Pareto frontier construction

Once a week, Mode 2 searches the candidate policy universe to find the policy that produces the most expected enterprise value across the realistic distribution of futures. The candidate universe is typically 200–500 policies — the current operating policy, user-defined variants explored over the week, and algorithmically generated nearby variants that perturb risk posture, objective weights, and constraint thresholds.

Naive evaluation of 500 candidates × 100 iterations each is 50,000 underlying solves — multi-day even with aggressive parallelization. Ranking-and-selection methodology compresses that cost. Clearly dominated policies get eliminated cheaply with low-fidelity early evaluation; full evaluation is reserved for survivors. The result is a defensible Pareto frontier in a tractable weekly batch.

The Pareto frontier is a persisted, versioned object. Each frontier policy carries its full KPI distribution — its EVA distribution, its service distribution, its working-capital distribution. The knee policy is identified as the recommended active policy. The current policy is marked on the frontier. The Delta PVA between current and knee is computed and displayed. The CSCO and CFO sit in the same room and look at the same number when they decide to migrate the active policy or stay where they are.

Continuous policy intelligence is the bigger consequence. The frontier refreshes weekly with the latest learned distributions and the latest realized variabilities. As the business changes — new customers, new suppliers, shifted demand patterns, geopolitical events — the recommended policy shifts. The migration is human-confirmed, not automatic. The system tells the CSCO "the knee policy has moved; here's the Delta PVA at the new shape; here's what you'd give up if we stay where we are." The argument that used to happen in S&OP — should we run lean or run safe — becomes a math result the CFO can defend to the board with numbers.