Dependency-Closed Evidence Contracts for Quantum Machine Learning Comparisons

Comparative quantum machine learning (QML) claims rest on chains of evidentiary prerequisites that isolated checks leave implicit. We introduce an executable dependency-closed evidence contract for structured QML comparison records. Six local obligations form a declared prerequisite partial order. From the locally satisfied obligations, the contract extracts the unique greatest downward-closed support state and identifies the order-minimal frontier where admissible support ends. All 14 constructed cases yielded the manually specified expected states and agreed with a separately implemented reference evaluator. Exhaustive enumeration confirmed agreement between two implementations of the support operator across all 64 local-predicate profiles. In a public IQP-kernel versus RBF-SVC benchmark, withholding train/test identities restricts support to the claim and method branches. Adding the published split extends support through evaluation, selection and result/resource validity under the declared resource policy. Accuracies of 0.80 and 0.90 place the focal superiority claim beyond the support frontier under the specified comparison rule. Dependency closure converts local evidentiary judgments into a precise characterization of comparative support that preserves independently established obligations and identifies the exact evidentiary boundary governing each claim.