V34 Release Notes

34.0.16 (20 Jun 2023)

Settings

Fixed

Functionality and description of Submission Upload setting – We've fixed an issue that caused the Create Submission button to be enabled when the Submission Upload setting was disabled in the System Settings (Administration > System Settings). Also, we've updated the description of the Submission Upload setting with the correct name of the Create Submission button.

Submissions

Updated

Maximum number of files per upload – We've increased the default maximum number of files per upload from 100 to 1000. This value can be customized with the DATA_UPLOAD_MAX_NUMBER_FILES ".env" file variable.

Documents

Fixed

Viewing deleted documents – We've resolved an issue that caused errors to occur when users attempted to view images of documents that had been deleted (e.g., documents that were deleted during PII-data deletion).

Flows

Fixed

Settings for unarchived flows – We've fixed an issue that prevented the system from showing all of the available settings for a flow after the flow was unarchived.

Archiving or unarchiving a flow while adding a connection – We've resolved an issue that caused an "Unexpected error" message to appear when a user attempted to archive or unarchive a flow while adding a connection to it.

Message Queue Listeners

Updated

Handling of NullPointerException exceptions and non-2xx HTTP errors – We've made changes to ensure that NullPointerException exceptions and non-2xx HTTP errors are managed in ways that are transparent to the user and do not disrupt the user experience. Specifically, the system manages non-2xx HTTP errors as exceptions, and if one of these errors occurs or a NullPointerException is thrown, the process of handling the message is stopped.

Salesforce Listener

Updated

Timeouts and retries – We've increased the timeout from 10 seconds to 20 seconds and added a retry mechanism for lost connections to the Salesforce server. The system reattempts the connection up to three times with increasing wait times between attempts.

Reporting

Fixed

Data for Semi-structured documents in Usage Reports – We've resolved an issue that prevented data for Semi-structured documents from appearing in the following Usage Report files:

checkbox_machine_transcriptions_report.csv
supervision_transcriptions_report.csv
machine_transcriptions_report.csv

Downloading Usage Reports after completing QA tasks for Checkboxes or Signatures – We've fixed an issue that prevented Usage Reports from being downloaded after a QA task was completed for a Checkbox or Signature field.

Security

Fixed

Addressing security vulnerabilities – To ensure security, we've updated:

sqlparse to 0.4.4,
requests to 2.31.0,
com.fasterxml.jackson.core:jackson-databind to 2.14.2,
ddtrace to 1.7.2, and
the version of Golang used to compile Filebeat to 1.20.4.

Upgrades

New

Monitoring the progress of bigint migrations – The migrate_to_bigint_with_monitoring.sh script allows you to monitor the status of bigint migrations, including the total number of queries in the migration, the number of queries that have been completed, and which queries have been completed and which are currently running.

SaaS

Fixed

User-Agent HTTP headers in Client Library authentication requests – To address security concerns, the system now requires User-Agent HTTP headers in all Client Library authentication requests.

API

New

Swagger UI and schema endpoints – We've enabled the following Swagger endpoints:

/api/v5/swagger – Displays information about Hyperscience API endpoints that currently have OpenAPI specifications
/api/v5/schema – Downloads the OpenAPI specifications for the instance's version of Hyperscience
/api/v5/schema.json – Provides the same data as /api/v5/schema, but in JSON format

Fixed

Checking files in Submission Creation requests – We've fixed an issue where sending a Submission Creation request with an empty files parameter caused a 500 (Internal Server) error to be thrown rather than a 400 (Bad Request) error. The updated response also includes more details on what caused the error.

34.0.15 (6 Mar 2023)

Field Identification Models

Fixed

Training with documents containing fields with multiple occurrences or multiple bounding boxes – We've resolved an issue that caused training to fail if the training set contained documents with fields that had multiple occurrences, multiple bounding boxes, or both.

Input Connections

Fixed

Retrying submissions – We've fixed a flow-run ID issue that sometimes caused images from submissions ingested through folder-based connections to be deleted prior to processing. The issue prevented affected submissions from being resubmitted.

Classification Supervision

Fixed

"Perform Tasks" link in Submissions table – We've fixed an issue that prevented the Perform Tasks link from appearing in the Submissions table for submissions with Classification Supervision tasks. The issue affected submissions whose first page was classified by the machine.

Security

Fixed

Updating certifi and oauthlib – To address security vulnerabilities, we've updated certifi to 2022.12.7 and oauthlib to 3.2.2.

34.0.14 (20 Jan 2023)

Submission Processing

Updated

Size of submission metadata – To reduce the size of submission metadata in the database, we've eliminated unnecessary whitespace in the submission metadata JSON.

Reporting

Fixed

Records of tasks performed on Table Identification training data – We've fixed a task-reporting issue that caused the system to create duplicate records for user tasks that were performed on Table Identification training data.

PII Data Deletion

Fixed

Deleting PII data from large submissions in instances with PostgreSQL databases – We've fixed an issue that caused database CPU usage to spike when PII data was deleted from large submissions. The issue affected instances with PostgreSQL databases.

API

New

Swagger UI and schema endpoints – We've enabled the following Swagger endpoints:

/api/v5/swagger – Displays information about Hyperscience API endpoints that currently have OpenAPI specifications
/api/v5/schema – Downloads the OpenAPI specifications for the instance's version of Hyperscience
/api/v5/schema.json – Provides the same data as /api/v5/schema, but in JSON format

34.0.13 (22 Dec 2022)

V34 is not recommended for nested tables. V35 will contain several enhancements and fixes for nested tables. To learn more about known issues and limitations in v34, see Known Issues and Limitations in v34.

Classification

Fixed

Generating Manual Classification tasks – We’ve fixed an issue that prevented Manual Classification tasks from being generated.

Connections

Fixed

Sending a large number of objects through Message Queue Listener connections – We’ve fixed an issue with sending a large number of objects through Message Queue Listener connections. This issue caused timeouts to occur and prevented submissions from being processed.

Exported Data

Fixed

Redacting restricted information from CSV exports– We've removed restricted information from the exports of page-, field-, and table-level data, both in the application and in our API endpoints. Examples of restricted information include the UUIDs of submissions and flows that the user generating the exports does not have access to.

