• qa_deck_100#slide 6 · What RAIDT is and what it governs

In RAIDT, a run is one configured use of a generative AI system for a specific task, at a specific time, in a specific organisational context. The definition is intentionally concrete. A run includes the prompt or instruction, the active model and tool configuration, any retrieved context, the output produced, and the human or automated checks that follow. RAIDT therefore treats the run not as a vague interaction, but as a bounded governance object whose contents can later be inspected.

The reason RAIDT defines the run so carefully is that GenAI risk materialises in use, not only in model design. Small changes in prompt structure, retrieval context, decoding settings, adapter versions, or oversight steps can materially alter outputs even when the underlying model remains the same. For that reason, RAIDT adopts the run as the unit of governance. A run is the level at which an organisation can show what actually happened, under which controls, and with what consequences for review, challenge, or improvement.

Once captured as a run-level evidence pack, the run becomes a reconstructable record rather than a transient exchange. That record is then assessed across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability). In this sense, a run is both the immediate act of configured GenAI use and the evidential unit from which a RAIDT score profile can be produced. The framework's central claim is that responsible organisational use depends on making each material run inspectable rather than relying only on policy statements or model-level documentation.

In the healthcare vignette, a clinician uses GenAI to summarise a chest-pain consultation into structured headings such as symptoms, treatment, and red flags. In RAIDT terms, that single configured use is the run. The run includes the exact prompt, the rule that unsafe invention is banned, the active model and any alignment controls, the generated summary, and the clinician's follow-on review before the note is used.

This matters because a later reviewer may need to know whether uncertainty about pending tests was stated, whether escalation cues were requested, and whether high-risk use was flagged for human oversight. If those elements are preserved in the run-level evidence pack, the clinic can reconstruct and evaluate the run; if only the final summary is kept, the run cannot be properly governed.

• 00-RAIDT_Wording_v2
• 08-RAIDT_Foundations_M_V50