3 New Features, Enhancements and Limitations in this Release
The OFS MMG 26.2.0 release introduces strategic architectural updates that are designed to enhance system governance, operational resilience, and analytical precision.
New Features
General Features
User Provisioning and Identity Diagnostics
User provisioning and identity handling now include improved SAML validation, error handling, and diagnostics; stronger user-provisioning and identity-mapping validation; and safer fallback behavior with clearer responses when user or group data is missing or invalid.
Business Impact: Provides clearer validation and diagnostic information for troubleshooting SAML, user-provisioning, and identity-mapping issues.
HTTP Strict Transport Security
HTTP Strict Transport Security (HSTS) has been added to response headers to enforce HTTPS browser behavior. Tracking reference: OFSMMG-16684.
Business Impact: Enforces HTTPS browser communication through the HSTS response header.
Workspace
Data Sourcing and Model Upload Diagnostics
Workspace Data Sourcing and model upload processing have been enhanced with clearer object-level execution status and failure visibility, better progress tracking and failure isolation for large uploads and batches, dependency-aware handling and reporting for database objects, improved troubleshooting for partitioned database objects, and separate AAI-DMM service logs.
Business Impact: Provides more granular status and failure information for diagnosing Data Sourcing and model upload operations.
Optional Synonym Creation
Workspace Data Sourcing now supports optional synonym creation.
Business Impact: Allows users to avoid synonym creation when it is not required, supporting the stated objective of reducing storage costs.
Datasets
Dataset Feature Restriction Removed
The Dataset screen no longer restricts datasets to 100 features.
Business Impact: Allows users to work with datasets containing more than 100 features on the Dataset screen.
Independent Custom SQL Queries
During Step 3 of dataset creation, users can provide complex custom SQL queries independently against a single datastore.
Business Impact: Provides flexibility to define dataset logic with independent custom SQL queries on a single datastore.
Model Pipelines
Open Session Monitoring and Management
Model Pipelines now supports monitoring and managing open sessions.
Business Impact: Provides visibility and management controls for open sessions.
Model Monitoring Metadata Management and Persistence
Model Monitoring metadata can be managed and persisted.
Business Impact: Retains Model Monitoring metadata for continued management and use.
Model Monitoring Pipeline View
A Pipeline View is available for Model Monitoring.
Business Impact: Provides a pipeline-oriented view of Model Monitoring information.
Model and AI Risk Management (MARM)
Model Dashboard
- Clickable KPIs with drill-down into filtered Model Inventory
- Risk-prioritized Critical Validation Backlog for timely attention to validation demand
- Model-risk distribution by business domain and production-model validation status
- Model lifecycle-stage and production-model vintage distribution
Business Impact: Improves oversight of model risk and validation demand by bringing the most relevant inventory, risk, and backlog indicators into a single decision-oriented view.
Model Registration in MARM
- For models already available in MMG, MARM performs a readiness check to confirm that the selected model is ready to enter evaluation or validation.
- For external models, a dedicated registration path enables models from external environments to be registered through a streamlined process within MARM.
Business Impact: Provides a clearer, faster path to bring both MMG and external models into the MARM validation lifecycle, while establishing readiness before validation begins.
Model Tiering
Model Tiering now includes automatic tier score calculation, configuration-driven standard tiering criteria, an improved tiering-criteria layout, display of the previous tier, and controlled manual override.
Business Impact: Reduces manual calculation through automatic tier scoring while retaining controlled override and visibility into the previous tier.
Backtesting
Backtesting now supports an As-of Date, allowing validators to select the date as of which date a model is to be validated. Backtesting has also been expanded to probability of default (PD), classification, multiclass, and traditional models.
Business Impact: Enables validation against the relevant historical point in time, helping validator persona capture the real business value of the backtesting process.
Audit Logging
MARM now records validation-related activities as part of the original pipeline audit logs.
Business Impact: Improves traceability by providing a consolidated audit record of validation activity alongside the underlying pipeline history.
Role-Based Access Controls
MARM now supports role-based access controls through three seeded personas: Model Owner, Model Validator, and Lead Validator / Validation Approver. Access is organized through the MMG/OFSAA User-to-User Group-to-Role-to-Function model, with granular view, validation-execution, and approval functions.
Business Impact: Strengthens governance through clear accountability and segregation of duties: Model Owners receive read-only access, Model Validators perform validation work, and Lead Validators / Validation Approvers review and finalize outcomes.
Findings and Observation Tracking
MARM now provides a flow to track findings and observations identified in a validation. Open findings can be carried forward into subsequent validations, enabling validators to continue tracking them until closure.
Business Impact: Provides continuous visibility of unresolved validation matters and supports disciplined follow-through across validation cycles.
Performance Improvements
- The Product now carries updated versions of the following:
- Data Studio version 26.1.2
- Oracle Database 26.ai support in addition to existing ones
- Java version 17, 21, 25
- RAPID version 20
- OJet version 19.1
- All 3rd party libraries upgraded to latest stable versions for improved performance and security.
Known Issues and Limitations
Workspace
- When Views or Tables are selected during Data Sourcing, referenced database object dependencies are automatically replicated to the target schema. Automatically included dependencies are processed by the backend but are not displayed as selected in the user interface. Dependent synonyms are not replicated automatically and must be selected manually.
- Deletion of database tables is not synchronized between source and target schemas during Workspace Edit operations.
Data Population: Error log table creation is not supported when source tables contain LONG, CLOB, BLOB, BFILE, or ADT columns.
Datasets
- Cache operations are not supported for dataset definitions based on the Modin library.
- Profiler and drift functionality have limited support on Python 3.13 and later. Python 3.12 is the last fully qualified version stated in the supplied content.
Model Catalog: Model deployment from the Model Catalog is for test purposes. Models intended for deployment should be published from Model Pipelines.
Model Pipelines
Reports: The PDF produced by Generate Report does not include data in the output sections.
Deployment: Publishing a model from the Model Summary page does not automatically include dependent objects such as Graphs, Parameter Sets, Datasets, and Models. Publish from the Pipeline Canvas to promote dependencies.
Installer
Python 3.12 and Later with Apache Flink: Apache Flink does not provide full support for Python 3.12 and later. Installation warnings or errors related to Flink on these Python versions are expected and can be ignored.
Guardian AI Installation: The oracle-guardian-ai package is no longer installed automatically with the mmg-python library. Install and configure it separately in the appropriate Conda environment when Guardian AI functionality is required.