• workshop_dense_100#slide 5

The ring-by-ring prioritisation logic is evidential dependency, not presentational convenience. The first priority is always to stabilise the governable object: the run as the unit of governance. The second is to define the minimum run-level evidence pack that makes that run reconstructable. The third is to assess the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability) through a score profile rather than through narrative impression. Only after those three inner layers are clear does it make sense to discuss how influence methods as governance interventions may improve or weaken governance readiness.

This ordering is necessary because each ring presupposes the previous one. Without a bounded run, there is nothing specific to govern. Without the run-level evidence pack, the five pillars cannot be scored credibly. Without a score profile using anchors 1=missing / 3=partial / 5=audit-ready, intervention claims remain speculative: one cannot say that retrieval augmentation improved Traceability, or that preference-based alignment improved Responsibility, unless the relevant evidence was preserved and scored. Implementation therefore comes after the inner logic is established, because automation depth, logging design, and review workflows are means of sustaining the evidence architecture rather than substitutes for it. Policy and empirical work come later still. They translate and test the same inner design; they do not replace it. The priority logic is therefore cumulative: define, evidence, score, intervene, implement, translate, then evaluate more widely.

Consider cybersecurity alert triage. A team may be tempted to discuss whether a stacked configuration is better than baseline prompting. Under RAIDT, that is not the first question. The first question is whether one alert-triage run can be reconstructed: the prompt, retrieval context, tool outputs, recommendation, hashes, and analyst review. The second is whether the score profile shows weak Dependability or weak Traceability.

If Dependability is low because repeat runs vary sharply, the team can then test structured prompts, retrieval grounding, or preference constraints as influence methods as governance interventions. If those methods improve the score profile and the evidence remains complete, the team can move to implementation and policy. The rings therefore enforce the right order of argument.

• 08-RAIDT_Foundations_M_V50
• 12-RAIDT_DSR_Theory_M_v8