• qa_deck_100#slide 64 · Scoring anchors, profiles, and trade-offs

RAIDT keeps a score profile because the composite is only a summary indicator. The scoring paper allows a composite score to be reported as the mean of the five pillar scores, but it states that the five-pillar profile remains the primary governance output because trade-offs are often the substantive result. Two runs can share the same composite while representing very different governance conditions. One may be strong on responsibility and interpretability but weak on auditability and traceability; another may show the reverse pattern. A single mean conceals that difference.

This matters because RAIDT separates governance readiness from task fluency or apparent usefulness. A persuasive answer can still be non-auditable or non-traceable if identifiers, retrieval snapshots, or oversight records are absent. Conversely, a fully logged run can still lack dependability if repeated runs vary materially. By retaining the score profile, RAIDT makes those asymmetries visible and resists the temptation to treat governance as one headline number.

The profile is also operationally useful. RAIDT scores the run-level evidence pack so that organisations can decide what to fix: provenance capture, human oversight, uncertainty communication, or variance testing. If one composite were treated as decisive, teams could improve averages while masking weak pillars that matter most in high-stakes settings. The score profile therefore supports contestability, calibrated review, and better-targeted governance improvement.

Consider public-service eligibility advice. A run might receive Responsibility 4, Auditability 2, Interpretability 4, Dependability 3, and Traceability 2. The composite would be 3.0, which could appear acceptable on a dashboard.

The score profile says something more important. Citizens can read the answer and staff applied some oversight, but reviewers cannot reconstruct the exact rule text used because retrieval snapshots and provenance identifiers were not preserved. Repeated runs also show some instability. If the council relied only on the composite, it might continue deployment and miss the core governance weakness. Looking at the profile instead points directly to the needed interventions: preserve the retrieved regulation version, hash the run artefacts, and tighten repeat-run controls before using the workflow in contested cases.

• 00-RAIDT_Scoring_v1
• 08-RAIDT_Foundations_M_V50