• qa_deck_100#slide 13 · Boundaries, scope, and limitations
• integrated_82#Q1.6

RAIDT is not a prompt-engineering study and it is not a RAG system. The papers consistently treat prompting, retrieval augmentation, PEFT/LoRA, and preference-based alignment as influence methods as governance interventions rather than as the framework itself. In other words, RAIDT does not ask whether one technical method is best in the abstract; it asks what evidence a particular configuration produces, whether that evidence can be reviewed later, and how that changes governance readiness for a specific run. Retrieval, for example, may improve Traceability or Auditability, but only if the retrieved context is preserved in the run-level evidence pack.

RAIDT is also not an auditability-only framework, not a model card, and not a generic checklist. Its object is the run as the unit of governance, and its assessment spans the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability). A model card may describe a model in general terms, but it does not show what happened in one configured organisational use. Likewise, a generic checklist may list desirable controls, yet RAIDT requires a scored, reconstructable evidence object for the run itself and retains a score profile so trade-offs across pillars remain visible.

Finally, RAIDT is not a guarantee of factual correctness, safety, or legal compliance. The papers are explicit that scoring does not replace domain judgement or sector procedures. Instead, RAIDT provides a disciplined way to inspect and compare governance readiness through a run-level evidence pack, a score profile, and anchors 1=missing / 3=partial / 5=audit-ready.

Consider a public-service team using a language model to explain eligibility rules to citizens. If the team says, 'we use RAG, so the system is governed', RAIDT would reject that claim. RAG is only one technical mechanism. The governance question is whether the run-level evidence pack records the exact prompt, model version, retrieval query, retrieved passages, document identifiers, hashes, output, and review steps for that specific case.

If those artefacts are missing, the team may have a technically grounded answer but still have weak Auditability and Traceability. A later complaint would be difficult to investigate because the organisation could not reconstruct what rule version informed the advice. RAIDT therefore shows what it is not: not a synonym for RAG, not a prompt recipe, and not a compliance badge. It is a run-level evidence method for making a contested use inspectable and governable.

• 00-RAIDT_Wording_v2
• 08-RAIDT_Foundations_M_V50
• 11-RAIDT_Academic_Logic_M_v11