chronos-mesh

ADR-0002: Optimizing Conditional Value-at-Risk (CVaR95) Over Expected Cost Alone

Context

Traditional ERP MRP systems minimize nominal expected landed cost, converging on 100% single-sourcing from the lowest-cost offshore foundry via the lowest-cost maritime corridor. Under tail geopolitical or seismic disruptions, single-sourced networks suffer catastrophic stockouts.

Decision

  1. Enforce an explicit anti-concentration constraint ($\sum_{\ell \in \text{out}(s)} q_\ell \le 0.65 D$) inside the MILP polytope.
  2. Evaluate policies via Monte Carlo VaR95 and CVaR95 (Expected Shortfall) (src/chronos_mesh/risk/montecarlo.py) paired with Simchi-Levi’s Time-to-Recover (TTR) vs. Time-to-Survive (TTS) contagion metric (src/chronos_mesh/graph/topology.py).