Runtime Advocates: A Persona-Driven Framework for Requirements@Runtime Decision Support

TL;DR

RAVEN introduces an event-driven framework that transforms static, design-time personas into runtime advocate personas to support human-on-the-loop decision making in safety-critical sUAS emergency response. By grounding three advocates—Safety Controller, Ethical Governor, and Regulatory Auditor—in established standards and using an in-context LLM workflow, RAVEN delivers just-in-time, context-aware guidance as mission conditions evolve. The approach is validated through a proof-of-concept with 15 formative scenarios, demonstrating context-sensitive, domain-specific recommendations and coherent multi-advocate alignment. This work advances runtime requirements engineering by surfacing operational requirements in real time, enhancing safety, ethics, and regulatory compliance in dynamic missions, with future work focusing on broader evaluation, additional advocates, and richer world-state representations.

Abstract

Complex systems, such as small Uncrewed Aerial Systems (sUAS) swarms dispatched for emergency response, often require dynamic reconfiguration at runtime under the supervision of human operators. This introduces human-on-the-loop requirements, where evolving needs shape ongoing system functionality and behaviors. While traditional personas support upfront, static requirements elicitation, we propose a persona-based advocate framework for runtime requirements engineering to provide ethically informed, safety-driven, and regulatory-aware decision support. Our approach extends standard personas into event-driven personas. When triggered by events such as adverse environmental conditions, evolving mission state, or operational constraints, the framework updates the sUAS operator's view of the personas, ensuring relevance to current conditions. We create three key advocate personas, namely Safety Controller, Ethical Governor, and Regulatory Auditor, to manage trade-offs among risk, ethical considerations, and regulatory compliance. We perform a proof-of-concept validation in an emergency response scenario using sUAS, showing how our advocate personas provide context-aware guidance grounded in safety, regulatory, and ethical constraints. By evolving static, design-time personas into adaptive, event-driven advocates, the framework surfaces mission-critical runtime requirements in response to changing conditions. These requirements shape operator decisions in real time, aligning actions with the operational demands of the moment.

Figures (4)

• Figure 1: RAVEN leverages domain-specific Safety, Regulatory, and Ethics personas to generate contextualized runtime requirements that guide runtime decision-making by Human-on-the-loop operators.
• Figure 2: Comparison of traditional static personas used for design-time requirements specification with dynamic advocate personas supporting runtime decision-making. Static personas rely on fixed assumptions and user needs established prior to deployment, while dynamic advocate personas adapt in response to evolving mission context, environmental conditions, and operational constraints to support informed decision making.
• Figure 3: Baseline persona generation process for RAVEN.
• Figure 4: RAVEN: Our framework where the evolving world state triggers event monitoring, initiates persona selection, and delivers 0–3 advocate responses to the operator, who then adjusts drone behavior, completing the decision-feedback cycle.