• qa_deck_100#slide 98 · Empirical programme, calibration, procurement, and assurance

RAIDT supports procurement decisions by moving buyer review away from generic vendor claims and towards inspectable use-level proof. The policy pathways paper states that procurement teams can request evidence packs and score reports from suppliers, while public procurement can require minimum RAIDT thresholds and contract clauses for ongoing updates. This works because RAIDT treats the run as the unit of governance. Rather than asking whether a vendor is responsible in the abstract, the buyer reviews a run-level evidence pack for a material use: run ID, prompt or template version, model and tool configuration, retrieval snapshot where relevant, output hash, and recorded human or automated checks.

Procurement decisions then become comparable across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability) through a score profile. The papers emphasise anchors 1=missing / 3=partial / 5=audit-ready, which gives buyers a disciplined way to distinguish narrative assurance from reconstructable evidence. RAIDT is especially useful where suppliers rely on external models, APIs, or proprietary components, because it supports layered disclosure rather than unrealistic demands for full source disclosure. Buyers can require audit rights, evidence retention duties, and incident-reporting pathways while still respecting intellectual-property limits. In effect, RAIDT turns procurement from a paper-based trust exercise into an evidence-based governance decision that can be repeated, challenged, and monitored after award.

In a healthcare procurement, an NHS trust evaluates two suppliers for a clinical note-summarisation assistant used in chest-pain triage support. Both vendors claim strong governance, but the trust requests a run-level evidence pack and score profile for the same representative workflow. Supplier A provides version-controlled prompts, model and version identifiers, retrieval snapshot IDs, output hashes, clinician sign-off rules, repeat-run logs, and explanation templates. Supplier B provides a policy statement and a system brochure but no reconstructable run record.

Using the five pillars, the trust can see that Supplier A is much stronger on Auditability, Dependability, and Traceability, while Supplier B sits closer to anchors 1=missing / 3=partial / 5=audit-ready. The trust awards the contract to Supplier A and includes clauses requiring refreshed evidence packs at payment milestones and after incidents. The procurement decision is therefore based on governed use evidence, not branding or generic compliance claims.

• 10-RAIDT_Policy_Pathways_M_V50
• 14-RAIDT-Policy-Motivation_M_v11