• workshop_dense_100#slide 64

In RAIDT, trade-offs are the visible tensions that arise when one governance gain creates cost, privacy risk, latency, reduced helpfulness, or greater operational complexity elsewhere in the same run configuration. This matters because RAIDT treats the run as the unit of governance and evaluates governance through evidence, not aspiration. The framework therefore asks not whether a configuration is simply ?better?, but how it changes the evidential standing of the run across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability).

The foundations and evidence-review papers both present trade-offs as an inherent feature of run-level governance. Richer evidence capture improves accountability, reconstruction, and organisational learning, yet it can also increase storage burden, surveillance concerns, privacy exposure, and security risk if full prompts and outputs are retained indiscriminately. Likewise, influence methods as governance interventions can improve one outcome while complicating another: structured prompts may improve Interpretability, retrieval may improve Traceability, and adapter-based specialisation may improve Dependability, but each also increases versioning, provenance, and change-control demands.

Trade-offs matter in RAIDT because governance decisions are substantive only when these tensions remain inspectable. A score profile that hides them would encourage false assurance. By contrast, a run-level evidence pack makes trade-offs contestable and actionable: reviewers can see what was gained, what new obligations were created, and whether the balance is proportionate for the task and risk setting.

In an HR performance-appraisal workflow, a manager may use a GenAI assistant to draft a review. Adding a tightly structured prompt can improve Interpretability by forcing explicit criteria and uncertainty statements, and detailed run logging can improve Auditability if the appraisal is later disputed. Yet the same design raises trade-offs: the organisation must manage sensitive employee information in prompts and outputs, store evidence securely, and maintain version control over templates and model settings.

RAIDT treats that situation as governance work rather than as an inconvenience around the edges of deployment. If the organisation captures only the polished appraisal text, it loses the evidence needed for contestation. If it captures everything without safeguards, it creates unnecessary privacy and confidentiality risk. The framework matters because it exposes that balance and makes the choice reviewable through the score profile and the run-level evidence pack.

• 08-RAIDT_Foundations_M_V50
• 13-RAIDT-Evidence-Review_M_v10
• 00-RAIDT_Scoring_v1