C0.11 - Core_innovation

C0.11 ? Core innovation

flowchart LR
    A[Traditional governance limits
principles, model cards, supplier assurances] --> B[RAIDT
run-level evidence framework]
    H[Practical fields
prompts, settings, outputs, review notes, sectors] --> C[[Core innovation
bounded proof object + anchored score profile]]
    B --> C
    C --> D[Evidence pack]
    C --> E[RAIDT score profile]
    D --> F[Reviewer reconstruction
reviewability and contestability]
    E --> G[Governance readiness
comparison, learning, policy alignment]

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

Star context: Defines the project identity of RAIDT by clarifying what is genuinely new about the framework: it turns responsible GenAI governance into a run-level evidential method built around reconstructable proof and an anchored five-pillar governance-readiness profile.

Definition / background

Core innovation refers to what is specifically new in RAIDT as a governance framework for organisational uses of generative AI. The innovation is not that RAIDT talks about responsibility, audit, explanation, reliability, or traceability in the abstract. Many frameworks do that. RAIDT's innovation is that it binds governance to the run and produces two linked practical outputs from that choice: a bounded run-level proof object, assembled as an evidence pack, and an anchored governance-readiness score profile justified across Responsibility, Auditability, Interpretability, Dependability, and Traceability.

This matters conceptually because AI governance often remains split between broad principles and fragmented technical records. Principles can be normatively useful, but they are often too general to inspect one actual use event. Technical logs may capture activity, but they often lack the contextual and evaluative information needed for governance judgement. RAIDT's innovation sits between these extremes. It creates a structured way to show what happened in one configured use of GenAI, what evidence exists for that use, and how that evidence supports a practical governance assessment.

The item belongs inside RAIDT Core because it explains what differentiates RAIDT from a generic governance vocabulary. RAIDT is not simply a taxonomy of desirable values. It is an operational architecture for evidencing and scoring governance readiness at the level where organisational accountability is most often tested: the specific run. That is why this item links directly to C0.03 ? Run-level evidence, C0.04 ? Evidence pack, C0.05 ? Score profile, and C0.06 ? Governance readiness.

The innovation therefore has two inseparable dimensions. The first is evidential: governance is attached to a reconstructable run-level record. The second is evaluative: that evidence is translated into an anchored profile across the five RAIDT pillars. Together, these dimensions allow comparison, review, challenge, and improvement across runs, tasks, suppliers, and organisational settings without collapsing governance into either vague principle talk or narrow technical logging.

Why this concept matters

This concept matters because it answers the question that many supervisors, reviewers, and practitioners eventually ask: what is actually new here? Without a clear account of the innovation, RAIDT could be mistaken for another responsible AI framework that restates familiar values. The concept clarifies that RAIDT's novelty lies in making governance operational through a bounded unit of proof and a structured scoring logic.

It also solves a practical problem in organisational GenAI governance. Organisations are often told to govern AI responsibly, but they are given little guidance on what evidence should exist for one concrete use event or how that evidence should be assessed consistently. RAIDT addresses this by specifying both the evidential object and the evaluative profile. This avoids the confusion of assuming that a policy statement, a model card, or a supplier assurance is sufficient to demonstrate governance readiness in practice.

If this innovation is missing, governance remains difficult to inspect and compare. Review becomes ad hoc, contestability weakens, audit discussions stay superficial, and organisational learning is easily lost because there is no stable object to review. RAIDT matters precisely because it creates a repeatable bridge from principle to evidence, from evidence to scoring, and from scoring to governance improvement.

Key idea: RAIDT's core innovation is that it turns responsible GenAI governance into a reconstructable and scoreable run-level practice rather than leaving it as a set of abstract commitments.

