• qa_deck_100#slide 28 · Design science and mid-range design theory

In RAIDT, design science does not primarily produce another abstract principle set, maturity model, or model-level document. It produces a bounded governance artefact for material GenAI use: the "run-level evidence pack" plus a "score profile" for the "five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability)". Because RAIDT defines the "run as the unit of governance", the design object is one configured use in context, captured so that it can later be reconstructed, reviewed, compared, and challenged. In that sense, design science produces a governance-ready proof object rather than narrative assurance alone.

The papers also show that this artefact is operational rather than merely conceptual. RAIDT includes the run record schema, evidence-pack checklist, reviewer rubric, prompt registry templates, and related policy crosswalks that make governance inspectable in practice. The score profile translates evidence into a comparable governance view through the standard "anchors 1=missing / 3=partial / 5=audit-ready". This means that governance readiness is assessed from preserved evidence, not from aspiration, and that weak Responsibility or Traceability can be identified at the level where risk actually materialises.

At the same time, RAIDT design science produces explanatory knowledge. The DSR contribution is formalised as a mechanism-based mid-range design theory explaining how governance artefacts and "influence methods as governance interventions" shape observable run-level outcomes under stated boundary conditions. So, what design science actually produces in RAIDT is dual: a concrete artefact set for governing one run, and a theory that explains why those artefacts should make organisational GenAI use more reviewable, comparable, and contestable.

In a finance setting, a bank uses GenAI to draft a credit adverse-action explanation for a customer. Under RAIDT, the relevant design output is not just a policy saying explanations should be fair or clear. The bank creates a run-level evidence pack for that specific run: the prompt template used, the model and deployment identifiers, any documented decision criteria, the generated explanation, reviewer notes, and the approval or escalation decision. It then scores the run across the five pillars using the score profile.

If the explanation is clear but the supporting criteria or provenance are weak, the score profile will expose that imbalance rather than hide it inside a general compliance statement. Managers or auditors can then see whether structured prompting, retrieval support, or stronger review steps are needed. If the customer later challenges the explanation, the organisation can inspect the preserved run as the unit of governance, rather than relying on a generic model card or staff recollection.

• 12-RAIDT_DSR_Theory_M_v8
• 11-RAIDT_Academic_Logic_M_v11
• 17-RAIDT-Sociotechnical_M_v6