Quality Assurance

Fixed

Clearing QA tasks on instances that use MSSQL databases – We’ve fixed an issue that prevented users from clearing more than 2000 QA tasks at the same time on instances that use MSSQL databases.

Model Validation Tasks

Fixed

Displaying multiline bounding boxes – We’ve fixed an issue that caused multiline bounding boxes to be displayed as separate segments during Model Validation Tasks.

Reporting

Fixed

Downloading the Usage report – We’ve fixed an issue that prevented the Usage report (Reporting > Usage) from populating data for the current date. This issue occurred when downloading the report with one of the following predefined date ranges:

Week to Date
Month to Date
Year to Date

Upgrades

Fixed

Migration of retry_submission_id values in Oracle databases– We've fixed a data-type issue that caused the migration of retry_submission_id values to fail when upgrading instances with Oracle databases from v32.

SaaS

Fixed

Access to the Administration page for System Admins – We’ve fixed an issue with SaaS instances that caused a 403 error to occur when System Admins attempted to open the Administration page.

API

Fixed

Submissions created with Content-Type: application/json; charset=utf-8– We've fixed an issue that caused internal server errors to occur when submissions with Content-Type: application/json; charset=utf-8 were created via the API.

Security

Fixed

Updating wheel and cryptography – To address security vulnerabilities, we've updated wheel to 0.38.4. and cryptography to 38.0.3

34.0.12 (10 Nov 2022)

Layout Editor

Fixed

Deactivating auto-cloned fields – Previously, all auto-cloned fields of a layout variation were linked to a single shared field. Deactivating any of the layout variation’s auto-cloned fields caused this layout variation’s Layout Editor page to crash when a user attempted to open it. A fix for this issue is included in v34.0.12.

Flows

Fixed

Adding connections to disabled flows – We’ve fixed an issue that prevented users from saving changes when adding connections to disabled flows if there were validation errors caused by the connections’ settings.

Submission Processing

Fixed

Excluding a layout variation and halted submissions – Previously, if you had a live layout with multiple variations and successfully processed documents with one of these variations, excluding this variation from a new release resulted in halted submissions when resubmitting the same documents. A fix for this issue is included in v34.0.12.

Submissions Table

Fixed

Names of disabled flows – We've fixed an issue that caused the incorrect names of disabled flows to appear in the Submissions table. Previously, a combination of the flow's identifier and UUID would be given as the flow's name.

Tasks

Fixed

Flow filter when switching tabs – We’ve fixed an issue that reset the flow filter when switching from the Perform Tasks tab to the Task Queue tab.

Manual Classification

Fixed

Manual Classification task after importing renamed layout variations – We’ve fixed an issue with importing renamed layout variations that caused these variations to be displayed with their old names in the Layout drop-down list during Manual Classification.

Connections

Updated

“Scope” and “Additional OAuth Request Parameters” settings for HTTP Notifier Blocks – We've added optional Scope and Additional OAuth Request Parameters settings for the OAuth 2.0 Client Credentials authorization type in HTTP Notifier Blocks.

Security

Fixed

Updating lxml – To address a security vulnerability, we've updated lxml to 4.9.1.

34.0.11 (20 Oct 2022)

Layout Editor

Updated

Deleting multiple fields in the Layout Editor – You can now delete multiple fields in the Layout Editor at the same time.

Making changes to shared fields – We've reduced the time it takes for the system to register changes made to shared fields in the Layout Editor.

Table Identification QA

Fixed

Clicking Mark Layout Variation Incorrect – We've resolved an issue that caused an error to occur when a user clicked Mark Layout Variation Incorrect during Table Identification QA tasks.

Reporting

New

Field Exceptions Report – We've added a Field Exceptions Report to the Accuracy tab in the Reporting section of the application. For each field exception, the report includes information about the field's submission, layout, document, page, and field.

Fixed

Count of submissions processed – We've fixed an issue in high-volume instances that caused the submission counts in our reporting to differ from the number of completed submissions found in the database. To determine if your instance was affected by this issue, contact your Hyperscience representative.

34.0.10 (7 Oct 2022)

Submission Initialization Block

Fixed

Names of files from Generic Web Storage (HTTP/HTTPS) file stores – Previously, the filenames generated for files from Generic Web Storage (HTTP/HTTPS) file stores would sometimes cause errors due to their length. A fix for this issue is included in this version.

API

Updated

Flow-based permissions for API endpoints – To protect API endpoints against unauthorized access from flow-restricted users, we’ve applied flow-based permissions. All v5 API endpoints that access flow-related resources are now protected by flow-based permissions. For example, users who do not have access to a given flow won’t have access to submissions that are processed by this flow.

If you use API calls to access submission resources, make sure that users are assigned permissions to all relevant flows that process those submissions.

For SaaS instances that have API accounts, make sure to assign all relevant flows to the API User permission group.

34.0.9 (28 Sep 2022)

User Experience

Fixed

Assigning a flow to a permission group – To improve the user experience of assigning a flow to a permission group, we’ve renamed the Available Live Flows label to Available Flows. The Available Live Flows label was incorrect as the flows table in a permission group’s configuration page included both live and disabled flows.

Layouts

Fixed

Adding a layout variation with a name that exists – We’ve fixed an issue that caused a 500 error to occur when attempting to add a layout variation with a name that has already been used.

Connections

Updated

Support for Microsoft 365 in the Email Listener – The Email Listener now supports connections to Microsoft 365 Outlook accounts. To ingest emails from Office 365 inboxes, select Microsoft 365 Outlook from the new Email Provider drop-down list in the Email Listener connection's settings. Then, enter your account and authentication information in the Connection settings fields.

Note that the Email Listener uses Microsoft Graph APIs for account authentication; OAuth authentication is not supported at this time.

Kubernetes

Fixed

Custom Code Blocks with Python libraries – We’ve fixed an issue with some tasks that prevented Custom Code Blocks with Python libraries from completing.

Backward-incompatible API – For v34.0.9, we’ve made the API backward incompatible for Kubernetes deployments. This application version is compatible only with version 4.0.0+ of the operator and version 6.0.0+ of the Hyperscience Helm chart.

34.0.8 (15 Sep 2022)

Classification

Fixed

Multiple high-confidence layout candidates for Structured pages – We’ve fixed a page-matching issue with Structured pages that had more than one high-confidence layout candidate. This issue caused errors during the recalibration of Classification models for Structured documents.

