• qa_deck_100#slide 34 · Socio-technical governance and accountability lineage

Audit and accountability lineage is central to RAIDT because the framework is built around reconstructable accountability for one configured organisational use of generative AI. The papers argue that high-level principles, model documentation, explainability methods, and episodic audits are all useful, but they remain too coarse when an organisation must justify a specific output, recommendation, record, or communication. RAIDT therefore defines the run as the unit of governance and makes that run reviewable through a run-level evidence pack and a score profile. This is how governance moves from narrative assurance to preserved proof.

In that sense, RAIDT inherits directly from traditions of audit trails, evidence-based review, answerability, and contestability. A governance claim is not treated as credible merely because a policy exists; it must be inspectable in relation to a specific run, its configuration, its sources, and its review path. The five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability) operationalise this lineage, while the anchors 1=missing / 3=partial / 5=audit-ready specify what sufficient evidence looks like in practice. This is also why RAIDT treats influence methods as governance interventions: prompting, retrieval, adaptation, and alignment are not just engineering choices, because they shape what can later be reconstructed, challenged, and defended. Without this lineage, RAIDT would become another abstract governance vocabulary. With it, RAIDT becomes a bounded, reviewable, and comparable governance object for live organisational use.

Consider a public-service eligibility advice workflow in which a GenAI assistant drafts an answer for a citizen asking why support was refused. If the organisation only keeps the final text, it cannot show which policy clause was retrieved, which prompt template was active, whether retrieval augmentation was used, or who reviewed the answer before release. In a contested case, accountability becomes weak because the organisation can offer only narrative reassurance.

Under RAIDT, the same run would preserve a run-level evidence pack containing the prompt version, model deployment, retrieval snapshot identifiers, relevant rule text, output, timestamp, and reviewer action. The score profile could then show whether Auditability and Traceability are strong while still revealing any weaker pillar. That example shows why audit lineage is central: the organisation can reconstruct and defend one disputed use through evidence, not recollection.

• 12-RAIDT_DSR_Theory_M_v8
• 11-RAIDT_Academic_Logic_M_v11