What this item enables

It enables governance to attach to one specific configured use of GenAI rather than only to a model, vendor, or policy statement.
It enables the creation of a bounded evidence pack that functions as a practical proof object for review.
It enables a five-pillar score profile that is anchored in evidence rather than impression or rhetoric.
It enables structured comparison across runs, workflows, suppliers, teams, and sectors.
It enables reviewability and contestability because another person can inspect how a governance judgement was reached.
It enables audit readiness by linking claims about responsible use to reconstructable records.
It enables organisational learning by turning individual use events into analyzable governance cases.
It enables RAIDT to position itself as a practical governance framework rather than only a conceptual responsible AI model.

Practical example / likely audience question

Audience question

What is genuinely novel about RAIDT if other AI governance frameworks also discuss accountability, transparency, and documentation?

Answer

The concern behind this question is that RAIDT might appear to rename familiar responsible AI ideas without adding a distinct methodological contribution. The direct answer is that RAIDT does not claim novelty simply because it values good governance. Its novelty lies in how governance is operationalised. RAIDT specifies what evidence should exist for one configured use of GenAI, how that evidence can be assembled into a bounded proof object, and how that proof can be assessed across five explicit governance pillars.

A practical example makes the difference clearer. A generic AI governance approach may say that an organisation should document uses of GenAI and ensure accountability. RAIDT goes further by asking whether, for one specific run, a reviewer can reconstruct the task, context, inputs, configuration, outputs, human interventions, and decision pathway, and then score the governance readiness of that run in a disciplined way. That is a more concrete and inspectable proposition.

RAIDT handles this issue better than a generic approach because it integrates evidence and assessment into the same governance workflow. It does not stop at saying that organisations should be responsible. It defines what evidence responsibility must be able to point to and how that evidence supports comparative governance judgement. That is what makes the framework methodologically distinctive rather than merely rhetorically aligned with existing responsible AI discourse.

Practical example in RAIDT terms

Consider a finance setting in which a bank uses a GenAI assistant to draft an internal credit-risk briefing for relationship managers. The use case appears routine, but the run-level issue is whether the briefing relied on appropriate inputs, whether the model configuration was suitable for the task, whether sensitive information was handled correctly, and whether human review prevented overconfident or misleading claims from entering downstream decision-making.

Under a generic governance approach, the bank might rely on policy approval, vendor documentation, and a statement that human oversight exists. Under RAIDT's core innovation, the organisation would instead assemble a run-level proof object containing the task purpose, prompt template, source data references, model or tool version, generated output, reviewer edits, approval pathway, and any warnings or exceptions. It would then assess that run across the five pillars. Responsibility concerns who authorised and reviewed the briefing. Auditability concerns whether a later reviewer can reconstruct the run. Interpretability concerns whether the basis of the output is understandable enough for internal use. Dependability concerns whether the process produced a stable and usable draft. Traceability concerns whether the run is linked to identifiable artefacts, timing, and workflow steps.

This improves governance readiness because the bank can defend or challenge a specific use event with evidence, not just with policy language. It also supports comparison across similar runs, making it easier to identify recurring weaknesses in prompt design, review practice, or workflow control. In RAIDT terms, the innovation is the move from general assurance to inspectable run-level governance.

Detailed link to RAIDT

Core innovation links to RAIDT in four ways.

First, it clarifies the RAIDT core idea that responsible governance should be judged from actual organisational use rather than from abstract declarations alone.

Second, it depends on the run as the unit of governance and on C0.03 ? Run-level evidence as the evidential basis for reconstructing one configured use.

Third, it becomes operational through C0.04 ? Evidence pack and C0.05 ? Score profile, which turn evidence into structured governance assessment.

Fourth, it advances reviewability, contestability, audit readiness, and organisational learning by making governance claims examinable at the level of concrete practice.

Core innovation ? Run-level evidence ? Evidence pack ? RAIDT score profile ? Governance readiness

The chain matters because the innovation is not a detached theoretical claim. It is the integrated movement by which RAIDT transforms one GenAI run into a governance object that can be reviewed, scored, compared, and improved.

Link to the five RAIDT pillars

Responsibility

Core innovation strengthens Responsibility by requiring governance claims to be attached to identifiable roles, decisions, and review actions within a concrete run rather than left at the level of organisational aspiration.