v0.45 - AI native screening = -80% false positive

Marble v0.45 brings you the most advanced options to get razor sharp screening for your payments

• Multiple screening per scenarios : set different types of screening, with dedicated parameters and lists, depending on the transaction you’re checking
• Fine tune your screening to drastically improve efficiency : remove numbers or recurring strings, select specific types of entities (companies, individuals, vehicules…) only, and avoid checking narrow strings generating noise.
• Activate the AI Name Entity Recognition for labels, removing the clutter from long strings to extract companies and individual names

Improvements

• Topics are displayed (sanctions, PEP, wanted...) on screening matches

Fix

• Details of the screenings are now also available in the decision/all endpoint
• Handle "index in creation" status when configuration data exploration from case manager

v0.44.0 - PEP & adverse media lists for screening, object annotations in case manager

New features

• You can now put notes (comments, tags, attach files) on the entities you ingested in the case manage's data viewer ! Those notes will be visible both in the investigation timeline of the case where they have been added, and in any other case related to the same user in the future. No need to switch to the main investigation screen to work, you can do most of the investigation while checking the transactions. Even better : the data you are adding stays connected to the object, and will be displayed and useable in any future investigation.

• Get access to all lists available in Opensanctions for you screening requirements: sanctions, PEP, adverse media, crypto wallets and more!

Feature improvements

• improve interface to configure decision grouping mechanism (formerly called "pivot values") in the data model

• recommend use of the latest version of yente for sanction screening

Bugfixes

• restore access link to already reviewed screening hits in case manager #916
• better persist left bar folded/unfolded state #918
• avoid error when trying to move extensions that may have been pre-created by AWS RDS managed admin user (or possibly similar on other managed databases) #1035
• more reliable whitelisting for screening checks, by better taking into account all "equivalent" ids of an entity in Opensanctions data #1040

v0.43.0 - handle Suspicious Acitivity Reports in case manager, bug fixes

Content

This version of Marble most notably adds the possibility to manage suspicious activity reports (SAR) in the case manager, by marking a case as "SAR to do" or "SAR done" and upload the file.
This is the first step on helping customers handle the full SAR workflow, as we are planning to add help on actually writing the report soon.

Bug fixes

• Allow reading from soft-deleted custom lists during scenario evaluation #1022
• Replace migration moving pg_trgm to check if we are extension owner, avoiding a bug for self-hosted AWS RDS users #1021
• Add time to list of previous cases of the customer in case manager right panel #898
• feat: persist user menu display preference (left panel folded) #899

v0.42.0 - improvements on case manager & bugfixes

This week we focus on some product improvements and bugfixes on the case manager, based on customer feedback from last week's big release.
The last missing pieces of the case manager revamp will be following next week!

Improvements

• New icons for decision outcomes, case status
• Better inbox navigation for the case manager:
  • more readable icons
  • resolved cases are now hidden by default in the inbox - optional filter to show only resolved cases
  • cases are now ordered by default by showing "waiting for action" cases (escalated to an inbox, reassigned etc) first, then by oldest cases first. Clicking on the "date" column header inverts the order.
  • new filter to show only unassigned cases
• The decisions table page now shows decisions of the past month only by default, for better response time in edge cases of filters (e.g. no rows matching, ...)

Bug fixes

• Improve translations on several parts of the case manager #883
• Case outcome (false positive, etc) no longer reset if a resolved case is reopened #881
• Navigation between inboxes properly updates the inbox if filters/pagination had already been applied #892
• Placeholder display in the case right panel if no pivot values have been configured #878
• Correct two bugs related to database field accessors in rule builder: one that caused transient errors when opening the rule page, one that caused some database field accessors to be missing in some cases (involving very nested data models) #1000
• Self-assign case on decision and sanction check review #1007
• avoid noisy logs on backend if Segment is disabled #1013

v0.41.0: major release : New case manager

New case manager

Today we’re proudly announcing the first brick to a 10x more efficient analyst work, with our new case manager. Not just a redesign, it’s the foundation for automation and the end of context switching, ensuring better compliance, less manual work, and giving space to let AI work alongside analysts.

It’s the first of a series of releases set for the next months, gradually setting up our vision for automation.

