• qa_deck_100#slide 8 · What RAIDT is and what it governs

RAIDT represents governance readiness as an evidence-based property of a specific use event, not as a generic statement about a model or policy. The framework works with the run as the unit of governance, so the central object is the run-level evidence pack for one configured task in context at a specific time. That pack records what later reviewers need in order to reconstruct and assess the run: prompts and templates, model and tool configuration, retrieved context where used, outputs, checks, version identifiers, and oversight actions. RAIDT then expresses the condition of that run through a score profile across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability). In this way, governance readiness becomes inspectable, comparable, and reviewable rather than merely asserted.

The representation is deliberately structured. Each pillar is assessed against anchors 1=missing / 3=partial / 5=audit-ready, with the run-level evidence pack as the scored object. A composite score may be reported, but RAIDT retains the full pillar profile because governance trade-offs matter; a run may be relatively interpretable yet still weak on auditability or traceability. The papers also emphasise influence methods as governance interventions, because prompting, retrieval augmentation, PEFT or LoRA, and alignment controls change both system behaviour and the evidence available for review. RAIDT therefore represents governance readiness as scored, reconstructable evidence about how GenAI was actually used, rather than as narrative assurance or a claim of factual correctness.

In a public-service eligibility case, a staff member uses GenAI to interpret a rule for a claimant. Under RAIDT, the organisation does not rely on the answer alone. It builds a run-level evidence pack containing the prompt, model version, retrieval snapshot of the exact policy passages, output text, and any review notes. The resulting score profile can then show whether Auditability and Traceability are genuinely strong, or whether the advice only appears well grounded.

If the retrieval snapshot is missing, the same fluent answer would score much lower because later reviewers could not prove which rule version informed the advice. RAIDT thus represents readiness through what can be reconstructed and checked after the fact, not through surface quality alone.

• 00-RAIDT_Wording_v2
• 11-RAIDT_Academic_Logic_M_v11