• workshop_dense_100#slide 1

In a 100-slide workshop edition, RAIDT should be taught as a sequenced governance architecture rather than as a loose collection of ethics themes. The workshop should begin with the problem statement developed in the foundations and academic logic papers: model-level documentation and policy statements are insufficient because important GenAI risk materialises at run time. From there, the next module should define RAIDT, explain run as the unit of governance, and establish the run-level evidence pack as the standard proof object for material uses. Only after that conceptual grounding should the workshop move into the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability), showing how each pillar is evidenced rather than merely asserted.

The middle of the workshop should then cover scoring and implementation. Participants should learn how the score profile is produced, why the five-pillar profile must remain visible even when a composite is reported, and how the anchors 1=missing / 3=partial / 5=audit-ready discipline reviewer judgement. A further block should cover influence methods as governance interventions, emphasising that structured prompting, retrieval augmentation, PEFT or LoRA, and preference-based alignment alter both behaviour and evidentiary requirements. The final workshop section should use the papers' sector vignettes to connect the framework to healthcare, public service, finance, HR, and cybersecurity, then close with standards alignment, monitoring, and organisational embedding. In short, the 100-slide edition should mirror RAIDT's logic from concept to evidence to scoring to implementation.

A strong workshop exercise would use the cybersecurity alert-triage vignette. Early slides would present the task and the risk of hallucinated indicators. Mid-workshop slides would ask participants to inspect a candidate run-level evidence pack containing the prompt structure, retrieved threat-intelligence snapshot, output, and review checks. They would then assign a score profile across the five pillars and discuss why dependability depends on repeat-run stability rather than on a single convincing answer.

Later slides would vary the influence configuration by removing retrieval snapshots or changing the prompt constraints. Participants could then see how governance readiness changes when evidence is weakened. This makes the workshop practical: RAIDT is learned not as abstract doctrine, but as a repeatable way of reconstructing, comparing, and governing configured uses.

• 08-RAIDT_Foundations_M_V50
• 11-RAIDT_Academic_Logic_M_v11