C0.01 - RAIDT

C0.01 — RAIDT

flowchart LR
    A[Background problem:
policy claims, model-level assurance,
limited case reconstruction] --> B[RAIDT framework]
    F[Finance, healthcare, public services,
education, enterprise work] --> B
    B --> C[Run as unit of governance]
    C --> D[Run-level evidence]
    D --> E[Evidence pack]
    E --> G[RAIDT score profile]
    B --> H[Reviewability and contestability]
    G --> I[Governance readiness]
    H --> I
    I --> J[Audit, learning, policy alignment]

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

Star context: Defines the project identity of RAIDT and anchors the whole mind map around the claim that responsible GenAI governance must be grounded in run-level evidence rather than broad assurance alone.

Definition / background

RAIDT is a run-level evidence framework for the responsible governance of generative AI systems used in organisational work. It defines the governed object not as the model in the abstract and not as policy at the organisational level, but as a specific configured run: one use of a GenAI system for a particular task, at a particular time, in a particular context. In that sense, RAIDT is both the name of the overall framework and the core architectural idea around which the rest of the project is organised.

Conceptually, RAIDT addresses a persistent gap in AI governance. Many governance approaches describe principles, controls, or model properties, but they do not always preserve enough evidence to reconstruct what happened in one material use event. RAIDT responds by making the run inspectable. It asks what was asked, which model or workflow was used, what contextual inputs mattered, what output was produced, what checks or interventions occurred, and what evidence was retained for later review.

This matters because responsible governance of GenAI rarely fails at the level of abstract principle alone; it fails when an organisation cannot explain or defend a specific use. RAIDT therefore sits above and connects C0.02 · Run, C0.03 · Run-level evidence, C0.04 · Evidence pack, and C0.05 · Score profile. The framework turns these parts into a coherent governance method.

RAIDT differs from a generic AI governance framework in four important ways. First, it is centred on run-level evidence rather than policy aspiration alone. Second, it produces practical outputs that can be reviewed and compared. Third, it uses the five RAIDT pillars, Responsibility, Auditability, Interpretability, Dependability, and Traceability, to make governance judgement more explicit. Fourth, it is designed for governance readiness: review, contest, audit, and improvement in organisational settings.

Why this concept matters

RAIDT matters because organisations need a way to govern actual GenAI use, not only models, vendors, or policy statements. Without a framework like RAIDT, governance discussions can remain too high-level to answer the practical question that supervisors, regulators, internal reviewers, and affected stakeholders eventually ask: what happened in this case, and what evidence supports the account?

The concept solves an operationalisation problem. It connects broad responsible-AI commitments with a repeatable evidential method for documenting and assessing use. It also reduces a common confusion between governance of systems in general and governance of specific uses in context. RAIDT shows that a system may be acceptable in principle while a particular run is poorly documented, weakly controlled, or difficult to defend.

If this concept is missing, an organisation may have impressive governance language but weak case-level accountability. That creates risks for audit, challenge handling, policy compliance, service quality, and organisational learning after failures or disputes. RAIDT provides a route from principle to inspectable practice.

Key idea: RAIDT matters because it transforms a single GenAI use from an ephemeral interaction into a structured, reviewable governance object.

What this item explains

RAIDT defines the overall project object and the logic that ties the whole C0 branch together.
RAIDT explains why the run, rather than the model alone, is the primary unit of governance in organisational GenAI use.
RAIDT explains how run-level evidence is collected and assembled into an evidence pack.
RAIDT explains why a five-pillar score profile is needed to interpret governance quality rather than relying on a single metric.
RAIDT explains how reviewability, contestability, audit readiness, and organisational learning emerge from evidential structure.
RAIDT explains how claims about responsible GenAI can be supported with inspectable records rather than assertion alone.

Practical example / likely audience question

Audience question

Is RAIDT simply another responsible AI framework, or does it add something operational that existing policies and documentation do not provide?

Answer

The concern behind the question is that the governance field already contains many principles, checklists, and assurance documents. A supervisor, reviewer, or practitioner may reasonably worry that RAIDT just renames existing good practice. The direct answer is that RAIDT adds an operational layer that many generic governance approaches leave under-specified: it treats each material GenAI use as a reviewable run and requires evidence sufficient to reconstruct that use.

For example, suppose a finance team uses a GenAI assistant to produce an explanation of a quarterly variance for executive review. A generic governance approach may confirm that an approved model was used, that a policy exists, and that staff received training. RAIDT goes further by capturing the prompt or task framing, relevant source context, system configuration, output, timestamp, reviewer action, and the resulting five-pillar profile. That means the explanation can later be reviewed as an accountable event rather than as a vague memory of system use.

RAIDT handles this issue better than a generic AI governance approach because it does not stop at institutional assurance. It creates a structured bridge between use, evidence, judgement, and governance action. That bridge is what allows challenge, reconstruction, and improvement when questions arise.

Practical example in RAIDT terms

Consider a public-sector benefits office using a GenAI tool to draft a case summary for a claimant review meeting. The use case appears efficient, but the governance issue sits at run level: if the summary omits a crucial fact, overstates a risk, or misrepresents supporting evidence, the organisation must be able to reconstruct what happened in that particular interaction.

In RAIDT terms, the run would record the task purpose, user role, prompt, supporting documents supplied to the system, model or workflow version, output generated, any human edits, and the final decision about whether the draft was accepted or revised. The evidence pack would need enough material to show how the summary was produced and how it was reviewed. The most affected pillars would be Responsibility, because human oversight and use conditions matter; Auditability, because the case must be reviewable later; Interpretability, because the reasoning behind the summary should be explainable; Dependability, because consistent drafting quality matters; and Traceability, because the output must be linked to sources and actions.

RAIDT improves governance readiness here by making the case summary a reconstructable event. Instead of relying on policy assurances alone, the office can show how a specific run was performed, what evidence exists, what scores were assigned, and what follow-up action was taken.

Detailed link to RAIDT

This item links to RAIDT in four ways.

First, it defines RAIDT itself as the project's central claim that GenAI governance should be organised around inspectable runs rather than abstract commitments alone.
Second, it establishes why C0.02 · Run and C0.03 · Run-level evidence are foundational rather than secondary concepts.
Third, it shows how C0.04 · Evidence pack and C0.05 · Score profile become the practical outputs through which governance is documented and judged.
Fourth, it links the framework to C0.08 · Core value: reviewability, C0.09 · Core value: contestability, audit readiness, and organisational learning.

RAIDT core definition → Run → Run-level evidence → Evidence pack → RAIDT score profile → Governance readiness

This chain matters because RAIDT is not only a conceptual umbrella. It is the mechanism that turns one GenAI use into evidential material that can support explanation, review, challenge, and improvement.

Link to the five RAIDT pillars

Responsibility

RAIDT strengthens Responsibility by requiring that a run be situated in a role, purpose, context, and review process rather than treated as anonymous system output.