• workshop_dense_100#slide 99

By the end of the workshop, several questions should now be answerable in a disciplined way. Participants should be able to say what counts as a material run in their context and why RAIDT treats the run as the unit of governance. They should be able to specify the minimum contents of the run-level evidence pack: prompts and templates, configuration identifiers, outputs, retrieval snapshots where relevant, hashes and timestamps, and human review or control steps. They should also be able to explain how those artefacts support the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability) and how the resulting score profile is interpreted using anchors 1=missing / 3=partial / 5=audit-ready.

They should also now be able to answer the practical design questions that follow from scoring. Which pillars are currently weak, and why? Which influence methods as governance interventions are active already, and which must now be explicitly logged and governed? What level of implementation is proportionate: manual pilot, semi-automated capture, or orchestration-layer evidence logging? How will reviewer calibration and repeat-run testing be used where Dependability or variance matters? Finally, the workshop should leave participants able to state the framework's boundaries. RAIDT is not a guarantee of correctness, does not replace professional judgement, and is strongest where bounded evidence can be preserved under suitable access, privacy, and retention controls. If those answers are not yet available, the workshop has not fully completed its job.

A healthcare governance group reviewing clinical note summarisation should now be able to answer concrete questions rather than offer general reassurance. Can one note-summarisation run be reconstructed? Are safety constraints, uncertainty wording, escalation prompts, and clinician review records preserved? Does the score profile show strong Responsibility but weaker Auditability because versioned prompts are missing? If so, is the next action to tighten logging or to redesign the prompt template?

The group should also be able to articulate boundaries. RAIDT can make the run more auditable and contestable, but it does not replace clinical judgement and does not prove that a summary is medically correct. That is what a completed workshop should make explicit.

• 08-RAIDT_Foundations_M_V50
• 11-RAIDT_Academic_Logic_M_v11