Network Resilience: Centrality, Articulation Points and Structural Failure
How to read the Network Resilience Lab without confusing graph metrics with operational causality.
What the showcase models
The lab represents dependencies as a graph and recomputes structural measures as nodes or edges are removed. It focuses attention on connectivity, shortest paths and bridge-like structure rather than treating dependency counts as sufficient.
Metrics
Degree is local connectivity. Betweenness centrality counts how often a node lies on shortest paths between other nodes, under the metric's assumptions. An articulation point is a node whose removal increases the number of connected components in an undirected graph.
Controls and interpretation
Removing a structurally important dependency can fragment the graph or lengthen routes. That is a topological statement, not automatically a prediction of production loss. Operational meaning depends on what nodes and edges represent and whether alternative paths are truly substitutable.
Assumptions and limits
Direction, weights, capacities and temporal behavior can materially change conclusions. The miniature network is intentionally legible; enterprise dependency graphs need explicit semantics and often require directed or weighted algorithms.
Validation checks
Removing an articulation point should increase component count by definition. Recomputing centrality after a structural edit is necessary; stale centrality is not a property of the new graph. Metrics should also be checked against small graphs with known answers before being trusted on large networks.