• integrated_82#Q2.15

RAIDT fits high-impact, repeated organisational use especially well because those settings create persistent demands for justification, comparability, and review that model-level documentation cannot meet on its own. The papers argue that repeated use exposes run-to-run variance, configuration drift, and uneven governance quality across teams. For that reason, RAIDT does not treat one output as self-explanatory. Instead, it uses the run-level evidence pack and a score profile to turn each material use into an inspectable and comparable proof object. That is particularly valuable where outputs influence credit explanations, public-service advice, clinical summaries, cybersecurity triage, or HR decisions, because these are settings in which an organisation may later need to show what happened, why it happened, and what checks were applied before reliance occurred.

This is also why RAIDT aligns closely with questions of legitimacy and trust, but in a bounded organisational sense rather than as a general theory of trust in AI. The programme repeatedly argues that legitimacy cannot rest on narrative assurance alone. Organisational legitimacy is strengthened when stakeholders can inspect evidence of governance, reconstruct contested runs, and see how influence methods as governance interventions were logged and reviewed. Trust, in this formulation, is better understood as justified reliance supported by evidence rather than blind confidence in a model. RAIDT is therefore strongest where multiple stakeholders need review capability, where runs recur often enough for systematic sampling and calibration, and where governance claims must survive challenge. In such contexts, evidence-centred review is more defensible than ad hoc explanation after the event.

Consider a bank using GenAI to draft adverse-action explanations after credit refusals. The use is high-impact, repeated, and legitimacy-sensitive because customers may challenge the explanation and regulators may inspect the process. RAIDT fits well here because the bank can treat each explanation run as a governed event: it records the prompt template, model version, criteria version, any retrieval used, the generated explanation, and the reviewer decision.

Over time, the bank can compare runs, calibrate reviewers, and detect whether a configuration improves interpretability while weakening traceability. That is a stronger basis for institutional trust than simply saying the model is responsible in principle. The organisation can show why a particular explanation was produced, how it was checked, and whether its evidence reached anchors 1=missing / 3=partial / 5=audit-ready. In legitimacy-sensitive environments, that ability to justify one contested run is exactly where RAIDT adds value.

• 11-RAIDT_Academic_Logic_M_v11
• 12-RAIDT_DSR_Theory_M_v8