• integrated_82#Q1.3

The run is the unit of governance in RAIDT because governance becomes operational only when an organisation can answer for one configured use event. The audit and accountability paper shows that answerability, reconstructability, and contestability all depend on preserved records of what actually happened in a specific case. For generative AI, that requirement is unusually pressing because outputs are materially shaped at run time. Even when the same underlying model is reused, variations in prompts, retrieval, tools, settings, and human review can change what was produced and whether the output was suitable for the task.

RAIDT therefore moves governance downward to the level where risk materialises and where disputes are adjudicated. A run-level evidence pack serves as the bounded proof object for one configured use, and the associated score profile translates governance quality into inspectable evidence rather than narrative claims. By scoring the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability) with anchors 1=missing / 3=partial / 5=audit-ready, RAIDT makes it possible to compare runs, identify recurring weaknesses, and support post-incident review. This is why run as the unit of governance is not a rhetorical choice but a methodological one: the run is the smallest organisationally meaningful object that can link context, configuration, output, and oversight in a form that can later be reviewed and challenged.

An HR manager uses a GenAI assistant to draft a performance appraisal, and the employee later disputes the justification that entered the personnel record. A model card may show the model's broad intended use, but it will not reveal which prompt template version was used, whether policy text was retrieved, what output was generated, or what review the manager performed before filing the appraisal.

A run-level evidence pack preserves those specifics, including the generated text and the review decision. Reviewers can then assess whether that run met responsibility, auditability, and traceability expectations. In a contested organisational setting, that is the governance unit that actually matters.

• 13-RAIDT-Evidence-Review_M_v10
• 16-RAIDT-Audit-Accountability_M_v05