Sampling only top-degree nodes in a degree-heterogeneous network can reverse the sign of an inferred structural breakpoint — a "hub-selection artifact"

In the same 2026 Roman–Byzantine trade-network study (claim-roman-byzantine-trade-network-decoupled-after-1082-chrysobull), the authors catch their own method producing a false signal. When a break-detection statistic is computed on a subgraph built by sampling only the highest-degree ("hub") nodes of a degree-heterogeneous historical network, the inferred trend becomes non-monotonic in sample size: per the capture's full-text read, "N=800 top-degree sampling produces lower Chow F-statistics than both N=400 and N=1,600." The abstract states the consequence directly — a "hub-selection artifact in degree-heterogeneous networks can reverse the sign of the inferred Phase III slope." The cause is that Western hubs dilute inter-regional cycle structure, so which hubs land in the sample determines whether a breakpoint appears at all.

The authors' fix is a minimum-coverage sampling rule that preserves regional representation ratios rather than sampling by degree alone — an explicit, pre-specified guard against the artifact.

The general shape — a striking pattern mined from a large structured dataset may be an artifact of how the data was sampled, not a real signal — recurs across domains the vault already tracks. It is the same worry behind claim-bridge-detection-lacks-pkg-validation (whether a graph "bridge" signals a real knowledge gap or a structural accident) and claim-structural-gaps-not-modularity-score-are-gap-signal (which network measure actually carries the signal). See moc-knowledge-gap-detection. Its messier, unresolved twin in quantitative history is claim-seshat-moralizing-gods-finding-disputed-as-data-artifact.