Sharp Target-Domain Certificates for Quantum-Kernel Advantage under Distribution Shift

Immutable reproducible artifact for Sharp Target-Domain Certificates for Quantum-Kernel Advantage under Distribution Shift. Version 1.1.5 establishes the sharp finite-batch identified interval for the predictive advantage of one fixed candidate over the best member of a prespecified fixed classical-kernel family, under an additive bounded loss and unrestricted completions of the unaudited labels. Both endpoints are attained, giving sharpness and minimality; for zero-one accuracy the update is pathwise exact after any realized audit subset, including adaptively selected subsets.

This release adds verified bibliographic positioning in the partial-identification tradition and distinguishes the estimand from minimax transductive aggregation, active evaluation, multi-model evaluation, and out-of-distribution performance estimation. It also situates finite-shot complexity alongside recent favorable but study-specific quantum-machine-learning evidence. No experiments, datasets, analyses, theorems, or results were added or changed.

The empirical artifact preserves the complete controlled quantum-kernel benchmark, including train-only regularization, target-label-free deployment selection, matched search budgets, factorial and shortcut analyses, external TableShift corroboration, separable-product versus entangling-ZZ strata, circuit resources, geometry diagnostics, repeated finite-shot perturbations, the frozen 30/60/115 reference-breadth frontier, the retrospective eight-shift certificate audit, and prospective Gate-2 corroboration. The artifact does not claim universal classical superiority, classical simulability, hardware advantage, quantum speedup, or population-wide ordering.

This study falls outside the scope of institutional ethics review because it consists exclusively of secondary analyses of publicly available, deidentified datasets, without recruitment, interaction, intervention, access to direct identifiers, or any attempt at reidentification. Ethics approval and additional informed consent were not required. No ethics approval or exemption is claimed.

The archive contains source code, frozen specifications, prediction locks, physically separated label archives, complete machine-readable outputs, tests, manuscript sources, final PDFs, release notes, and SHA-256 checksums. Raw benchmark source records are not redistributed.