Quality Assurance

Fixed

Deleting a child row during the child-table step in Table Identification QA – We’ve fixed an issue with Table Identification QA tasks for nested tables that caused child-parent associations to be incorrect during the parent-table step. This issue occurred if you had previously deleted a child row during the child-table step.

Reaching consensus during Table Identification QA – Previously, if you:

had added an extra row to your table during Table Identification, and
had submitted the document, and
had completed Table ID QA without making any changes,

then consensus was not reached. To reach consensus, you had to complete a second round of Table ID QA. A fix for this issue is included in this version.

Annotating child rows that span over multiple pages – We’ve fixed an issue with Table Identification QA tasks for nested tables that prevented child rows that spanned over multiple pages from being annotated.

Keyer Data Management

Fixed

Annotating parent rows that span over multiple pages – We’ve fixed an issue that prevented annotations of parent rows that spanned over multiple pages from being saved when using Keyer Data Management tools.

Models

Fixed

Memory usage when downloading Classification models and training data – We've fixed an issue that caused large amounts of memory to be consumed when users downloaded Classification models and their training data.

Reporting

Updated

Filtering by flow – We’ve added a Flow filter to the following reports:

Manual Accuracy vs Machine Accuracy (Reporting > Accuracy)
Performance Distribution (Reporting > User Performance)
All Users Performance Summary (Reporting > User Performance)

We’ve also added a Flow column to the following “.csv” files from the Keyer Projection Report (Reporting > User Performance):

HourlyReportingTaskOverview

HourlyReportingSubmissionOverview

HistoricalProcessing

KeyerPerformance

Fixed

Software Version column in the Usage report – We’ve fixed an issue that prevented the Software Version column in the Usage report (Reporting > Usage) from being populated.

Installations

Fixed

Installing the application on instances that use MSSQL databases – We’ve fixed an issue with installing the application on instances that use MSSQL databases. This issue caused timeouts to occur and prevented users from completing the installation process.

34.0.7 (24 Aug 2022)

Classification

Fixed

Optimized Machine Classification tasks – We’ve fixed an issue that caused Machine Classification tasks to take a few hours to complete when processing the first submission from a newly-created release.

Table Identification

Fixed

Using the “Next column” and “Previous column” shortcuts with Caps Lock enabled – Previously, when the Caps Lock key was enabled, using the keyboard shortcuts for Next column (E or ⬇) and Previous column (W or ⬆) reversed the order in which the columns were iterated over. A fix for this issue is included in this version.

Layout Editor

Fixed

Deactivating fields in Structured layout variations – Previously, deactivating fields in Structured layout variations prevented these fields from being displayed in the layout editor’s Inactive items list. If you clicked the Commit changes button after deactivating some fields in a layout variation, the same issue also prevented you from opening this layout variation. A fix for this issue is included in v34.0.7.

Integrations

Fixed

Using the Salesforce Listener in EST time zone – Before the Salesforce Listener Block starts running, it always checks if a record for the configured channel exists in Salesforce’s Last_Processed_Ingestion_Time object. If no record exists, the Salesforce Listener Block creates a new one. Using the Salesforce Listener in a machine that is configured in EST time zone caused the record in the Salesforce’s Last_Processed_Ingestion_Time object to be incorrect. A fix for this issue is included in this version.

Reporting

Updated

Filtering by flow – We’ve added a Flow filter to the following reports:

Manual Accuracy vs Machine Accuracy (Reporting > Accuracy)
Performance Distribution (Reporting > User Performance)
All Users Performance Summary (Reporting > User Performance)

We’ve also added a Flow column to the following “.csv” files from the Keyer Projection Report (Reporting > User Performance):

HourlyReportingTaskOverview
HourlyReportingSubmissionOverview
HistoricalProcessing
KeyerPerformance
- Daily breakdowns in KeyerPerformance

Security

Fixed

Upgrading okhttp – To address a security vulnerability, we've upgraded okhttp to the latest stable version, 4.10.0.

34.0.6 (11 Aug 2022)

Cases

Fixed

Refreshing the Cases page after editing or deleting cases – We've fixed an issue that prevented the data on the Cases page from being refreshed after a user edited or deleted a case. For example, a deleted case would still appear in the Cases table after being deleted.

Tables

Fixed

QA tasks for nested tables – We've fixed an issue in nested tables that caused the information shown during Table Identification QA to differ from that shown during the sampled Supervision task. Specifically, some row numbers did not appear, and the system would sometimes create multiple occurrences for a single value in a cell.

Security

Fixed

HttpResponse usage – To address security vulnerabilities, we've made updates to the system's usage of HttpResponse, including adding missing content-type information and output encoding.

34.0.5 (8 Aug 2022)

Model Validation Tasks

Fixed

Multiple bounding boxes for a single-line field values – We've fixed an issue that caused the system to create multiple bounding boxes for single-line field values in some model validation tasks (MVTs).

Keyer Data Management

Fixed

Saving changes to field locations – We've fixed an issue that prevented revised field locations from being completely overwritten by their updated versions.

SaaS

Updated

Access to /admin – We've restricted /admin access in SaaS instances to Hyperscience employees.

Fixed

Access to other users' API tokens – For security purposes, we've removed the ability of System Admins and Business Admins to view the API tokens of other users.

Security

Fixed

Caching of pages containing submission images – To guard against unauthorized access to sensitive information, the system now prevents pages containing submission images from being cached by intermediary proxy servers and local web browsers.

34.0.4 (5 Aug 2022)

v34 is not recommended for nested tables. v35 will contain several enhancements and fixes for nested tables. To learn more about known issues and limitations in v34, see Known Issues and Limitations in v34.

Flows

Updated

Enhancements to flow filtering – We've improved how the system filters flow-specific data in the application and in reports. In particular, the processes for filtering out restricted flows and determining whether a flow is a Notification flow are more efficient.

Fixed

Filtering by flows that a user cannot access – We've fixed an issue that allowed users to filter tables and reports by flows that they did not have access to.

Submissions

Fixed

Search-box placeholder text – Because we no longer support searching submissions by file name, we've removed the "file name" reference in the placeholder text in the Submission table's search box.

