Extreme Value Chaos Corpus & High-Cardinality Observability

In This Component

Stochastic supply chain orchestrators do not fail under average demand—they fail in the heavy tails. This final component details how Project Aegis synthesizes adversarial tail events and verifies runtime self-healing under continuous chaos injection.

Extreme Value Theory (EVT) & Minimum Covariance Determinant (MCD)

Following Charu C. Aggarwal’s outlier methodology, Aegis profiles structural regime shifts (such as the November 2017 Black Friday surge) using ExtremeValueTheoryFilter and RobustCovarianceFilter (Minimum Covariance Determinant). These tail realizations are frozen into data/fixtures/chaos_orders_v1.json, containing:

• 99.9th percentile multi-item basket surges across constrained regional hubs (${\lambda }_k=0.65$).
• Multi-byte UTF-8 Portuguese payloads and orphaned postal prefix stress cases.
• Correlated carrier delay shocks across primary highway corridors.

Property-Based Invariant Proofs & Wide Event Telemetry

Across $10^5$ randomized state transitions driven by hypothesis, the test harness mathematically verifies two hard invariants:

$\forall t,k,i:R_{t,k,i}-R_{t,k,i}^{\text{allocated}}\ge 0\text{and}{\lim }_{\tau \to \infty }{R}_{t+\tau ,k,i}^{\text{zombie\_locked}}=0$

Every decision epoch emits a high-cardinality, non-aggregated Wide Structured Event containing trace_id, order_id, node_k, sku_i, decision_class, p99_latency_ms, and gaap_contribution_cents, enabling sub-second root-cause slicing across any customer cohort or warehouse echelon.