• qa_deck_100#slide 48 · Outputs, review decisions, and retention

In RAIDT governance, a prompt registry and a prompt version perform different control functions. The prompt registry is the organisational control artefact: it is the managed catalogue of prompt templates, intended uses, constraints, and associated metadata that supports standardisation, ownership, and change governance. The Foundations paper refers to prompt registry templates as implementation artefacts, and the wider RAIDT argument places them alongside run schemas and evidence checklists. In governance terms, the registry answers the ex ante question: which prompt assets exist, for what purpose, and under what organisational controls?

A prompt version, by contrast, is the specific revision identifier for the exact template or instruction used in one run. The Evidence Review treats prompt template ID and version as configuration provenance, because the run-level evidence pack must show precisely what was active at the moment of use. This is an ex post governance function: it answers which exact revision shaped a contested output. The distinction matters because a registry without versions cannot reconstruct an incident, while a version without a registry gives weak assurance about approval status, ownership, scope, or change rationale. RAIDT therefore needs both, but for different reasons. The registry supports managed lifecycle control over influence methods as governance interventions; the version supports audit sampling, post-incident review, and run-level scoring. Since RAIDT treats the run as the unit of governance, the version is the field that makes one run inspectable, while the registry is the broader institutional mechanism that makes prompt change governable.

Consider an HR appraisal assistant. The organisation?s prompt registry contains an approved template family for drafting appraisal text, with rules such as draft only, no protected-characteristic inference, and mandatory manager review. Within that registry, version HR-APPRAISAL v2.1 is replaced by v2.2 after legal review strengthens fairness wording.

If an employee later challenges an appraisal, reviewers need both artefacts for different reasons. The registry shows that the prompt family had a defined owner, purpose, and approval route. The run record?s version field shows whether the disputed appraisal used v2.1 or v2.2. Without the registry, governance context is thin; without the version, the run cannot be reconstructed precisely enough for a defensible run-level evidence pack or score profile.

• 08-RAIDT_Foundations_M_V50
• 13-RAIDT-Evidence-Review_M_v10