• integrated_82#Q4.27

The clearest one-sentence takeaway of the RAIDT Project is this: responsible GenAI governance becomes operational only when each material use is captured as a run-level evidence pack and assessed as a score profile, because the run as the unit of governance is where risk actually materialises.

The wider significance of that sentence is that RAIDT is not presented as a new model, a vendor-specific toolkit, or a claim that governance can be solved by better rhetoric. The project argues for a new governance measurement object. By preserving reconstructable evidence for each run and evaluating it through the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability), organisations can move from abstract commitments to inspectable accountability. The papers also show why this matters methodologically and practically: run-level evidence supports repeat-run testing, policy and standards mapping, audit sampling, procurement, and organisational learning across sectors. At the same time, the project remains careful about non-scope. A strong score does not prove that a clinical summary is medically correct or that a legal duty has been fully discharged; it shows that the governance of the run is visible, reviewable, and open to challenge and improvement.

Consider a clinician using GenAI to summarise a chest-pain consultation. Without RAIDT, the hospital may keep only the final summary, leaving no reliable way to show what safety constraints were set, whether uncertainty about pending tests was required, or what oversight decision the clinician made. The output may look polished, but the governance of its production is opaque.

With RAIDT, the same event becomes a run-level evidence pack with prompt and system-message versions, model details, output, review checks, and oversight notes. A score profile then indicates whether the run was merely serviceable or genuinely audit-ready. That concrete shift from fluent output to inspectable run record captures the project's central message.

• 00-RAIDT_Wording_v2
• 11-RAIDT_Academic_Logic_M_v11