Submission Processing

Fixed

Resubmitting v31 submissions – Previously, submissions that had been successfully matched to layout variations in v31 weren’t matched to any layout variations after being resubmitted in v34. A fix for this issue is included in this version.

Cases

Updated

Cursor-based pagination in the Cases table – We've implemented cursor-paged pagination in the Cases table. As part of this change, we've removed the Per Page drop-down list from the bottom of the table, and we've changed the case count to display "100+ Cases" when the total number of cases exceeds 100.

Fixed

Multiple-page documents in the Documents table – We've resolved an issue that caused multiple-page documents to appear multiple times in the Documents table on the Case Details page, one time for each page in the document.

System responsiveness when searching for cases – We've fixed an issue that caused search results to take more than 3 seconds to load on the Cases page when there were a large number of cases in the system.

Response times for retrieving case information – We've updated the queries used to retrieve case data by external case ID, making the system more responsive when users navigate the Cases table, upload submissions, or complete Custom Supervision tasks.

Supervision

Fixed

Redirecting keyers after completing submissions – We've resolved an issue that caused keyers to be redirected to the Submissions page after completing a submission's Supervision tasks. The system now redirects keyers to the Submission Completed page.

Quality Assurance

Fixed

Classification QA and images marked for deletion – We've fixed an issue that caused Classification QA tasks to contain page images that had already been deleted. When generating a Classification QA task, the system now confirms that the page images included in the task are not marked for deletion.

Connections

Fixed

Deserialization of boolean and integer values in Message Queue Listener input blocks – We've resolved an issue that caused Message Queue Listener input blocks to convert boolean and integer values to strings during message deserialization.

Databases

Fixed

Column indexing in Microsoft SQL Server (MSSQL) databases – We've fixed an issue in MSSQL databases that caused ALTER TABLE errors when migrating to 64-bit integer column indexes.

34.0.3 (18 Jul 2022)

Layouts

Fixed

Queries during layout import – We've removed a query from the layout-import process that caused errors when importing layouts and upgrading Hyperscience.

Layout Editor

Fixed

Transcription of layout identifiers – We've resolved an issue that prevented layout identifiers from being transcribed.

Editing field metadata in the layout editor in IE 11 – We’ve fixed an issue with the layout editor that resulted in an error in IE 11. The issue occurred in the following cases after uploading layouts:

When a user cleared an existing field’s name and then duplicated that field.
When a user edited any existing field.

Interacting with the layout editor after taking a partial screenshot on a Mac – Previously, taking a partial screenshot on a Mac caused selected fields in the layout editor to automatically close when you clicked on those fields’ multi-option or input settings. A fix for this issue is included in this version.

Releases

Fixed

Deploying empty releases – We've fixed an issue that allowed users to assign empty releases to flows and then deploy those flows. Doing so resulted in errors when the system attempted to classify documents.

Flows

Fixed

Importing flows from v32 and earlier – We’ve fixed an issue that prevented imported flows from v32 and earlier that referenced secrets in their inputs from working.

Flow Blocks

Fixed

Submitting block outputs to localhost URLs – We've fixed an issue that caused blocks to submit their outputs to localhost URLs in some situations. This issue resulted in halted submissions.

Populating and changing field locations in a Custom Code Block – We've resolved an issue that prevented field locations from being passed to Custom Code Blocks and changed by the blocks' code.

Submissions

Updated

Populating corrected_image_url for unassigned pages – We now provide corrected_image_url values for pages that are not assigned to a layout. With this update, you can obtain URLs for transformed versions of these pages in submission JSON outputs and in Page data in API v5 responses.

Submission Processing

Fixed

Optimized data-aggregation queries upon submission completion – To increase performance, we’ve optimized the data-aggregation queries that run upon submission completion.

Cases

Fixed

Sorting the Cases table by Case ID – We've resolved an issue that caused sorting the Cases table by Case ID to take several seconds, even when there weren't a large number of cases in the system.

Text Classification

New

Ability to upload text classification models – To improve the user experience and give you more control over text classification, we’ve added the option to upload text classification models at /admin/form_extraction/textclassificationmlmodel/.

Tasks

Fixed

Performing tasks from the Cases page – We've resolved an issue that caused 404 errors to occur after keyers processed documents by clicking Perform Tasks on the Cases page.

Supervision

Fixed

All fields pointing to the same bounding box during Flexible Extraction – We've resolved an issue that caused all fields in a document to point to the same bounding box during Flexible Extraction tasks. The issue occurred when none of the fields in the document had shared_field_id values.

Supervision Widget

Updated

Smooth token updates and support for JSON Web Tokens (JWTs) – The embedded Supervision widget now accepts JWTs in addition to the existing token authentication. The third-party system the Supervision widget is embedded into handles the management of JWTs.

We’ve also added two distinct approaches to ensure smooth token updates without sacrificing user experience:

an onTokenExpiring callback that is called when the token is close to expiring. The callback should return a Promise object with the value of the renewed token.
an updateAuthToken function that updates the authentication token on demand.

Headers for JWT (JSON Web Token) authentication – We've updated the default header for JWT-authenticated API requests from a custom header to the standard Authentication: Bearer header. As part of this change, the system automatically detects whether a JWT or Django REST Framework (DRF) token is being passed in Authentication: Bearer .

Connections

Updated

Metadata support for Amazon SQS connections – We've added Additional SQS Metadata to the connection settings for Amazon SQS connections, which allows you to enter metadata as key-value pairs.

Fixed

Deserialization of boolean and integer values in Message Queue Listener input blocks – We've fixed an issue that caused Message Queue Listener input blocks to convert boolean and integer values to strings during message deserialization.

Reporting

Fixed

Accuracy and User Performance tabs – We’ve fixed an issue that resulted in an unexpected error when opening the Accuracy and User Performance tabs in the Reporting section of the application.

Finetuning Models

Fixed

Importing and exporting finetuning models – We’ve added a warning message that informs users that they cannot import or export finetuning models if their instance does not have any flows.

The warning message for exporting appears when clicking the Export button on the Export Finetuning Model page at /admin/forms_qa/finetuningmodelmeta/export-file/.
The warning message for importing appears when clicking the Import button on the Import Finetuning Model page at /admin/forms_qa/finetuningmodelmeta/import-file/.

