• workshop_dense_100#slide 97

RAIDT becomes testable because the research programme specifies not only what should count as good governance, but also what must be observed, compared, and challenged. The papers define the run as the unit of governance, make the run-level evidence pack the scored object, and retain a multidimensional score profile across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability). In the academic-logic paper, this is operationalised through a scenario-based cross-domain design: fourteen domains, twenty scenario-configuration cases per domain, six influence configurations, repeated 10-12 times, with reviewer calibration. That design makes the framework falsifiable. If evidence capture is incomplete, or if retrieval, adaptation, or alignment do not improve the expected pillar scores, the propositions can be challenged directly from the recorded evidence.

Sector playbooks add a second layer of testability: portability with constraint. They keep the RAIDT core stable while calibrating thresholds and examples for domain duties. The papers state that healthcare may require explicit red-flag escalation, finance may require reason codes and provenance, public-service eligibility may require the exact policy clause and version, and cybersecurity may require repeat-run stability and uncertainty signalling. In that sense, playbooks are not local exceptions; they are structured tests of whether the same run-level evidence pack, the same score profile, and the same anchors 1=missing / 3=partial / 5=audit-ready remain implementable across sector conditions. This is also why RAIDT frames influence methods as governance interventions: sector playbooks can assess whether those interventions improve governance readiness, not merely output fluency.

Consider a finance adverse-action explanation workflow. A bank asks GenAI to draft the explanation sent to a customer after a lending decision. A generic system may produce fluent text, but without a stored prompt version, the exact policy clause, retrieval snapshot, output hash, and reviewer sign-off, the run is weak on Auditability and Traceability. RAIDT therefore scores the run-level evidence pack rather than the surface quality of the prose.

A finance playbook makes the test concrete by requiring reason codes, provenance for each explanation element, and clear human decision rights. The organisation can then repeat the same case under structured prompting, retrieval augmentation, or a stacked configuration and compare the resulting score profile. If the more heavily governed configuration produces better evidence and more stable Dependability without losing Interpretability, RAIDT's claims are supported. If it does not, the framework's propositions are challenged in a way that is visible and auditable.

• 11-RAIDT_Academic_Logic_M_v11
• 12-RAIDT_DSR_Theory_M_v8