RASE — verifiable agent training as MBSE plus RLVR.

Gaius introduces RASE — Rapid Agentic Systems Engineering — a Python-native metamodel with SysML v2 semantics. Four coupled sub-models hold the pieces:

• OSM — Operational Scenario Model. BDD scenarios as executable specifications, given/when/then registered against step decorators.
• SSM — System State Model. NiFi (and adjacent systems) as a typed graph with declarative constraints over its structure.
• UOM — UI Observation Model. Set-of-Mark and Trace-of-Mark for grounding agent actions in the rendered UI.
• VM — Verifier Model. The RLVR oracle and reward computation, with the load-bearing constraint that the oracle uses ground truth — never UI observations.

UI traces are the training target; verification provides the reward signal. The discipline behind self-improving agents is not chains of prompts — it is a digital thread from requirement to verdict. RASE makes the thread explicit.