Separate high-cardinality effects from sampling asymmetry in per-dimension KS

Determine whether the near-100% false-positive behavior of Bonferroni-corrected per-dimension KS on the Trendyol benchmark is caused primarily by high-cardinality identifier columns, by the asymmetric reference/current sampling design, or by their interaction, using matched-sampling and identifier-removal ablations.

Background

Per-dimension KS is dominated by ID-like columns in the Trendyol benchmark. The reference and current samples are drawn from different stream regions and have different positional sample sizes, producing a distributional discrepancy on identifier columns that overwhelms the injected-label signal. The paper identifies matched sampling and separate removal of ID columns as the required ablations, but does not perform them.

References

Disentangling high-cardinality per se from sampling asymmetry requires an ablation (matched sampling, ID-column removal) identified as future work in Section~\ref{sec7}.

— Beyond Marginal Monitoring: Distributed Joint-Distribution Testing for Data Concept Drift in Large Scale E-Commerce Operations  (2610.08132 - Pullu et al., 6 Oct 2026) in Section 6.3.7, Failure modes: per-dim KS, MMD-RBF, and projected KS; discussed further in Section 7