• integrated_82#Q1.1

RAIDT produces two practical outputs for important GenAI use. The first is the run-level evidence pack, which preserves the structured evidence of one run: identifiers, prompts, configurations, retrieved context where relevant, outputs, checks, review actions, and governance linkage. The second is the RAIDT score profile for that same run. Together, these outputs allow an organisation to move from broad responsible-AI aspirations to inspectable, comparable proof about how a system was actually used. The evidence pack answers the question, ?What happened here??; the score profile answers, ?How governable was this run against RAIDT criteria??

The score profile is expressed through the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability), with judgement anchored in evidence rather than rhetoric. The scoring logic uses anchors 1=missing / 3=partial / 5=audit-ready, while preserving the full profile so that trade-offs remain visible rather than disappearing into a single average. This is important because a run may be strong on interpretability but weak on traceability, or robust on auditability but limited by poor oversight controls. The practical value of RAIDT lies precisely in producing these paired outputs: an auditable evidence object and a comparable governance measurement. They support monitoring, audit sampling, procurement review, incident investigation, and continuous improvement across repeated runs and changing influence configurations.

In cybersecurity, a security analyst may use GenAI to triage an alert and recommend next steps. RAIDT would first produce a run-level evidence pack containing the prompt, approved threat-intelligence retrieval snapshot, model configuration, output hash, and any reviewer notes. It would then produce a score profile showing how that run performed across the five pillars, especially Dependability, Auditability, and Traceability.

If repeated runs reveal unstable recommendations, the evidence packs make the variance visible and the score profile records that weakness formally under documented controls. The team can then adjust prompts, retrieval, or oversight controls and compare later runs on the same basis.

• 00-RAIDT_Wording_v2
• 08-RAIDT_Foundations_M_V50