• workshop_dense_100#slide 15

A run in RAIDT is one configured use of a GenAI system for a defined task in context at a specific time. It includes the instructions given, the active model and tool settings, any retrieved passages or tool outputs, the generated response, and the checks, oversight, or safety controls applied afterwards. RAIDT therefore treats the run as the unit of governance because this is the point at which organisational obligations, technical configuration, and actual output meet.

The practical significance of this definition is that RAIDT turns each material use into a run-level evidence pack and then into a score profile. The run is not assessed by impression. It is assessed against the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability), using anchors 1=missing / 3=partial / 5=audit-ready. This allows an organisation to compare different prompts, retrieval designs, or oversight arrangements on common governance criteria while retaining visible trade-offs between pillars rather than collapsing everything into a vague trust claim.

The concept matters because GenAI behaviour is shaped at run time. RAIDT therefore treats influence methods as governance interventions: retrieval augmentation can improve traceability only if snapshots are preserved; structured prompting can improve interpretability only if the prompt and output schema are logged; adaptation or alignment can improve fit or safety only if versions and controls are recorded. In short, the run matters because it is the smallest organisationally meaningful evidence object from which reconstruction, contestability, and learning can proceed.

In the finance vignette, a bank uses GenAI to draft an adverse-action explanation after a credit refusal. The run consists of the constrained prompt template, the model version, the documented decision criteria, the generated explanation, and the review step checking that the explanation matches recorded reasons rather than inventing new ones.

RAIDT matters here because an explanation can sound persuasive while still being unfit for governance. By preserving the run-level evidence pack, the bank can show which criteria informed the text, whether uncertainty or recourse was stated, and whether the explanation remained tied to the decision record. That evidence supports a score profile across the five pillars and makes the explanation contestable if a customer later challenges it.

• 00-RAIDT_Wording_v2
• 11-RAIDT_Academic_Logic_M_v11