Security

Fixed

Authorization for /admin URLs – To address security vulnerabilities, we've updated the permissions required to access several /admin URLs that allow users to upload files or execute code.

34.0.2 (4 Jul 2022)

Transcription

Updated

Support for multiple bounding boxes in recalibration and auto-thresholding – To allow you to continuously optimize transcription automation and minimize human effort needed to process documents, we’ve added support for multiple bounding boxes in recalibration and auto-thresholding.

Quality Assurance

Updated

Optimized Table Transcription QA queries – To increase performance for instances that contain a large number of documents with table cells, we’ve removed unnecessary JOIN clauses from Table Transcription QA queries.

Models

Updated

System Upgrade Evaluation tab and increased performance – To increase performance, we’ve split the request for loading the Current System Version and System Upgrade Evaluation tabs in the Model Details page. We now have a separate request that loads the System Upgrade Evaluation tab only when a user tries to open the tab.

Fixed

Field ID models and multiple bounding boxes – We’ve fixed an issue that caused Field ID models to predict multiple bounding boxes for fields whose values are located on the same line. With the fix included in this version, values that are located on the same line are predicted as single bounding boxes.

Recalibration of Classification models for Structured documents – We've fixed an issue that caused the page IDs of Flexible Extraction documents to point to pages without fields in some situations. This issue caused errors during the recalibration of Classification models for Structured documents.

API

Fixed

Client ID validation in JSON Web Token (JWT) authentication – We’ve enabled client ID validation in JWT authentication. We now properly handle errors when JWT authentication fails due to being disabled.

34.0.1 (1 Jul 2022)

This version of v34 is not officially supported. The next patch version of v34, v34.0.2, is officially supported and contains all of the updates described below.

For more information, contact your Hyperscience representative.

Text Classification

New

Text Classification preview – With our new Text Classification feature, you can analyze and organize extracted text by user intent, sentiment, topic, or any custom labels based on your own business rules.

This update to v34 includes a preview version of Text Classification, which will be released in full in a future version of Hyperscience.

Languages

Updated

Supporting recalibration, finetuning, and auto-thresholding for Korean and Arabic languages – We’ve added support for recalibration, finetuning, and auto-thresholding for Korean and Arabic languages.

Support for multiple bounding boxes in Arabic language during Machine Transcription – We’ve added support for transcribing multiple bounding boxes in Arabic during Machine Transcription tasks.

Ordering of multiple bounding boxes in Arabic language – Transcription Supervision tasks, Transcription QA tasks, Field ID QA tasks, and Keyer Data Management now present multiple bounding boxes in Arabic language to users from top to bottom and right to left.

Layout Editor

Fixed

Transcription of layout identifiers – We've resolved an issue that prevented layout identifiers from being transcribed.

Copying of field-specific languages – We've fixed an issue that prevented a field's language from being copied along with the rest of the field's settings when copying the field from one layout variation to another.

Clicking "Undo" when adding a field – We've fixed an issue that caused the Layout Editor to enter an invalid state when a user clicked Undo after creating a field with no Name or Data Type values. In these situations, a "Missing Info" notification would be shown to the user, but the user could not provide the "missing info" or commit their changes.

Text of "Read only" label in the Layout Editor – Previously, the "Read only" label in the Layout Editor had white text on a white background, making the text illegible. The issue is fixed in this version, and the "Read only" text is now legible.

Layouts

Fixed

Importing field dictionaries – We've fixed an issue that prevented field dictionaries from being imported successfully. The same issue also caused users to be redirected to the Import Field Dictionary dialog box after attempting to import a field dictionary.

Archiving layouts in a release assigned to a flow – We've resolved an issue that caused the Archive option to be available for layouts in releases assigned to flows. These layouts cannot be archived, so the Archive option should not have been available.

Releases

Fixed

Release validation error when no release is available – Previously, the release validation error shown when no release was available showed (missing) null as the selected release. A fix for this issue is included in this version.

Auto-saving of releases when selecting a different layout version – We’ve fixed an issue that prevented releases from auto-saving when a user selected a different version of a layout.

Models

Fixed

Memory usage when downloading Classification models and training data – We've fixed an issue that caused large amounts of memory to be consumed when users downloaded Classification models and their training data.

Flows

Fixed

Importing flows in ZIP files – We’ve fixed an issue that caused the flow-import process to fail when importing a ZIP file that was created manually by compressing the parent directory.

Counts of inserted and deleted lines of code – We've resolved an issue that caused the counts of inserted and deleted lines of code shown during the flow-import process to be incorrect in some situations.

Availability of flow settings – We've fixed an issue that caused some flow settings to be hidden when they should not have been.

Blocks for future tasks in Flow Execution diagrams – We've resolved an issue that caused the background for future tasks in Flow Execution diagrams to be transparent instead of gray.

Active finetuning models and using the Flow Studio – We’ve fixed an issue that caused errors for transcription settings to appear in the Flow Studio if you had at least one active finetuning model in the system for the current version of the application.

Submission Processing

Fixed

Memory usage for large submissions – We've fixed an issue that caused the system's processing of large submissions to consume a large amount of memory, with the system using all available memory in some circumstances.

Submissions

Fixed

Downloading submissions when the Submissions table is filtered – Previously, if multiple values were selected for a Submissions table filter and a user attempted to download submission-level data, data was not downloaded for all of the submissions in the table. Instead, data was downloaded only for submissions that matched the last filter criteria listed in the URL query parameters. A fix for this issue is included in this version.

Submission Output

Fixed

Field Identification percentage in the Automation card – We’ve fixed an issue that caused the Field Identification percentage in the Automation card on the Submission Output page to be incorrect in the following cases:

Multiple occurrences of a field were on different pages.
The first page did not have any fields.

Rejected documents in the Submission Output page – We’ve fixed an issue that caused documents that were rejected during Transcription Supervision to be accessible in the Submission Output page.

Table Identification

Fixed

Annotating multi-page documents – We’ve fixed an issue that caused multi-page documents’ annotations to be incorrectly visualized on all pages except the first one.

Duplicated table cells – We’ve fixed an issue that caused table cells to be duplicated during Table Identification tasks when using a flow that had been created in v32 or earlier.

