• workshop_dense_100#slide 77

In RAIDT, automated orchestration means implementing the framework through wrappers or workflow layers that capture governance evidence automatically during execution rather than assembling it manually after the fact. The practical purpose is to ensure that each significant use produces a run-level evidence pack and an associated score profile across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability). The papers show that this matters because generative AI governance is shaped at run time by prompts, retrieved context, tool chains, adapter versions, policy layers, and post-generation checks. If these are not logged as the run occurs, later review becomes partial, contestability weakens, and governance claims remain largely narrative.

Automated orchestration is therefore not only a convenience layer. It is the mechanism that turns fragmented traces into a bounded governance object. In RAIDT terms, it operationalises the run as the unit of governance and makes influence methods as governance interventions visible and assessable. That is why the technical foundation paper distinguishes RAIDT from generic observability: engineering logs alone are insufficient unless they are assembled into a governance-ready object that managers, auditors, compliance teams, and domain reviewers can inspect. Using the anchors 1=missing / 3=partial / 5=audit-ready, orchestration matters because it raises the likelihood that evidence is complete enough to reconstruct a run, compare workflows, challenge outputs, and improve controls over time. In short, it gives organisational governance a reliable memory.

A healthcare note-summarisation service offers a concrete example. A hospital deploys a GenAI assistant to draft discharge summaries, but wraps it in an orchestration layer that records the run ID, clinician-approved prompt template, model version, active safety-policy layer, any retrieved guideline passages, their identifiers and hashes, the generated summary, and the clinician's approval or correction. RAIDT then produces a score profile for that run and flags whether the record is sufficiently complete for high-stakes use.

This matters because a clinically plausible summary is not enough. If a patient later contests an omission, the hospital needs evidence of what the system saw, which controls were active, and what review took place. Automated orchestration supplies that evidence routinely instead of relying on memory or informal notes.

• 08-RAIDT_Foundations_M_V50
• 18-RAIDT-Technical-Foundation_M_v04
• 09-RAIDT_Empirical_M_V50.docx