• workshop_dense_100#slide 7

RAIDT can be stated in one paragraph as follows. RAIDT is a run-level evidence framework for responsible governance of generative AI systems used in organisational work. It defines one configured use in context as the governed object, requires a run-level evidence pack that records the material artefacts of that use, and produces a score profile across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability). Its purpose is not to replace domain judgement or guarantee correctness, but to make governance readiness inspectable, comparable, and contestable by linking outputs to prompts, configuration, sources, checks, and oversight decisions. In this way, RAIDT moves organisations from principle-heavy assurance towards reconstructable evidence about what happened in a specific use.

That paragraph is faithful to the papers because it captures both the operational and theoretical levels. Operationally, RAIDT specifies a minimum evidence object and a scoring method interpreted through anchors 1=missing / 3=partial / 5=audit-ready. Theoretically, it treats governance as an evidence problem and formalises the framework as a mechanism-based design theory linking artefacts, context, and governance outcomes. A good one-paragraph account of RAIDT therefore has to mention purpose, unit of governance, evidence object, scoring output, and practical limits all at once.

In finance, a bank may use GenAI to draft an adverse-action explanation after a credit refusal. A one-paragraph description of RAIDT matters because it tells compliance, operations, and technical staff the same thing in a compact form: the relevant object is the specific run, the organisation must preserve a run-level evidence pack, and the resulting explanation must be assessed through a score profile rather than accepted because it sounds plausible.

That means the bank keeps the prompt template, model deployment identifier, any retrieved policy text, the output, and the review decision, then scores interpretability, traceability, and the other pillars against the evidence. If the explanation is later contested, the organisation has more than a document; it has a reconstructable governance record.

• 00-RAIDT_Wording_v2
• 08-RAIDT_Foundations_M_V50