, the block records incorrect last-processed times. Specifically, the time stored in the Salesforce record is 4 hours ahead of the current UTC time.
Infinite submissions created by the Email Listener Block – If the folder specified in Post Process Archive Folder does not exist, the Email Listener Block creates an infinite number of submissions. The system marks these submissions as "Completed," which can make the issue difficult to detect.
Infinite submissions created by the Box Folder Listener Block – If the folder specified in Folder to Move Completed Files does not exist, the Box Folder Listener Block creates an infinite number of submissions.
Health Check processes for Box Folder Listener Block – The Health Check API processes for the Box Folder Listener Block can be halted indefinitely in some circumstances, as there is no timeout configured for these processes.
Jobs
"Search by ID or other metadata" search box – The text in the Search by ID or other metadata search box is misleading, as you can only search for jobs or flows by Submission UUID.
Training
Training a GENERIC_FREEFORM_GT model with fields with multiple occurrences – Documents that failed training a GENERIC_FREEFORM_GT model due to having fields with multiple occurrences are not listed on the Model Details page.
Status for training a future model – When you run a training for a future model, the training will have a “Training in Progress” status instead of “In Training Queue”.
Starting a model training job without a trainer – If you start a model training job without a deployed trainer, you will receive a message that training has started, even though the training is only queued and cannot be completed without a trainer.
Viewing the Trainer page when there are no trainers – If there are no trainers in the instance and a user goes to the Trainer page (Administration > Trainer), the application displays an error page.
Checking trainer availability in SaaS deployments – In SaaS deployments, the trainer automatically turns off when not in use. However, the pages for Classification models and finding potential layouts may indicate that there is no trainer connected to the application when the trainer is actually present but turned off.
Reporting
Machine errors and System Transcription Sampled Errors report – The System Transcription Sampled Errors report (Reporting > Accuracy) shows the machine as incorrect even if the machine’s guess was below the transcription threshold.
Filters for Manual Accuracy vs Machine Accuracy report – If the Classification filter is selected for the Manual Accuracy vs Machine Accuracy report (Reporting > Accuracy), the All Field Types filter is still present, even though it doesn't apply to Classification data.
API
Not providing flow_uuid when creating a submission via API – If no flow_uuid value is provided when creating submissions via the API, an error may occur if the "Document Processing" flow for the current version is not live.
Workaround: Always provide a flow_uuid value when creating a submission via the API.