1. 26.04.2 Release
These release notes describe the changes introduced in Foretellix products in version 26.04.2. Features are classified as New, Changed, or Deprecated. "Changed" indicates either interface modifications or behavioral changes to existing functionality.
1.1 Release Highlights
- Multiple V-Plan views can now be created, named, and stored within the Foretify Manager workspace. Users can switch between different V-Plan configurations to explore coverage results from multiple perspectives without losing their customized layouts — streamlining coverage analysis and reporting for complex test suites.
- A new Safety metric, Deceleration Rate to Avoid Crash (DRAC), has been added to the metric package.
- Coverage bucket visualizations in Foretify Manager can now be displayed as either a list or a histogram, giving users the flexibility to choose the representation best suited to their analysis needs. This enhancement makes it easier to interpret coverage distribution patterns across large scenario populations.
1.1.1 Backward Compatibility
Compatibility Guidance for Foretellix Tool Versions
To ensure full support and optimal performance, we strongly recommend updating both Foretify and Foretify Manager to the same version. Backward compatibility between versions cannot always be guaranteed, although we are trying to keep it for two releases. When data is uploaded to Foretify Manager, it is automatically validated for compatibility.
Please note that forward compatibility—using results from newer versions of Foretify in older versions of Foretify Manager—is intentionally not supported.
Key Non-Compatible Changes in Recent Foretellix Releases
- The previous off_road-checker functionality has been reinstated within the Global Evaluator library and is now available as a default evaluator named road_departure for all Foretify users. As a result of this, user action is required in case any user-defined code has been using or referencing this checker:
- As a result of this change, any KPIs, coverage metrics, post-processing scripts, test plans, or result analysis that reference off_road will no longer match.
- Update all references from off_road to road_departure in scenario files, configurations, and analysis tooling.
Advance Notice for Upcoming Changes in 26.05
- The plain_object_group_position_along_road and position_along_road modifiers have been updated to be traffic-side agnostic. In 26.05, the old leftmost_lane and rightmost_lane will be deprecated. Please update to innermost and outermost.
- The following vehicle physical checks are deprecated and will be removed in release 26.05:
- speed_physical
- acceleration_physical
- turn_physical
- Instead, use Kinematic Behavior Monitoring.
- The following checkers are deprecated and will be removed in release 26.05:
- jitter_policy
- minimum_highway_vru_distance_policy
- unplanned_standing_time_policy
- car_does_not_move
- Instead, use Static Object Too Close.
- The following vehicle.policy fields are deprecated and will be removed in release 26.05:
- may_jitter — Instead, use [Unplanned Standing Checker][unplanned-standing].
- unplanned_standing_time — Instead, use [Unplanned Standing Checker][unplanned-standing].
- minimum_highway_speed — Instead, use [Slow Driving Checker][slow-driving-checker].
- minimum_highway_vru_distance — Instead, use [Static Object Too Close][static-object-too-close].
- too_close_TTC — Instead, use [Static Object Too Close][static-object-too-close].
- max_does_not_move_time — Instead, use [Unplanned Standing Checker][unplanned-standing].
- allow_no_movement — Instead, use [Unplanned Standing Checker][unplanned-standing].
- min_total_distance_to_indicate_movement — Instead, use [Unplanned Standing Checker][unplanned-standing].
- The following kinematic check infrastructure items are deprecated and will be removed in release 26.05:
- kinematic_behavior watcher modifier (ftlx_kinematic_checker.osc) — Instead, use Kinematic physical checker.
- DEPRECATED_KINEMATIC_ISSUE_REASON field (ftlx_global_eval_config_top.osc) — Instead, use the new kinematic checker config.
- The following vehicle_checker_params fields are deprecated and will be removed in release 26.05:
- vehicle_checker_params.car_jerk_check_flag
- vehicle_checker_params tolerance field
- vehicle_checker_params.enable_physical_acceleration_check
- vehicle_checker_params.enable_policy_acceleration_check
- Instead, use the kinematic physical checker.
Please review all Features marked as [Changed] in this release for further details on backward compatibility.
1.2 Evaluation V-Suite
Extended the evaluation V-Suite with coverage metrics for additional L0 scenario nodes, expanding behavioral coverage across the drive_in_lane, nav_intersection, and nudge scenario families.