Custom Supervision

Fixed

Bounding boxes of read-only transcription fields – We’ve fixed an issue that prevented users from selecting bounding boxes of read-only transcription fields.

Table editor and unidentified table columns – We’ve fixed an issue that caused the table editor to close automatically. The issue occurred when users selected cells in columns that were referenced in the Custom Supervision’s template but had not been identified during Machine or Manual Identification tasks.

Transcribed fields with overridden formatting validations – We’ve fixed an issue that caused fields with overridden formatting validations to be blank in Custom Supervision even if they had been transcribed in a previous task.

Transcription

Updated

Optimized database queries for Flexible Extraction tasks – To increase performance, we’ve removed unnecessary JOIN clauses from the queries that generate Flexible Extraction tasks.

Truncating transcription fields – We’ve fixed an issue that prevented the system from truncating transcription fields that exceed the 2000-character limit. Transcription fields that exceed the limit are now automatically truncated to 2000 characters.

Fixed

“None” normalized transcription values and the Semi-structured Transcription Confidence Boost setting – We’ve fixed an issue that prevented the Semi-structured Transcription Confidence Boost setting from handling “None” normalized transcription values.

Quality Assurance

Fixed

Fields without locations and Transcription QA – Previously, if a user manually classified a Structured document and it did not go through Flexible Extraction, fields without locations were sent to Transcription QA. This issue resulted in an error when trying to complete the Transcription QA task. With the fix included in this version, fields without locations are no longer sent to Transcription QA.

Model Validation Tasks

Updated

Optimized database queries for Model Validation Tasks (MVTs) – To increase performance, we’ve removed unnecessary JOIN clauses from the queries that generate Field ID MVTs and save their results.

Fixed

Model Validation Tasks (MVTs) for nested tables – We’ve fixed an issue that prevented the system from validating if all table columns in a nested table should go through MVTs. As a result, a few extra MVTs were created for nested tables, even though columns from these MVTs had already been validated.

Cases

Updated

Optimized queries for loading the Case Details page for instances with database encryption – To increase performance, we’ve optimized the queries for loading the Case Details page for instances with database encryption.

Keyer Data Management

Fixed

Long segments of text in the right-hand sidebar – Previously, if you had added 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 changed depending on the text’s length. With the fix included in this version, the right-hand sidebar’s size is consistent, regardless of the text’s length.

Training

Updated

Reusing training data after layout changes – Adding new fields or changing fields’ data types will no longer exclude all previous documents from training. If you add a new field or change a field’s data type, the field will become unsupported, but you can still start model training with existing documents on all other fields. To start model training that includes a new or a changed field, you need to annotate enough documents with this field first.

Fixed

Eligible documents for Table ID model training – We’ve fixed an issue that caused documents that had not reached consensus after completing a round of Table ID QA to be eligible for Table ID model training. The issue occurred when the Tables Nonconsensus Training Disabled setting was enabled in /admin.

Projected automation for Field ID models – We’ve fixed an issue that caused Field ID models to have projected cell automation on the Model Details page, even though Field ID models should not have cell automation.

Training with multiple-occurrence fields and checkboxes or signatures – We've fixed a filtering issue that created empty occurrences if both multiple-occurrence fields and checkboxes or signatures were present in a dataset. These empty occurrences caused training to fail.

Documents without field locations and Field ID training – We’ve fixed an issue that prevented documents without any field locations from being eligible for Field ID training.

Uploading models to future trainers – We’ve fixed an issue that prevented users from uploading models to future trainers in the System Upgrade Evaluation tab.

Warning message for instances without a trainer – We’ve added a “No Active Trainer” warning message to the Model Details page. If there is no trainer attached, the following message informs users that training tasks can only be queued but cannot be completed: “Trainer tasks can be created and queued, but they will not be completed because there is no active trainer.”

Connections

Fixed

Saving Folder Listener settings when required settings are missing – We've fixed an issue that allowed users to save their changes to Folder Listener Input Block settings even when the required Folder to Scan for Submissions and File Extensions settings had no values.

Messaging for HTTP REST Block failure – Previously, if the HTTP REST Block received a response with a 202 status code and an empty request body, the block would throw a JSON exception without displaying an explanatory message. A fix for this issue is included in this version, and a message appears explaining that the response's JSON is invalid.

Submissions created from the Box Folder Listener Block – If the folder specified in Folder to Move Completed Files did not exist, the Box Folder Listener Block created an infinite number of submissions. With the fix included in this version, the Box Folder Listener Block creates submissions only if the folder specified in Folder to Move Completed Files exists.

Telemetry

Fixed

Enabling Telemetry — We've resolved an issue that caused a 500 error when a user enabled the Telemetry feature in Settings (Administration > Settings).

Permissions

Updated

Warning message when assigning permission groups to flows – We’ve added a warning message that informs users that flow-based permissions are not applied to existing tasks. The warning message appears when assigning a permission group to a flow.

Security

Fixed

Updating Django to 2.2.28 and 3.2.13 – To address security vulnerabilities, we've updated Django to the latest security releases, 2.2.28 and 3.2.13.

Changing the javax.el library to jakarta.el and updating it – To address a security vulnerability, we’ve changed the Maven javax.el library to jakarta.el and updated it to 3.0.4.

Referencing system secrets with the Edit Flows permission – We've fixed an issue that allowed users with the Edit Flows permission to reference existing system secrets when editing flows. These users can now create references to new system secrets only, preventing system secrets from being exposed when cross-referencing fields or flows.

SaaS

Updated

Restricting access to SQL Explorer – In SaaS deployments, we've restricted access to the SQL Explorer to Hyperscience employees only.

Upgrades

Fixed

Optimized queries for database migrations – To decrease upgrade time, we’ve optimized the queries for database migrations.

34.0.0 (14 June 2022)

User Experience

Updated

New color scheme for field annotations – To differentiate among the fields you need to identify, we’ve implemented a new color scheme. This color scheme includes 12 different colors and is applicable to:

Identification Supervision tasks
Identification QA tasks
Model Validation tasks
Keyer Data Management

Each field is highlighted with a different color. If a field has multiple occurrences or multiple bounding boxes, all of the field’s occurrences and bounding boxes are highlighted with the same color.

