• workshop_dense_100#slide 96

The listed policy instruments reuse the same evidence grammar by converging on a common operational question: what retained evidence allows a reviewer to reconstruct, test, and challenge one consequential use of generative AI? The motivation paper argues that contemporary governance pressure increasingly demands documentation, transparency, oversight, monitoring, and post-deployment review, yet does not itself supply a standard proof object for one configured use. The policy pathways paper answers that gap by treating the run as the unit of governance and by translating cross-framework requirements into a shared run-level evidence pack structured through the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability). In practice, this means that governance claims are attached to stable evidence fields such as run ID and timestamp, prompt or template version, model and tool configuration, retrieval query and retrieval snapshot, output hash, and reviewer decision.

Reuse therefore occurs at the evidence layer rather than through legal uniformity. The same run-level evidence pack can be read by different instruments for different purposes: an EU-oriented reviewer may emphasise logging, transparency, human oversight, and post-market learning; an ISO/IEC 42001-oriented reviewer may read it as management-system evidence; and a NIST-oriented reviewer may use it to support mapping, measurement, governance, and control improvement. The score profile makes that reuse inspectable and comparable, while the anchors 1=missing / 3=partial / 5=audit-ready provide a disciplined way to judge evidence completeness across contexts. This is also why influence methods as governance interventions matter: if prompting, retrieval augmentation, or adaptation change what the system does, the grammar must preserve those changes in the same bounded record so that procurement, audit sampling, complaint handling, and organisational learning can all work from one evidence object rather than from fragmented narratives.

A public-service eligibility team deploys a retrieval-augmented assistant to draft explanations for housing-support decisions. For each material run, the organisation stores the prompt template ID, the specific policy clause version retrieved, the retrieval snapshot identifier, the generated explanation, its output hash, and the human reviewer decision in the run-level evidence pack. That same evidence grammar is then reused across policy instruments without rewriting the case from scratch. A procurement team can ask whether the supplier preserves audit-ready records; an internal auditor can sample the run and inspect its score profile; and a complaints officer can reconstruct what the citizen was told and why.

If the explanation cites a rule but the retrieval snapshot was not preserved, the grammar exposes the weakness immediately. Traceability and Auditability would fall on the score profile even if the prose explanation looked plausible. The value of reuse is therefore practical: the organisation does not maintain separate proof formats for procurement, standards assurance, and challenge handling. It keeps one evidence-bearing record for one configured use, and multiple governance instruments interrogate that same record from different angles.

• 10-RAIDT_Policy_Pathways_M_V50
• 14-RAIDT-Policy-Motivation_M_v11
• 16-RAIDT-Audit-Accountability_M_v05