S10.02 - 14_domains

S10.02 ? 14 domains

flowchart LR
    A[Background problem:
AI governance stays abstract or single-sector] --> B[RAIDT
Run-level evidence framework]
    B --> C[[14 domains
Cross-sector testbed for governance transferability]]
    C --> D[Run-level evidence pack]
    C --> E[Five-pillar score profile]
    C --> F[Sector playbooks and calibration]
    D --> G[Reviewer reconstruction]
    E --> H[Governance readiness]
    F --> H
    I[Healthcare] --> C
    J[Finance] --> C
    K[Law and public services] --> C
    L[Cybersecurity] --> C
    M[Education] --> C
    N[Environment] --> C
    O[Crisis response] --> C
    P[Supply chain] --> C

? Star S10 - Empirical Programme, Domains and Sector Playbooks

Star context: Positions RAIDT as an empirical programme that must travel across multiple sectors, showing that run-level governance is not confined to a single use case but can be tested, compared and adapted through domain-specific playbooks.

Academic picture

Definition / background

In RAIDT, 14 domains refers to the deliberately broad cross-sector empirical space used to test, calibrate and apply the framework. The concept matters because RAIDT is not presented as a theory that works only in one organisational niche. It is designed as a run-level evidence framework whose governance logic must remain intelligible and useful when generative AI is used for different tasks, under different constraints, by different professional communities and with different consequences if a run fails.

Conceptually, the domain layer sits between abstract governance principles and individual runs. A run is always specific: one configured use of a generative AI system for a task, at a time, in a context. A domain groups together families of runs that share regulatory pressures, risk patterns, accountability expectations, workflows and evidence requirements. The purpose of the 14-domain structure is therefore not simple categorisation. It is to create a disciplined basis for examining whether RAIDT's core logic travels across sectors without becoming vague or detached from practice.

This differs from a generic sector list in policy writing. In a generic governance document, domains may simply illustrate breadth. In RAIDT, the domains become part of the empirical programme. They shape scenario design, evidence expectations, scoring interpretation and the drafting of sector playbooks. That is why the item belongs within RAIDT's architecture: it supports the claim that run-level evidence, evidence packs and five-pillar score profiles can be compared across heterogeneous organisational settings while still remaining context-sensitive.

The 14-domain framing also helps explain how RAIDT links methodological consistency with practical variation. The framework holds the run as the unit of governance, but accepts that the content of a strong evidence pack will not look identical in healthcare, finance, education or emergency response. Domain coverage therefore strengthens rather than weakens RAIDT. It demonstrates that the framework can preserve a stable governance structure while accommodating differences in risk, justification, documentation and review.

Why this concept matters

Without a domain architecture, a framework like RAIDT could be criticised as either too narrow to generalise or too abstract to implement. The 14-domain design addresses both problems. It gives the project enough breadth to test transferability, while also forcing governance claims to confront domain-specific realities such as safety criticality, legal contestability, operational tempo, professional judgement and public accountability.

This concept also prevents a common confusion in AI governance: the assumption that one good example is enough to validate a framework. RAIDT rejects that assumption. A governance method that appears sensible in one low-risk productivity setting may fail when moved into a public service, a legal workflow or a crisis-response environment. The 14 domains therefore provide a structured way to examine where the framework is robust, where it needs calibration and how playbooks can support domain-sensitive application without abandoning a common governance core.

For organisations using generative AI, the benefit is practical. Cross-domain evidence makes it easier to distinguish universal governance requirements from sector-specific controls. That helps move governance away from slogans and towards operational design choices: what must be logged, what reviewers need to reconstruct, which pillar is under pressure in a given context and what would count as readiness for deployment or continued use.

Key idea: The 14-domain structure matters because it tests whether RAIDT can turn run-level governance into a transferable, evidence-based practice across very different organisational settings.

What this item enables

A cross-sector empirical testbed for checking whether RAIDT's run-level governance logic travels beyond a single domain.
Systematic comparison of evidence requirements across different professional, regulatory and operational contexts.
Development of sector playbooks that adapt RAIDT without fragmenting it into unrelated local methods.
More credible interpretation of five-pillar score profiles because scores can be understood against domain-specific risk expectations.
Stronger supervision, viva and publication arguments that RAIDT is empirically grounded rather than purely conceptual.
Organisational learning across domains, where lessons from one sector can inform controls, prompts, review workflows or evidence standards in another.

Practical example / likely audience question

Audience question

Why so many domains?

Answer

The concern behind that question is usually that breadth may produce superficiality. A reviewer may worry that spreading the framework across many sectors weakens analytical depth or creates an impressionistic validation strategy. The direct answer is that RAIDT needs multiple domains precisely because governance needs vary by sector, and cross-domain work tests whether the core logic travels.

A practical example makes the point clearer. A generative AI run used to draft patient-facing discharge advice in healthcare and a run used to summarise procurement risk in supply chain may both involve prompting, model selection, human review and organisational accountability. However, the evidence needed to justify the run, the harms that matter, the acceptable failure thresholds and the form of reviewer reconstruction are different. If RAIDT can still produce a coherent evidence pack and interpretable five-pillar profile in both settings, then the framework has demonstrated more than isolated usefulness.

This is where RAIDT is stronger than a generic AI governance approach. Generic frameworks often state that governance should be context-aware, but they do not specify a repeatable unit of analysis through which context can be operationalised and compared. RAIDT does. By anchoring comparison at the level of the run, the framework allows multiple domains to be examined without losing methodological discipline.

Practical example in RAIDT terms

Consider an education domain run in which a university uses a generative AI assistant to produce first-pass feedback summaries for student assignments. The run-level issue is not simply whether the output is useful. It is whether the system can be governed as a traceable, reviewable and contestable component of academic work.

The evidence needed would include the task definition, the intended role of the model, prompt configuration, any institutional guidance applied, reviewer oversight steps, examples of acceptable and unacceptable outputs, and records showing how staff correct or reject the model's suggestions. The most affected RAIDT pillars would be Responsibility, because staff must remain accountable for the pedagogic judgement; Auditability, because reviewers need to see how the summary was produced and checked; Interpretability, because users must understand what the model is and is not doing; and Traceability, because the run needs a reconstructable history.

The domain framing improves governance readiness by making clear that this educational use case should not be treated as identical to, for example, a healthcare triage-support run or a cybersecurity incident-summary run. RAIDT can use the same run-level structure across those settings, but the domain lens clarifies which evidence is decisive, which risks dominate and how scoring should be interpreted in context.

Detailed link to RAIDT

14 domains links to RAIDT in four ways.

First, it supports RAIDT's core idea that governance should be based on evidence from actual configured uses of generative AI rather than on abstract assurance claims.
Second, it shows that the run remains the stable unit of analysis even when the organisational setting, risk profile and professional workflow change from one domain to another.
Third, it strengthens both the evidence pack and the score profile by showing how comparable governance outputs can still be produced under different sector conditions.
Fourth, it supports reviewability, contestability, audit readiness and organisational learning because cross-domain comparison exposes where governance holds, where it needs calibration and where sector playbooks must become more specific.

14 domains -> cross-sector runs -> comparable evidence packs -> RAIDT score profiles -> governance readiness

Link to the five RAIDT pillars

Responsibility

The domain structure clarifies who is accountable for a run and how responsibility is distributed in different organisational settings. It prevents RAIDT from assuming that the same governance role structure applies everywhere.