• workshop_dense_100#slide 42

In RAIDT, a Run ID is the unique identifier assigned to one configured use of a generative AI system for a specific task, time, and organisational context. Because RAIDT treats the run as the unit of governance, the Run ID is the field that links all artefacts belonging to that governed event: prompt and template version, model deployment identifier, decoding settings, retrieval snapshot IDs and hashes, output record, and oversight decision. The foundations paper presents it as part of the minimal illustrative run record, where a stable run reference supports later reconstruction.

An example given in the papers is the illustrative form RUN-2026-01-20-001. The value itself is less important than its governance function: it gives reviewers an unambiguous handle for one run-level evidence pack. That handle allows the organisation to retrieve full content from secured repositories, verify hashes, and connect the run to audit trails and management processes without relying on memory or narrative explanation. In this sense, the Run ID is a practical mechanism for turning technical traces into a bounded evidence object.

Its importance in RAIDT is therefore wider than simple record-keeping. The Run ID underpins the score profile across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability), because reviewers must score a specific run rather than a model in general. It also supports anchors 1=missing / 3=partial / 5=audit-ready by making evidence presence testable. Where influence methods as governance interventions change outputs at run time, the Run ID preserves which exact configuration was active and therefore makes reconstruction, contestability, and organisational learning feasible.

In a healthcare note-summarisation run, a hospital might assign RUN-2026-01-20-001 to one emergency-department summarisation event. That identifier links the prompt template used for conservative summarisation, the deployment ID of the model, any retrieval snapshot of local guidance, the generated note draft, the output hash, and the clinician's review flag. If a later safety review asks why uncertainty was or was not communicated, the Run ID lets the team recover the exact governed event rather than a generic description of the tool.

That matters because two runs on the same patient pathway may look similar while differing in prompt version, retrieved policy text, or oversight action. With a stable Run ID, the hospital can assemble the correct run-level evidence pack, justify the score profile, and identify whether the governance issue lay in Responsibility, Auditability, or Traceability rather than treating the whole system as uniformly sound or unsound.

• 08-RAIDT_Foundations_M_V50
• 18-RAIDT-Technical-Foundation_M_v04