• workshop_dense_100#slide 90

In RAIDT, procurement means using acquisition and supplier-management processes to require governance evidence before and after purchase. The procurement object is not only the model or vendor brochure; it is the governed use, evidenced through a run-level evidence pack and a score profile. Buyers can specify the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability), ask for representative run records, and set acceptance thresholds using anchors 1=missing / 3=partial / 5=audit-ready. This definition aligns with the policy pathways paper, which places procurement alongside sandboxes, conformity assessment, certification, and post-market monitoring as a route by which RAIDT turns principle into performance.

Procurement matters in RAIDT because many organisations use external providers, APIs, or platform layers. Under those conditions, governance cannot rest on model ownership or policy declarations alone. The policy motivation paper shows that procurement specialists need operational proof, while the audit and accountability paper shows why run-level evidence is necessary for reconstruction, contestability, workflow comparison, and complaint handling. RAIDT therefore makes procurement a governance lever: contract clauses can require audit rights, evidence retention, incident reporting, and updated evidence packs over time. In other words, procurement operationalises run as the unit of governance in buyer-supplier relationships and ensures that accountability survives outsourcing rather than disappearing at the contract boundary.

A public health authority buying a clinical note-summarisation tool for chest-pain consultations requires shortlisted suppliers to submit sample run-level evidence packs for the same use case. Each pack must include version data, output hashes, clinician oversight records, repeat-run stability logs, and retrieval snapshot IDs. One supplier performs well overall but shows weak Interpretability for non-specialist readers.

Instead of rejecting the bidder immediately, the authority asks for a revised explanation template, reruns the scenario, and checks whether the score profile improves. Only after the evidence becomes audit-ready does the authority award the contract, with terms requiring refreshed evidence packs after major model updates and after any reportable incident. That is procurement in RAIDT: a continuing governance control, not a one-off tender questionnaire.

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