In this release, you’ll have access to :

• A complete redesign of the case manager, with dynamic action panels on the right
• A dedicated, personal inbox, giving you a view of all cases assigned to you, no matter in which inbox it’s located
• Inboxes filtered by status
• Escalation : set a target inbox to escalate your cases in a single click
• Assignment : any action in a pending case will assign the case to you. You can also reassign cases to other team member
• New case status : the case status have been overhauled for more clarity : Pending (non assigned), in progress (someone’s working on it), closed. The former status are automatically converted.
• Detection improvement status : you’ll be asked to select the value of the alert on each case (False positive, Valuable Alert, Risk confirmed). It’s the foundation to let you train your own ML algorithm using the labels available within Marble, and allow Marble to provide you in the near future with even more efficient detection.
• 1-click “next unassigned case” button for faster navigation
• Case snoozing to automatically reactivate a case at a certain date (great for reminders)
• A brand new investigation view that can be filtered for clearer insights
• A new view on alerts, with clear actions and quick view, and the possibility to expand on a panel all the details. The rules triggered are displayed in the quick view ordered by score (higher to lower)
• The detail of the past alert for the same customer / pivot, with their status, to get quick insights on previous risks
• A dedicated panel for rule snoozing, to make things faster
• A complete data exploration to replace your back office : access all other objects linked to the customer : transactions, events… Navigate within them, filter them, get details.
• A better view on object, with ordered fields and fields hidden by default to remove cluttered data

Configuration

To start working efficiently with this release, we invite you to configure your escalation inboxes and data display, following this 1 minute video presentation

Deployment

⚠️ This release exceptionally introduces some breaking changes to the case status. Please note that once the database migrations included in the version v0.41.X have been run, the backend may return errors as long as it has not been upgraded to the same version, after which the frontend may show errors as long as it has not been upgraded to version v0.41.X.

v0.40.2

Hotfix - Tags creation

• Fixed a regression where tag creation was impossible

v0.40.1

Hotfix - sanction match thresholds

• Check sanctions screening threshold values to two decimals instead of one

v0.40.0 : Fuzzy matching agregates and fixes

This week is the last release, focused on improving agregates with strings and fixes, before our new case manager interface :)

Product improvements

• Fuzzy matching strings within agregates is available, making it easy to check if a string is similar to any past string ingested in an object. See demo here .

Note : the functionality necessitates to add the following extensions to your PostgreSQL : PG_TRGM and FUZZYSTRMATCH. When installing this release, a migration will add those extensions.

• Comment is no longer mandatory when updating a "block and review" decision status

Bugfixes

• Better error message on "table not found"
• Defer freshness call to client-side to avoid possible timeout breaks
• Value in operand menu made multiple hover due to duplicates
• Fix sanction dataset freshness banner UI

Version updates

• Vite dependency bumped from 6.2.4 to 6.2.6
• Remix bumped to 2.26.5

v0.39.2

Bugfixes

• Error on form submit when whitelisting a sanction check match #780
• Accept long form uuids in url in the app #777
• Fix panic leading to server restart if decisions are created on a non-live scenario version #944 (especially in batch mode)
• Return a 500 status code in case of unexpected database error within the authentication middleware #952
• Correct an error where some DB field read options were randomly not available in the rule builder, in some cases of data model structure #954
• Correct an error where the "prepare" button to activate a scenario, sometimes needed to be run several times #950
• Make Marble DB migrations run with a user other than "postgres" #950
• Regression: null values handled gracefully in case name template workflow #953

v0.39.1 - Design improvements, bugfixes, preparation for updated case manager

This week is a small release focused on design improvements and bugfixes again, with preparatory work continuing for our revamped case manager interface.

Product improvements

• Free-form sanctioned entity search + rule snoozing in the public API v1beta #923
• "sanction check" renamed to "screening" in the public API v1beta #935
• Display a warning on the sign up page if no organization and first user have been created yet, to help getting started for self-hosted users #761
• Display the number of values present in a given list, from the table view #769
• Improve display of field selection in the rule builder, especially in the case of long field/table names or several layers of depth in the data model #764

Bugfixes

• Handle "null" index version returned by Opensanctions gracefully #916
• Fix Memory Leak on Firefox by #763
• Correct an error in the helm chart docker image configuration