- Added new scenario family files for drive_in_lane, nav_intersection, nudge, and their base variants, with coverage structure reorganized per L0 node for clearer separation across behavioral categories.
- Introduced new coverage items applying to all scenario families and a centralized enum configuration consolidating shared enums across the V-Suite.
- Added a new change_lane evaluation example with XODR map and run script to accelerate onboarding; additional L0 nodes will follow in subsequent releases.
See NVIDIA Evaluation Scenarios.
Added a Deceleration Rate to Avoid Crash (DRAC) metric, which measures the deceleration a vehicle would need to apply to avoid a collision.
- Provides V&V engineers with a standardized, objective measure of safety margins that can be tracked consistently across test campaigns.
- Also, an overview page for all Safety metrics has been added to the documentation.
See vehicle.get_drac().
Extended the scenario library with two evaluation scenarios:
- Vehicle Cut-Out Behind Ego detects a following vehicle that cuts out from behind the Ego with the intention of overtaking. See Vehicle Cut-Out Behind Ego.
- Adjacent Vehicle detects a vehicle driving in the lane adjacent to the Ego. See Adjacent Vehicle.
Added platform capabilities required to fully implement the JAMA 58 logical scenario library used as a safety assessment standard by Japanese automakers.
- Added detection of Merge and Branch road geometry zones.
- Corrected the vehicle.veering_to_opposite_lane and vehicle.oncoming modifiers to properly support left-hand traffic (LHT) configurations.
See vehicle.veering_to_opposite_lane and vehicle.oncoming.
The previous off_road-checker functionality has been reinstated within the Global Evaluator library and is now available as a default evaluator named road_departure for all Foretify users.
road_departure detects when the Ego leaves the drivable road surface and is active in both generative and evaluation flows. As a default evaluator, it is enabled automatically, allowing V&V engineers to monitor road-departure behavior across all scenarios without requiring custom checks.
Note - this is a non-backwards compatible change - Action required:
- As a result of this change, any KPIs, coverage metrics, post-processing scripts, test plans, or result analysis that reference off_road will no longer match.
- Update all references from off_road to road_departure in scenario files, configurations, and analysis tooling.
1.3 Foretify
Enhanced the error handling capabilities within the evaluation pipeline to provide clearer and more actionable error messages.
- Error messages now include detailed context on the origin and reason for failures, enabling users to identify and resolve issues without requiring support engagement.
- Workflow failures at the interval-creation level are automatically surfaced in the UI, dashboards, and logs to improve error transparency.
- When an interval fails to contribute to coverage, the system provides a clear explanation covering scenarios such as parsing errors, missing data fields, or invalid extracted scenarios.
For more information, refer to the user documentation: Error Classification, Severity Levels, and SLA Escalation Rules.
- Physical parameters that were represented by integer semantics have been migrated to real semantics.
- There is no change to SDK apps APIs.
- There might be some small changes in physical parameters and numerical representation due to more accurate real semantics.
1.4 Foretify Manager
Added a DataSourceType attribute to all run data ingested into Foretify Manager, enabling the system to distinguish between real-world and synthetic simulation data at the run level.
- V&V engineers can query and filter runs by data source type, providing a consistent way to analyze results across heterogeneous data pipelines.
- This is Phase 1; UI-level preset filters for data source type will be added in a subsequent release.
See Run attributes.
- Users can now add synthetic pedestrians and cyclists.
- Their trajectory can be composed of multiple line segments, each travelled at a constant speed.
- Use new Custom Actions to import or export multiple ClipGT files at once.
- Imported clips are grouped into a single test suite result, while batch export creates one ZIP file with all exported clips.
Added a License Usage Report API to Foretify Manager for real-time monitoring of license consumption across a deployment.
- Exposes a REST endpoint and Python SDK access returning total, used, and available license counts per feature, including active checkout details such as username, hostname, and start time.
- Enables automated tracking of license consumption to optimize allocation and identify usage patterns.
Added the ability to sort coverage metric items within a coverage group in the VPlan tree by Name or Grade in ascending or descending order.
- Sorting options are available directly in the VPlan tree header for each coverage group, improving navigation and readability for complex VPlans.
- The selected sort order persists across sessions, ensuring a consistent view for each user.
See Sort display of metric items.
Added the Analysis page to Foretify Manager, providing a consolidated view of coverage and performance trends over time.
