C0.05 - Score_profile

C0.05 ? Score profile

flowchart LR
    A[Traditional scoring limits
single score, generic rating, hidden trade-offs] --> B[RAIDT
run-level evidence framework]
    H[Practical fields and domains
prompts, settings, review notes,
healthcare, finance, education, public services] --> C[[Score profile
five-pillar assessment of one run]]
    B --> C
    D[Run-level evidence and evidence pack] --> C
    C --> E[Reviewer reconstruction
and contestability]
    C --> F[Governance readiness
and organisational learning]
    C --> G[Targeted improvement across
Responsibility, Auditability,
Interpretability, Dependability, Traceability]

? Star C0 - RAIDT Core, Definition, Values, Claims and Innovation

Star context: Defines the project identity of RAIDT by showing that responsible governance of GenAI in organisational work should be expressed as a structured, run-level profile across five pillars rather than as a single simplified score or a vague assurance claim.

Definition / background

The RAIDT score profile is the structured assessment of one run across the framework's five pillars: Responsibility, Auditability, Interpretability, Dependability, and Traceability. It is designed to show how governance performance is distributed across different dimensions of a concrete GenAI use event, rather than pretending that one composite figure can adequately represent the run as a whole.

Conceptually, the score profile sits downstream of run-level evidence and the evidence pack. A run produces evidence; the evidence pack organises that material for review; the score profile interprets the evidential position of the run across the five pillars. This means the profile is not merely a dashboard output or management summary. It is an evidence-linked judgement structure that makes governance characteristics visible in a disciplined and reviewable way.

This matters because governance in generative AI is rarely one-dimensional. A run can be well documented but poorly explained, dependable in routine use but weakly contestable, or clearly assigned in responsibility while still missing enough trace to support later audit. A profile preserves those distinctions. It therefore differs from generic risk scoring, maturity models, or procurement ratings that reduce varied governance conditions to a single headline assessment.

Within RAIDT, the score profile belongs in the core architecture because it operationalises the framework's claim that governance should move from principles and assertions toward evidence, reviewability, and organisational learning. It provides a practical bridge between evidential capture and governance readiness. Without the profile, evidence remains descriptive; with it, evidence becomes structured for comparative scrutiny, explanation, and intervention.

Why this concept matters

The score profile solves a recurring governance problem: organisations often need a concise assessment of a GenAI run, but oversimplified scoring can hide exactly the trade-offs that matter most. If one number is used, important weaknesses may be masked by strengths elsewhere. The profile avoids that distortion by making the pattern of performance visible.

It also prevents confusion between evidence collection and governance interpretation. An evidence pack can contain rich documentation, but decision-makers still need a way to understand what that evidence implies. The score profile provides that interpretive layer without detaching judgement from the underlying run record. It gives supervisors, auditors, policy designers, and practitioners a structured way to ask where the run is strong, where it is weak, and what should improve next.

If this concept is missing, organisations risk superficial assurance, poorly targeted interventions, and a false sense of control. A run may appear acceptable overall while containing a serious deficiency in one pillar that is critical for the use context. RAIDT uses the score profile to move beyond generic governance rhetoric and towards operational visibility of trade-offs, weaknesses, and readiness.

Key idea: The score profile matters because RAIDT needs a way to express the governance quality of one run across multiple pillars without hiding important evidential trade-offs inside a single number.

What this item measures

The degree to which one run demonstrates Responsibility through accountable roles, checks, and decision ownership.
The extent to which the run is Auditability-ready, meaning another reviewer can inspect and reconstruct what happened.
The practical level of Interpretability available for understanding how the output emerged and how it was reviewed.
The Dependability of the run in terms of stable, usable, and trustworthy performance within its task context.
The Traceability of the run across inputs, settings, actors, timestamps, outputs, and downstream actions.
The distribution of governance strengths and weaknesses across the five pillars rather than a flattened overall impression.
The evidential basis for judging whether the run contributes positively to wider governance readiness.

Practical example / likely audience question

Audience question

Why does RAIDT use a profile instead of giving each run one overall governance score?

Answer

The concern behind this question is usually a desire for simplicity. Managers, reviewers, and supervisors often want a single headline result because it seems easier to compare and communicate. The direct answer, however, is that one overall score can be misleading when governance quality is uneven across dimensions that matter differently in practice.

For example, a university might use GenAI to draft formative feedback for students. One run may score strongly on Traceability because the prompt, source material, timestamps, and reviewer notes are all preserved. It may also score reasonably on Responsibility because an academic is clearly assigned to approve the feedback. Yet the same run may score weakly on Interpretability if the reasoning behind key feedback statements cannot be clearly reconstructed, and weakly on Dependability if similar prompts yield inconsistent outcomes across comparable assignments. A single score could blur those weaknesses and imply a level of assurance that the evidence does not justify.

RAIDT handles this better than a generic AI governance approach because it does not treat governance as a single abstract property. It treats governance as a profile of evidence-backed conditions at the level of the run. That allows a reviewer not only to judge the run, but also to identify what kind of improvement is needed and why.

Practical example in RAIDT terms

Consider a finance setting in which a GenAI assistant is used to draft a quarterly internal risk summary for senior management. The run-level issue is not only whether the draft is useful, but whether its governance condition is visible in a way that supports review before the summary influences management discussion or downstream reporting.

The evidence needed includes the task brief, the prompt or template used, any source reports supplied to the model, the system version and settings, the generated summary, analyst edits, reviewer comments, approval records, and a note of any unsupported or contested statements. Responsibility is affected because the organisation must show who owned review and sign-off. Auditability is affected because another reviewer should be able to reconstruct how the draft was produced. Interpretability is affected because the team must understand how claims in the summary connect to the source material and instructions. Dependability is affected because the drafting process should be sufficiently stable and accurate for repeated reporting cycles. Traceability is affected because the run must be linked to the relevant inputs, timings, actors, and final approved artefact.

The score profile improves governance readiness here because it shows whether the run is uneven across pillars. A run might be well traced and well assigned but still weakly dependable if the model inserts unsupported emphasis, or weakly interpretable if reviewers cannot easily explain why key phrasing appeared. That visibility supports targeted control improvement rather than generic reassurance.

Detailed link to RAIDT

Score profile links to RAIDT in four ways.

First, it gives evaluative form to RAIDT's core idea that responsible GenAI governance should be evidenced at the level of the individual run.

Second, it depends on the run and on run-level evidence, because the profile can only be justified if one concrete use event has been sufficiently captured and reviewed.

Third, it sits alongside the evidence pack as one of RAIDT's two practical outputs: the evidence pack organises the proof, and the score profile expresses the governance condition revealed by that proof.

Fourth, it strengthens reviewability, contestability, audit readiness, and organisational learning by making the structure of strengths and weaknesses visible rather than implicit.

Score profile ? Run-level evidence ? Evidence pack ? Governance readiness

Link to the five RAIDT pillars

Responsibility

The score profile measures Responsibility by showing whether accountability, authority, and review duties were clearly allocated in the run and supported by evidence rather than assumption.