• workshop_dense_100#slide 74

The G. Implementation & Operations branch is the part of RAIDT that turns the framework from a conceptual model into an operating routine. Across the papers, its organising logic is staged implementation: organisations begin by capturing evidence for individual runs, then stabilise scoring and reviewer practice, and only later add orchestration, automated completeness checks, and repeat-run testing where risk justifies it. The branch therefore centres on how the run-level evidence pack and score profile are produced, reviewed, sampled, and fed back into governance decisions.

Operationally, this branch covers at least four linked activities. First, evidence capture: preserve prompts, outputs, identifiers, provenance material, checks, and oversight notes so the run can be reconstructed. Second, scoring and calibration: reviewers apply the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability), use evidence pointers, and align judgements through worked examples and anchors 1=missing / 3=partial / 5=audit-ready. Third, governance response: low scores trigger escalation, remediation, stronger constraints, or instrumentation fixes. Fourth, maturity progression: manual scoring can evolve into partial automation for objective fields and higher automation for repeat-run stability testing, while judgement-heavy questions remain under human review.

The branch also matters because RAIDT treats influence methods as governance interventions. Prompting, retrieval augmentation, adaptation layers, and alignment controls are not merely engineering choices; they are governed configuration whose evidence must enter the operational record. In that sense, Implementation & Operations is where run as the unit of governance becomes organisational practice rather than theory.

In a cybersecurity alert triage workflow, the branch begins with analysts manually saving the prompt, retrieved threat context, output, and reviewer checks for sampled runs. Each sampled alert receives a score profile so the team can see whether Dependability is weak under repeat runs or whether Traceability suffers because retrieval snapshots were not preserved.

Over time, the operations team adds light automation: run IDs, hashes, and retrieval snapshot identifiers are captured automatically, while analysts still review uncertain or high-impact cases. When repeated low scores appear, the branch does real governance work: triage prompts are revised, evidence capture is tightened, and escalation rules are updated. That is the branch in action - evidence capture, scoring, monitoring, and control updates tied to specific runs.

• 08-RAIDT_Foundations_M_V50
• 18-RAIDT-Technical-Foundation_M_v04