• qa_deck_100#slide 12 · Evidence pack and score profile

In RAIDT, governance readiness means the degree to which a concrete GenAI use can be justified, reconstructed, reviewed, challenged, and improved from recorded evidence. The construct is therefore about governability at run level, not simply about whether an output looks plausible. RAIDT defines the run as the unit of governance and asks what evidence must exist so that a stakeholder who was not present at the time can later inspect what happened. Readiness is expressed through the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability), each assessed from the run-level evidence pack rather than from intentions or policy claims. A high-readiness run is one for which purpose, constraints, provenance, checks, and oversight are visible enough to support contestability and learning.

This is why the papers distinguish governance readiness from factual correctness, legal compliance, or model intelligence. A model can produce a convincing answer and still have poor governance readiness if key artefacts are missing. Conversely, a run may reveal limitations clearly yet still be governance-ready because the evidence allows reviewers to understand what occurred and decide whether reliance was appropriate. RAIDT therefore treats readiness as an observable governance outcome grounded in reconstructable artefacts. The score profile is useful because it shows where readiness is strong or weak across the five pillars, making improvement practical rather than rhetorical.

Consider a healthcare note-summarisation run for chest pain. A clinician asks the system to produce sections for Symptoms, Diagnosis, Treatment, and Red Flags. Governance readiness is high only if the run-level evidence pack shows the constrained system role, the instruction to avoid unsafe invention, any uncertainty statements, the output, and the human oversight decision.

If the clinic retains only the final summary, the output may still read well, but the run is not governance-ready in RAIDT terms because another reviewer cannot tell what safeguards were active or whether escalation cues were required. The difference lies in the evidence available for later challenge and improvement.

• 00-RAIDT_Wording_v2
• 08-RAIDT_Foundations_M_V50