Languages

New

Support for Polish submissions – We now support automation on Structured and Semi-structured documents written in Polish. The Polish language model allows our system to extract data from Polish documents, and keyers can complete Transcription Supervision and Transcription QA tasks by entering Polish text.

Note that the Alphanumeric data type is not fully compatible with Polish at this time.

Updated

Enhanced support for Korean documents – We've improved the system's ability to automate the transcription of Korean documents in the following ways:

The system can transcribe documents written in Korean, regardless of whether they're Structured or Semi-structured or if the text is handwritten or printed.
We now support the transcription of the Korean won symbol ("₩").

Layout Editor

New

Support for field-specific languages – If you know that certain fields' text will be written in a language different from the surrounding text, you can select the new Not in option for each of those fields when creating the layout. Then, you can choose specific languages for each of those fields. For example, if you wanted to process an Arabic document that contains an Email field written in English, you would set the layout's language to Arabic, select the Not in Arabic option for the Email field, and then select English from the drop-down list that appears.

The available languages for each field depend on the type of layout you are creating, as different languages are supported for Structured and Semi-structured layouts, in general. As long as a language is supported for the type of layout you are creating, you can select that language for any text field in your layout, regardless of the layout's language family.

Note that information about field-specific languages is not included in the document's JSON output. However, the language for each field is preserved when importing and exporting the fields' layout.

Replacing the “Identification Supervision” checkbox with a multi-option “Identification Supervision” setting (Semi-structured only) – To expand the system’s field-level identification capabilities, we’ve replaced the Identification Supervision checkbox with a new setting called Identification Supervision. The new setting allows you to skip or force Field ID Supervision for a given field. The Identification Supervision setting includes a drop-down list with the following options:

Auto-identify – Skips Field ID Supervision and identifies the field automatically.
Default – The field undergoes machine identification. If the machine’s confidence level on its identification is above the set threshold, the machine will automatically identify the field. If the confidence level is below the threshold, the field will be sent to Field ID Supervision.
Always – Forces the field to always go to Field ID Supervision.

Updated

Renaming the “Supervision” setting to “Transcription Supervision” – We’ve renamed the Supervision setting in the Layout Editor to Transcription Supervision.

Flows

New

Managing archived flows – To allow you to manage archived flows in your instance, we've made the following updates:

When you enable the Show Archived flows option at the top of the Flows page, the list of flows on that page includes both archived and available flows.
Enabling this option gives you the ability to export, duplicate, and restore each archived flow, as well as copy the flow's UUID.

Updated

Importing new versions of flows containing Custom Code Blocks – When you import a new version of a flow that contains Custom Code Blocks, the system now compares not only the current and imported JSON files, but also the current and imported Python files for the Custom Code Blocks. When you select a JSON or Python file in the import tool, the system shows you the portions of the code that will be overwritten if you choose to proceed with the import.

You then have the option of:

downloading the current JSON file for the flow,
committing the changes, or
canceling the import process.

Note that you cannot selectively import particular files in a flow.

Visibility of "Document Processing Notification" flows – The list of flows on the Flows page now includes the “Document Processing Notification" flows available in your instance. These flows had previously been hidden by default, and making them visible required assistance from Hyperscience.

Note that the “Document Processing Notification” flows for v34 and later are named “Submission State Notifications for Document Processing ([version number]).”

Flow Blocks

Updated

Inclusion of flow blocks from previous versions of Hyperscience – Previously, each version of Hyperscience contained only the blocks that were created for that version. During upgrades, a Hyperscience representative needed to manually upload any blocks from previous versions that were included in the instance's flows. In v34, blocks from v30 through v34 are included in the application.

Flows SDK

Updated

Support for custom Python packages in Custom Code Blocks – If you would like to use custom Python libraries in your Custom Code Blocks, you can give your system administrators the ability to install them on one of your application machines. To do so, add the HS_ALLOW_EXTERNAL_PYTHON_PACKAGES_UPLOAD variable to your “.env” file and set it to true. Any Python packages installed will appear in the Python Packages page (Flows > Python Packages).

To learn more about installing and uninstalling Python packages, see our Flows SDK documentation.

Document Classification

Updated

Document Classification improvements – To enhance the Document Classification experience—and give you more flexibility in how you group pages—we've made the following updates:

When classifying pages in a submission or case, you can group and regroup pages into documents. You can also change the order of the pages in Semi-structured documents and split a document into multiple documents, if needed.
Note that you cannot regroup or reorder pages that have already been matched to a Structured layout.
As you classify the pages, you can see the page's page number in the submission, along with the submission's notes and attachments. This information can give helpful context when grouping pages into documents.

Field Identification

New

Ability to search for text segments within a document (Semi-structured only) – To improve the user experience and searchability during Field Identification tasks, we’ve added a search bar at the top of the Field Identification page. This search bar allows you to search for keywords across all pages of a document whose fields are currently being identified.

We’ve also added a number of keyboard shortcuts that help you use the search functionality of the Field Identification page more effectively.

Identifying multiple bounding boxes for a single field (Semi-structured only) – You can now capture fields that contain line or page breaks. We’ve added the option to create multiple bounding boxes for a single field during Field Identification Supervision, Field Identification QA, and model validation tasks (MVTs). Creating multiple bounding boxes allows you to be more flexible in capturing data and reduces the margin of error.

For example, if a patent application has a multi-page description that you need to extract, you can use multiple bounding boxes to capture the text on each page. The keyer completing Field Identification Supervision can identify the first part of the description in the document, indicate that there are more parts that need to be identified on other pages, and then identify the remaining parts consecutively.

If a document contains multiple bounding boxes for a field, its Document Output page shows all identified text segments for the field. The text segments are also included in the JSON outputs for the document and its submission.

