Computable Design Space Exploration via Governance Frameworks: Joint Optimisation of Permission Boundaries, Monitoring Thresholds and User Transparency Policies (for Proxy AI)

This dataset serves the paper 'Computable Design Space Exploration via Governance Frameworks: Joint Optimisation of Permission Boundaries, Monitoring Thresholds and User Transparency Policies (for Proxy AI)‘, addressing risks of unauthorised access, irreversible operations, and accountability gaps arising from tool invocation, memory read/write operations, and multi-agent delegation chains in ’proxy/agent-based AI‘. It proposes and evaluates the ’Computable Governance Design Space Search (CG-DSS)" framework: formalising permission boundaries (P), monitoring and escalation thresholds (τ), and user transparency policies (U) into a computable hybrid design space. Under hard constraints, it performs constrained multi-objective Bayesian optimisation with verification filtering, yielding deployable, provably feasible Pareto-optimal governance solutions. The dataset provides reproducible, auditable trajectory-level hyper-simulation data and audit fields (aligned with NIST risk governance closed-loop and audit record intent), comprising 600 replayable governance trajectories and 7,150 event-level audit records. This is accompanied by a six-level granular cognitive graph (Nodes/Edges/Triples) and schema for locating and attributing ‘evidence-mechanism chains’ in failure samples, thereby ensuring conclusions are falsifiable, reproducible, and auditable.