Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Four Coupled Models

The RASE metamodel consists of four tightly coupled models. They form a coherent verification framework where changes to one model often require updates to others.

Coupling Matrix

If you change…Also update…
SSM (system state)VM constraints that reference state structure
OSM (scenarios)VM requirements derived from scenarios
UOM (marks/traces)VM verification cases that consume traces
VM (verification)Ensure reward strategies align with constraint semantics

SSM – System State Model

The SSM represents the system under test as a typed graph. The primary domain is NiFi, modeled as NiFiInstance containing ProcessorGroup, Processor, FlowConnection, and ControllerService nodes.

from gaius.rase.domains.nifi import NiFiInstance, Processor, ProcessorGroup

state = NiFiInstance(
    root_group=ProcessorGroup(id="root", name="NiFi Flow", processors=[
        Processor(id="abc", name="GetFile", type="org.apache.nifi.GetFile"),
    ])
)

SSM constraints are declarative, composable, and immutable. Examples: ProcessorExists, AllProcessorsRunning, NoBackpressure, FlowIsEquivalent. Compose with AllOf, AnyOf, Not.

OSM – Operational Scenario Model

The OSM captures BDD (Behavior-Driven Development) scenarios as executable specifications. Each scenario is a sequence of Given/When/Then steps that map to SysML v2 action definitions.

from gaius.rase.osm import Scenario, StepType, StepUsage

scenario = Scenario(
    name="CreateBasicFlow",
    steps=[
        StepUsage(step_type=StepType.GIVEN, text="NiFi is running"),
        StepUsage(step_type=StepType.WHEN,  text="I create a processor group named 'ETL'"),
        StepUsage(step_type=StepType.THEN,  text="the group 'ETL' exists"),
    ],
)

Step definitions (StepDef) are reusable patterns with {param} placeholders. The StepRegistry maps step text to executable actions via @given, @when, @then decorators.

UOM – UI Observation Model

The UOM provides grounding between language and UI actions using two complementary structures:

  • SoM (Set-of-Mark): A ScreenshotWithSoM annotates a screenshot with numbered Mark objects, each with a BoundingBox, UIRole, and optional mapping to an SSM element.
  • ToM (Trace-of-Mark): A TraceOfMarks records a sequence of ActionFrame entries (click, type, scroll) referencing marks by number, forming the agent’s action trajectory.
from gaius.rase.uom import Mark, BoundingBox, PixelCoord, UIRole

mark = Mark(
    mark_id=1,
    bbox=BoundingBox.from_xywh(100, 200, 50, 30),
    ui_role=UIRole.BUTTON,
    label="Add Processor",
)

The SoM/ToM pattern enables precise UI grounding: agents reference elements by mark number rather than pixel coordinates.

VM – Verifier Model

The VM implements RLVR verification. It connects OSM scenarios to executable verification cases with oracle-based reward computation. See Verification for full details.

Key components:

  • Requirements: StepRequirement (atomic, from a BDD step) and ScenarioRequirement (composite, grouping steps with invariants)
  • Verification Cases: APIVerificationCase (ground truth via API) and UIVerificationCase (agent UI actions, final state checked via API)
  • Oracle: NiFiOracle queries the NiFi REST API for authoritative state verification
  • Reward Strategies: BinaryReward (sparse) and GradedReward (partial credit)