• integrated_82#QC3

The project should be read from the centre outward because RAIDT begins with the smallest governable object, not with abstract policy language. The centre is the claim that the run as the unit of governance: one configured use of a generative AI system for one task, at one time, in one context. Once that object is fixed, the next question is what must be preserved so that the run is inspectable later. That is why the next layer is the run-level evidence pack: prompts, configuration digests, retrieval snapshots where relevant, hashes and timestamps, outputs, and review logs. The papers consistently argue that governance becomes credible only when those artefacts are preserved as bounded, reviewable records rather than free-text assurances.

From there, the reading moves to the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability) and the score profile built from the evidence pack. The score profile is not a decorative summary. It is the analytical bridge between raw evidence and governance judgement, using anchors 1=missing / 3=partial / 5=audit-ready so that reviewers can compare runs without collapsing trade-offs into a single opaque claim. Only after that inner logic is clear should a supervisor move outward to influence methods as governance interventions, implementation choices, standards and policy translation, and finally the wider empirical programme. This centre-outward order mirrors the design-theory argument: first define the governable object, then define evidence, then define outcomes, then examine mechanisms, operationalisation, and cumulative research.

In a healthcare note-summarisation case, a supervisor should not begin by asking whether the hospital has an AI policy. The first question is what happened in one run: which prompt template was used, whether uncertainty and escalation wording were required, whether any retrieval source was used, and what the clinician reviewed before relying on the summary. That is the centre.

Only once that run-level evidence pack exists should the supervisor inspect the five pillars and the score profile. If Responsibility and Interpretability look reasonable but Auditability is weak because prompts or review logs were not preserved, the next conversation moves outward to the intervention and implementation layers: stronger prompt controls, logging, and review capture. Policy mapping becomes meaningful only after those inner layers are visible.

• 08-RAIDT_Foundations_M_V50
• 12-RAIDT_DSR_Theory_M_v8