- Helps V&V engineers and managers monitor verification progress and quickly identify areas that require attention.
- Displays trend data across test suite runs, confirming that the project is moving in the correct direction.
See Analyzing Test Suite Results.
- New layout option maximizes the visualizer and tucks away detailed panels, which can be expanded on demand.
- Layout switching is now preset-based, so you can always return to a known layout.
- Admins can set the default layout for new users, but each user’s layout configuration still persists in their preferences.
Extended VPlan Views in Foretify Manager with the ability to assign a custom weight to any node in the VPlan tree on a per-view basis.
- Users can increase, decrease, or set a node’s weight to 0 to remove that node and its entire subtree from the coverage grade calculation, without affecting the shared base VPlan or other users’ views.
- Weight adjustments are stored per VPlan view, allowing different teams and stakeholders to maintain independent grading perspectives on the same test plan data.
See VPlan views.
Improved the naming of Test Suite Results (TSRs) launched from the Scenario Designer to use the format <scenario_name>_<timestamp> instead of the previous <flow_definition_name>_<long_UID_hash>. This helps distinguish run results of different scenarios executed with the same flow definition.
Added a collapse/expand toggle to the Trends section in the Workspace test suite results view in Foretify Manager.
- Users can hide the trends charts to reclaim vertical space and focus on the test suite results table below.
- The toggle provides a cleaner working area during detailed result analysis without navigating away from the view.
See Analyzing Test Suite Results.
The Compound Intervals Playground is now accessible while a compound interval rule is being applied in the workspace, removing the previous restriction that blocked playground actions until the apply operation finished.
Users can continue exploring temporal relations, tweaking filters, and saving new compound interval rules in parallel with an in-progress apply; a clear indicator is shown when rule application is already running.
The compound interval rules status table now displays each rule's save date and apply date side-by-side, making it easy to see at a glance whether a rule's applied results are up-to-date with its latest saved definition.
- Replaced the single Signals toggle in Visualizer's View Tools configuration with independent Traffic lights, Stop signs, and Yield signs toggles.
Changed the coverage bucket visualization in Foretify Manager so that the selected display mode — list view or histogram — is saved and restored automatically across sessions.
- Histogram view is now the default, providing immediate visual insight into coverage distribution upon first use.
- Users can switch between views at any time; their preference is restored on the next login without requiring reconfiguration.
See View buckets as histogram chart.
- Previously, you could only override settings that already existed in the selected Environment Settings.
- Now, enable Show all environment settings to view and modify any available setting, including specifying settings not yet configured.
Moved buttons that trigger long-running operations in the VPlan viewer into a three-dots menu to prevent accidental execution during analysis.
- Actions that could degrade system performance or produce inconsistent results are no longer exposed as primary buttons.
- Users must explicitly open the menu to initiate these operations, reducing the risk of unintended side effects.
See VPlan viewer.
- You can now save your progress, reopen tests later for editing, and update them as needed with a standard save-and-load workflow.
- Tests created in earlier Foretify Manager versions remain executable, but they cannot be reopened for editing.
Foretify Manager will support PostgreSQL up to v18.x (instead of up to v13.x previously).
1.5 Evaluation Pipeline
- The parameter “Confidence of detection” estimates the level of certainty that an object was correctly detected.
- It is composed of 3 different parameters:
- Detection - Confidence in whether the object was detected.
- Classification - Confidence in the object’s classification.
- Localization - Confidence in the object’s location.
For more information, refer to the user documentation.
- Stationary vehicle extrapolation automatically extends detected stationary vehicles (primarily parked cars) across the full clip duration.
- This enables the generation of ground truth data, including data that was not fully recorded by the Ego (such as occluded vehicles and VRUs), particularly when modifications are applied to create new scenarios that were not part of the original recording.
For complete documentation, see Viewing the ingested scenario.
1.6 Known Limitations
Foretify Manager
A future fix will add a warning to inform users about potential data loss.
- Attempting to export ClipGT from the Runs page will fail with the error: "Operation failed. Could not find required flow definition."
- As a workaround, export ClipGT from the Triage page.
The Kubernetes dispatcher and orchestrator currently only support 1 replica.
Only one Orchestrator pair is supported per Foretify Manager. Each Orchestrator should use its own namespace.