Evaluation Choices Shape Apparent Quantum Kernel Robustness under Distribution Shift

Reproducible artifact for "Evaluation Choices Shape Apparent Quantum Kernel Robustness under Distribution Shift". The study implements a controlled kernel-swap comparison of classical and quantum kernels on fixed EMBER, UNSW-NB15, and ToN-IoT distribution-shift case studies, with prospective external corroboration on three TableShift domain shifts. The confirmatory protocol uses target-label-free ID-validation selection, per-configuration regularization tuned on training data alone, complete 60-versus-60 structurally matched candidate budgets, three-source-dataset-equal aggregation, and conditional small-cluster uncertainty. The apparent out-of-distribution advantage of fidelity kernels does not survive the controlled reference. Version 0.7.0 resolves the confirmatory budget-description inconsistency, clarifies the associational scope of the specification curve, adds BCa and leave-one-q-split sensitivity reporting, and extends the repeated finite-shot analysis with an explicitly heuristic train-Gram intervention and a coherent Nyström out-of-sample PSD extension.