• qa_deck_100#slide 29 · Design science and mid-range design theory

RAIDT is framed as a mid-range design theory because its explanatory ambition is deliberately bounded. It does not seek a universal account of all AI governance, law, or ethics; instead, it addresses one defined class of problems: organisational GenAI runs whose outputs may shape records, recommendations, decisions, or communications and must later be reconstructed, compared, and challenged. In that setting, RAIDT defines the run as the unit of governance and supplies a designable governance object rather than another principle set. The run-level evidence pack and the score profile across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability) make governance readiness inspectable at the point where risk materialises.

It is equally not a checklist. A checklist can verify the presence or absence of controls, but RAIDT is meant to explain how governance artefacts and influence methods as governance interventions affect observable outcomes under stated boundary conditions. That is why the design-theoretic layer matters: it specifies constructs, mechanisms, propositions, and limits, so claims can be tested cumulatively across domains and configurations. The anchors 1=missing / 3=partial / 5=audit-ready operationalise the score profile without collapsing governance into a single compliance tick-box. Mid-range design theory therefore gives RAIDT the right level of abstraction: concrete enough to guide design and audit of real runs, but theoretical enough to explain why some governance configurations produce stronger evidence, reviewability, and contestability than others.

In a hospital discharge-summary workflow, a trust may already have a responsible-AI policy and a clinical checklist. Neither is enough if a patient later challenges a summary that omitted a contraindication. RAIDT would require a run-level evidence pack for that specific use: the prompt template, retrieved guideline excerpt, model and tool settings, output, reviewer notes, and sign-off. The score profile would show whether the run was strong or weak across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability), using anchors 1=missing / 3=partial / 5=audit-ready.

That is more than a checklist because the hospital can compare configurations. If structured prompting and retrieval improve Traceability but weak reviewer capture leaves Auditability at 3, the organisation learns where governance failed and which intervention should change.

• 12-RAIDT_DSR_Theory_M_v8
• 11-RAIDT_Academic_Logic_M_v11