Table Identification

New

Nested tables (Semi-structured only) – To make processing of tables with complicated nesting structures possible, we’ve added support for nested tables. With this update, you can take advantage of the following new functionalities:

Defining a parent and a child table in Semi-structured layouts – You can define a parent and a child table with their respective columns in the Layout Editor for each Semi-structured layout.
Table ID Supervision tasks for nested tables – During Table ID Supervision, the system creates a single task where you first have to identify your child table’s cells. Then, you have to identify your parent table’s cells.
Table ID QA tasks, model validation tasks, and automation for nested tables – To reduce the need for human input while still reaching the desired output accuracy, we allow automation for nested tables. You can train Table ID models with nested tables. With the addition of Table ID QA tasks and model validation tasks (MVTs) for nested tables, you can improve the system’s performance by adding, confirming, and fixing cell annotations.
Transcription Supervision tasks for nested tables – During Transcription Supervision for nested tables, you can transcribe the cells in your child and parent tables.
Transcription QA tasks for nested tables – With the addition of Transcription QA tasks for nested tables, humans can give feedback to the machine and improve transcription accuracy over time. Table cells from your child and parent tables are combined together in a single Transcription QA task.
Document Output page – To make sure the correct data is extracted from a nested table, you can now review this data in the Document Output page.

Transcription

New

Transcribing fields with multiple bounding boxes – We’ve added support for transcribing fields with multiple bounding boxes in Machine Transcription, Transcription Supervision, and Transcription QA tasks.

During Machine Transcription, the machine transcribes each text segment of a field separately and then concatenates all text segments together.
During Transcription Supervision and Transcription QA, the system generates a single task for each field with multiple bounding boxes. Within this single task, you can navigate between the different text segments.

To improve transcription accuracy and automation, the system uses the consensus data from fields with multiple bounding boxes for auto-thresholding training.

Custom Supervision

New

Transcription and custom decisions for tables – To expand the capabilities of Custom Supervision, we’ve added support for table transcription and table decisioning. You can now view and select table columns from the Custom Supervision interface’s right panel. Selecting a table column from the right panel reveals a table editor at the bottom of the middle panel. This table editor allows you to do the following:

Review and edit table cells’ transcriptions.
Apply cell-level custom decisions.

We’ve also added keyboard shortcuts for Table Transcription tasks during Custom Supervision.

Removing documents from cases – To allow users to correct case collation mistakes that were made before the Custom Supervision task, we’ve added the option to remove documents from cases.

“Needs Review” feature for transcription fields – To make sure a user reviews a particular field before completing a Custom Supervision task, you can now set the optional needs_review parameter for transcription fields in the Custom Supervision input JSON payload. If this parameter is set to true for a field, users need to click on the field and review its contents before they are able to complete the task.

Validations for the Custom Supervision JSON – To prevent misconfigurations of the Custom Supervision task, we’ve added new JSON validations to catch errors during flow upload. More details on these validations can be found in the Flows SDK documentation.

Introducing data_identifier in the Custom Supervision JSON – We have replaced layout_field_id with the more generic data_identifier in the configuration of Custom Supervision tasks. Any component that has a type of transcription, crop_transcription, or table_transcription requires a data_identifier. Specifically:

The transcription type uses a data_identifier corresponding to the layout field UUID.
The crop_transcription type uses a data_identifier corresponding to the UUID of the crop.
The table_transcription type uses a data_identifier corresponding to the layout table UUID.

Pre-populated default options for decision choices - To enable flow developers to pre-select which decision choice is shown by default in a decision drop-down menu, we’ve added the optional decision parameter default_value to the Custom Supervision JSON payload for decisions. This new parameter accepts any of the const values supplied in the decision schema. Single-select decisions only accept a single value, and multi-select decisions can accept multiple values for default_decision.

Keyer Data Management

New

Support for fields with multiple bounding boxes – To give you more control over your Field Identification data and automation, we’re introducing support for fields with multiple bounding boxes in the Keyer Data Management tools. You can now view all bounding boxes that are associated with the ground truth of their respective fields, adjust bounding boxes, add additional text segments, and remove bounding boxes.

Training

Updated

Maximum of 1500 documents for Field ID training – To ensure a smooth training process and prevent crashes, we’ve changed the requirement for training a Field ID model to a maximum of 1500 documents.

Reporting

Updated

Custom Supervision data in the KeyerPerformance.csv – We’ve added the following columns to KeyerPerformance.csv, which is part of the Keyer Projection report (Reporting > User Performance):

Custom Supervision Table Cells Transcribed
Custom Supervision Table Characters Keyed

Adding table cell transcriptions from Custom Supervision tasks to the Supervision Volume report – We’ve updated the Supervision Volume report (Reporting > User Performance) to include data about table cell transcriptions from Custom Supervision tasks.

Support for multiple bounding boxes in the Transcription Sampled Errors and System Transcription Sampled Errors reports – The Transcription Sampled Errors (Reporting > User Performance) and System Transcription Sampled Errors (Reporting > Accuracy) reports now include data about all bounding boxes for fields.

Permissions

Updated

Reorganizing and renaming permissions – The permissions listed on the Permission Groups details page have been reorganized, and some have been renamed. Significant changes to permission names are listed below.

View Debug Tools is now Access Advanced Administration and Debug Tools.
Edit Advanced Settings is now Edit Settings.
View Advanced Settings is now View System Health and Settings.
Edit Task Queue is now Adjust Task Queue Processing Deadlines.
Complete Layout Variation Model Validation is now Complete Identification Model Validation.
Edit Document Restrictions is now Edit Layout Task Restrictions.
Edit User Groups is now Edit Permission Groups.
View Document Restrictions is now View Layout Task Restrictions.
View Users and Group Settings is now View Users and Permission Groups.

Integrations

New

Select endpoints available in MuleSoft Anypoint Exchange – We have published OpenAPI specifications for the following endpoints on the MuleSoft Anypoint Exchange:

Submissions
Create Submission

With this update, mutual customers of Hyperscience and MuleSoft no longer need to add the Hyperscience API to MuleSoft manually.

For more details, see MuleSoft's Anypoint Exchange.

Databases

Updated

Deprecating support for Oracle 12.x – Beginning with v34 of the application, Hyperscience will no longer support the use of Oracle 12.x databases.

API

New

OpenAPI specifications for select endpoints – We’ve generated OpenAPI specifications for the following endpoints:

Submissions
Create Submission

We plan on generating OpenAPI specifications for other endpoints in the future. These efforts will help us to enhance the overall API user experience and optimize the development of future integrations.

You can download these specifications from MuleSoft's Anypoint Exchange.

Added Submission IDs to the Cases endpoint - To aid users in tracing data between Cases and Submissions, we’ve added Submission IDs to the Cases endpoint.