• workshop_dense_100#slide 63

Within RAIDT, the score profile is the set of five pillar scores derived from the run-level evidence pack for one governed run. The composite score is the arithmetic mean of those five scores. The profile is therefore the primary measurement object, while the composite is only a compressed summary. Both use the same anchors 1=missing / 3=partial / 5=audit-ready, but they serve different purposes: the profile preserves structure, whereas the composite supports quick comparison or aggregation.

A worked logic from the papers shows why the distinction matters. Suppose a healthcare note-summarisation run is readable and includes limitation wording, so Interpretability is relatively strong. However, if repeated runs vary substantially, Dependability remains moderate or weak; if the exact retrieved guidance, prompt version, and model version are not preserved, Traceability and Auditability also remain weak. The composite may sit near the middle, yet that average hides the decisive point: the run is not governance-ready in the same way across the five pillars.

In RAIDT, this matters because run as the unit of governance and influence methods as governance interventions require reviewers to see which mechanism changed which outcome. Structured prompting, retrieval grounding, alignment controls, and adapter versions do not have uniform effects. The score profile therefore supports contestability, audit sampling, and targeted remediation, whereas the composite alone could encourage false equivalence between very different risk positions.

Consider an HR performance-appraisal draft produced with GenAI. The run-level evidence pack includes the prompt template, model deployment identifier, retrieved policy excerpt, output hash, and manager approval. The resulting score profile might be Responsibility 4, Auditability 4, Interpretability 5, Dependability 3, and Traceability 2; the composite is 3.6.

The definition is simple: 3.6 is the mean, while 4/4/5/3/2 is the profile. Why it matters is equally simple: the weak Traceability score means the organisation cannot fully prove which policy version informed the draft if the employee later contests it. Treating 3.6 as the whole story would miss the governance weakness that actually matters in dispute resolution.

• 00-RAIDT_Scoring_v1
• 08-RAIDT_Foundations_M_V50
• 13-RAIDT-Evidence-Review_M_v10