• integrated_82#Q2.11

RAIDT relates to auditing and accountability traditions by translating their core logic into a run-level governance method for generative AI. Audit traditions seek inspectable trails, stable evidence objects, integrity of records, and disciplined sampling or review. Accountability traditions seek answerability: who relied on the system, on what basis, under which controls, and with what route for challenge. RAIDT aligns with both traditions by making one configured use event inspectable rather than leaving governance at the level of policy statements, static documentation, or periodic attestations.

More specifically, RAIDT complements rather than replaces model cards, explainability tools, and internal audit routines. It provides the bounded object those practices often lack: a run-level evidence pack that can be sampled, compared, escalated, and reviewed after the fact. The accompanying score profile across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability) keeps governance multi-dimensional rather than collapsing judgement into one opaque metric. The anchors 1=missing / 3=partial / 5=audit-ready give auditors, managers, and compliance teams a shared way to judge whether evidence is sufficient for reconstructability and contestability. This makes RAIDT especially useful in environments where governance expectations are increasingly expressed through documentation, record-keeping, oversight, and traceability duties. In short, RAIDT stands in the lineage of auditing and accountability, but it updates that lineage for stochastic, configured, and workflow-embedded GenAI use by treating the run as the unit of governance.

Consider an HR shortlist justification workflow reviewed months after a candidate complaint. A conventional audit might find policy documents, model-level descriptions, and general hiring controls, yet still struggle to reconstruct the specific run that produced the shortlist rationale. The missing details may include the prompt template, scoring criteria, model deployment, adapter version, and the reviewer who approved the output.

With RAIDT, internal audit can sample that run as a transaction-like governance object. The run-level evidence pack shows the exact configured use, while the score profile highlights whether the workflow was merely partially documented or genuinely audit-ready. The organisation can then test preserved controls against expected controls, rather than relying on broad assurances that the recruitment system was governed in general.

• 11-RAIDT_Academic_Logic_M_v11
• 17-RAIDT-Sociotechnical_M_v6