This models separation / segregation of duties (SoD) — the standard defensive control against a single insider sabotaging a system alone. Ten sensitive permissions form five toxic pairs, each a request/approval split, rendered here as a live employee × permission access matrix instead of a 3D graph:
Initiate Wire Transfer ↔ Approve Wire Transfer
Grant System Access ↔ Approve Access Review
Deploy Code to Prod ↔ Approve Code Deploy
Export Customer Data ↔ Approve Data Export
Delete Audit Logs ↔ Modify Backup Retention
Each employee is granted permissions independently at random with probability p = avg/10. With enforcement OFF, nothing stops one row of the matrix from lighting up both columns of a pair. With enforcement ON, the policy layer blocks the second grant of any pair a person already holds:
SoD risk score R = Σ_employees 1[row holds both columns of any toxic pair]
Enforcement ON → R = 0 for every pair, every row
Enforcement OFF → R grows with headcount × avg. grants
The matrix is sorted so the highest-risk rows float to the top and is scrollable (drag vertically) once the roster outgrows the visible rows. A histogram panel shows how many employees hold 0..5 toxic pairs, and the probe lane animates "Simulate Insider Attempt": red means the highest-risk row alone reached the system core (breach), green means it stalled at the segregation boundary — the action needed a second, independently-privileged approver.
Real-world relevance: this is the model behind SOX financial controls, PAM policy engines, and code-review "two-person rule" gates — the matrix is a live audit report of who could sabotage what, alone.