Behavior-Driven Explainability

As modern technical systems grow in complexity, ensuring trust and maintainability becomes increasingly difficult. Traditional testing methods, while essential for verification, do not inherently provide transparency into a system's internal logic. The authors argue that explainability—the ability of a system to provide its own justifications for its actions—is a critical non-functional requirement for safety-critical and complex systems. Current approaches to self-explainability often remain abstract or lack actionable implementation steps.

Behavior-Driven Explainability (BDX) leverages the existing framework of Behavior-Driven Development (BDD). In BDD, system behavior is defined using structured scenarios (typically in the Gherkin language) consisting of 'Given' (context), 'When' (trigger), and 'Then' (outcome) steps. BDX treats these scenarios as a source of truth for generating explanations. When a user or another system queries why a specific event occurred, BDX identifies the relevant scenario and reformulates the chain of actions into a logical explanation. This approach effectively bridges the gap between high-level requirements and technical implementation.

BDX is designed to be versatile, applicable across various stages of the System Development Life Cycle (SDLC), including requirements definition, design, implementation, testing, and maintenance. By automating the derivation of explanations from specifications, BDX reduces the cognitive load on developers who would otherwise need to manually parse extensive documentation. The authors demonstrate the utility of this concept through a case study involving exception handling in a RISC-V processor, showing how it supports design-time understanding and debugging.