S11.01 - Boundary_conditions

S11.01 — Boundary conditions

flowchart LR
    A[Generic AI governance claims
all uses treated alike] --> B[RAIDT
run-level evidence framework]
    H[Practical contexts
healthcare, finance, public services,
compliance drafting, repeated workflows] --> C[[Boundary conditions
where RAIDT is strongest]]
    I[Low-stakes ad hoc uses
lighter-touch governance may suffice] --> C
    B --> C
    C --> D[Run-level evidence and evidence pack]
    C --> E[RAIDT score profile]
    D --> F[Reviewer reconstruction and contestability]
    E --> G[Governance readiness and organisational learning]

← Star S11 - Boundaries, Limitations and Future Questions

Star context: Clarifies where RAIDT should and should not be expected to deliver value, so the framework is used proportionately and not overclaimed as a universal solution for every generative AI use case.

Academic picture

Definition / background

Boundary conditions are the scope conditions under which RAIDT is analytically appropriate, operationally useful, and evidentially defensible. In this item, the concept refers to the circumstances in which a run-level governance framework adds real value: typically where generative AI is used in organisational work, where outputs shape records, decisions, services, communications, or accountability relationships, and where review of a specific run matters.

Conceptually, boundary conditions are not the same as limitations. A limitation describes what RAIDT does not yet do well, cannot fully solve, or should not be expected to guarantee. Boundary conditions instead define the kinds of contexts in which the framework is most suitably applied in the first place. They are also distinct from proportionality. Proportionality determines how much governance effort is justified; boundary conditions determine whether RAIDT is an appropriate governance lens at all, or whether a lighter-touch approach is more sensible.

This matters in GenAI governance because many frameworks are presented too broadly. They appear to apply everywhere, but they often become vague or burdensome when pushed into trivial, ad hoc, or non-accountable contexts. RAIDT avoids that problem by being explicit about its centre of gravity: it is strongest when the run can be identified, the context can be described, the evidence can be captured, and the organisational consequences of the output justify scrutiny.

Within RAIDT, boundary conditions matter because the framework depends on run-level evidence, evidence packs, and five-pillar scoring. Those mechanisms are most meaningful when there is a sufficiently discrete run, a reason to reconstruct it, and a governance need to review how GenAI was used in practice. Boundary conditions therefore protect the integrity of RAIDT by linking the framework to settings where its evidential logic genuinely supports governance readiness.

Why this concept matters

Boundary conditions solve a persistent governance problem: without clear scope conditions, a framework can be overclaimed, misapplied, or criticised for failing in contexts it was never designed to govern in depth. In generative AI, this risk is substantial because organisations often move quickly from broad principles to broad promises. Boundary conditions force a more disciplined question: in what kinds of work does a run-level evidence framework actually improve accountability?

The concept also avoids a damaging confusion between universal relevance and practical fit. RAIDT is highly relevant as a way of thinking about responsible GenAI use, but it is not equally heavy or equally necessary in every setting. If an interaction is trivial, ephemeral, purely private, or has little consequence for records or decisions, insisting on full run-level capture may create burden without much governance gain. If the interaction is repeated, high-impact, or auditable, the same capture may be essential.

Without this concept, organisations may either under-govern consequential uses or over-govern inconsequential ones. Both outcomes are problematic. The first leaves weak reviewability, poor contestability, and fragile accountability. The second produces bureaucracy, weak adoption, and performative compliance. Boundary conditions therefore help RAIDT move from abstract responsible-AI language to operational governance design.

Key idea: Boundary conditions matter because they show where RAIDT's run-level evidence model is genuinely worth applying and where a lighter or different governance approach is more appropriate.

What this item explains

The types of organisational GenAI use for which RAIDT is most appropriate.
Why repeated, consequential, reviewable workflows are the strongest fit for run-level governance.
Why low-stakes, one-off, or non-accountable uses may not justify full RAIDT evidence capture.
How identifiable runs, reconstructable context, and feasible evidence collection shape RAIDT applicability.
How boundary conditions protect RAIDT from overclaiming while strengthening its credibility.
How scope clarity improves interpretation of evidence packs, score profiles, and governance-readiness claims.

Practical example / likely audience question

Audience question

Can RAIDT be applied to every use of generative AI, or is it only meant for certain kinds of organisational work?

Answer

The concern behind this question is usually one of scope and burden. Reviewers, supervisors, or practitioners want to know whether RAIDT is being presented as a universal framework for all AI use, or whether it has a clearly defended domain of strongest applicability. The direct answer is that RAIDT can inform thinking across many GenAI settings, but it is most valuable where a specific run can affect organisational records, decisions, services, or communications and where reconstruction and review are meaningful.

For example, a staff member casually asking a chatbot for brainstorming ideas for a workshop title is not the same as a caseworker using GenAI to draft correspondence that will be placed on a service user file. In the first case, a full evidence pack and detailed scoring exercise may add little. In the second case, the output has traceable organisational significance and may later need to be reviewed, challenged, or defended. That is where RAIDT is strongest.

RAIDT handles this issue better than a generic AI governance approach because it does not stop at saying that governance should exist. It provides a practical reason for why stronger governance effort is warranted in some cases and lighter governance is warranted in others. The run-level logic makes the boundary condition operational rather than rhetorical.

Practical example in RAIDT terms

Consider a public-services setting in which a local authority uses a GenAI drafting assistant to prepare housing-benefit decision letters for claimants. The use case falls within RAIDT's boundary conditions because the workflow is repeated, the outputs enter an accountable communication process, and mistakes could materially affect how a claimant understands their entitlements or next steps.

The run-level issue is not only whether the model performs well in general. It is whether one particular letter was produced under an appropriate prompt, using the right case information, with adequate human review, and with a record of edits and approval. The evidence needed includes the task definition, the prompt template, the source case facts, the generated draft, staff amendments, the reviewer identity, timestamps, and the final communication outcome.

Responsibility is affected because the organisation must show who remained accountable for the letter. Auditability is affected because an internal reviewer may need to reconstruct how the draft was produced. Interpretability is affected because staff must understand why the output was acceptable or problematic in context. Dependability is affected because the drafting workflow should be reliable across similar cases. Traceability is affected because the run must be linked to a specific case, actor, tool configuration, and output history.

Boundary conditions improve governance readiness here by justifying why RAIDT should be applied in depth. By contrast, if the same team used GenAI informally to suggest meeting titles or rough brainstorming prompts with no record-level consequence, those interactions would sit much closer to the edge or outside the core RAIDT boundary.

Detailed link to RAIDT

Boundary conditions link to RAIDT in four ways.

First, they connect directly to RAIDT's core idea that governance should be evidence-led and tied to real organisational use rather than framed only as broad principle.

Second, they clarify when a run is a meaningful unit of governance. RAIDT depends on the run being identifiable, context-rich, and worth reconstructing; boundary conditions define when that assumption is reasonable.

Third, they determine when an evidence pack or score profile is substantively useful rather than merely formal. If the use case sits inside the boundary conditions, evidence capture and scoring can support real review. If it sits outside, the same apparatus may be disproportionate.

Fourth, they strengthen reviewability, contestability, audit readiness, and organisational learning by ensuring RAIDT is applied where those governance outcomes are actually needed.

Boundary conditions → Run-level evidence → Evidence pack → RAIDT score profile → Governance readiness

Link to the five RAIDT pillars

Responsibility

Boundary conditions support Responsibility by identifying when GenAI use is consequential enough that role clarity, review obligations, and accountability should be formalised.