• workshop_dense_100#slide 98

Supervisors should not confuse the core contribution with any peripheral proof of feasibility, adjacent theory, or claim RAIDT explicitly rejects. The papers state that RAIDT is not a general theory of all trust in AI, not a guarantee that outputs are correct, not a replacement for clinical judgement, legal duties, or formal decision procedures, and not merely an empirical performance study. Equally, the contribution is not the existence of a new tool, a sector vignette, a standards map, or repeat-run testing as such. Those elements support the argument, but they are not its centre.

The centre is a bounded governance framework in which the run as the unit of governance is rendered inspectable through a run-level evidence pack and a score profile across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability). That is why discussions should return to whether a point strengthens reconstruction, review, comparison, or challenge of a configured use event. The anchors 1=missing / 3=partial / 5=audit-ready matter because they operationalise this claim, but even the rubric is secondary to the deeper design-theory contribution: governance readiness is made observable at run level. Supervisors should therefore guard against rebranding RAIDT as only model documentation, only LLMOps instrumentation, only explainability, only compliance mapping, or only sector playbook development. Each is relevant, but only as a supporting pathway into the same core contribution.

In the HR shortlist scenario, a supervisor might be tempted to say that the important achievement is simply having a versioned LoRA adapter or a tidy standards mapping for procurement. RAIDT requires a stricter interpretation. Those features are useful, but they are not the core contribution unless they support reconstructable governance for the actual hiring run.

If a candidate later challenges the shortlist, the organisation needs the run-level evidence pack: prompt template version, criteria version, model deployment ID, any adaptation identifiers, the generated justification, and the human review decision. It also needs a score profile showing whether the run was responsible, auditable, interpretable, dependable, and traceable enough for the employment context. Without that, the discussion has confused a supporting instrument with the actual governance contribution.

• 08-RAIDT_Foundations_M_V50
• 11-RAIDT_Academic_Logic_M_v11
• 12-RAIDT_DSR_Theory_M_v8