• workshop_dense_100#slide 9

The claim that 'the run is the missing governance unit' captures the central gap identified across the three papers. Existing artefacts remain valuable, but they usually operate at model, system, process, or programme level. Principles tell organisations what ought to hold, model cards and datasheets describe assets in general terms, and audits or assurance routines review controls over time. What they often do not provide is a standard proof object for one configured generative AI use in context. That missing layer matters because many organisational disputes concern one output that shaped a record, recommendation, communication, or decision.

RAIDT fills that gap by defining run as the unit of governance and by specifying the run-level evidence pack as a bounded, structured, versioned, and reviewable bundle of records for one configured use event. This shifts governance from abstract commitment to inspectable proof. The run-level score profile then assesses whether that evidence supports the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability), using anchors 1=missing / 3=partial / 5=audit-ready. In this sense, the missing governance unit is not the model, because model-level artefacts cannot usually reconstruct the configuration and oversight that produced a contested output. The missing unit is the run, because it is the smallest evidence-bearing object that makes reviewability, comparability, and contestability operational in organisational generative AI use.

In a public-sector advice service, a citizen disputes eligibility guidance generated by a GenAI assistant. The organisation may have a governance programme, a policy manual, and model documentation, yet those artefacts still leave a practical gap if nobody can show the exact prompt, retrieved rule text, model settings, output, and oversight action for that case.

RAIDT closes that gap by treating the specific configured use event as the governed object. The run-level evidence pack becomes the proof bundle for the contested case, allowing reviewers to compare what was done against expected governance standards rather than relying on generic assurances. It therefore supplies the missing unit that older artefacts leave implicit.

• 08-RAIDT_Foundations_M_V50
• 13-RAIDT-Evidence-Review_M_v10
• 16-RAIDT-Audit-Accountability_M_v05