发表机构
City University of Science and Information Technology (CUSIT); National University of Computer and Emerging Sciences (FAST-NUCES)（城市科学与技术大学; 国立计算机与新兴科学大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本综述整合量子编译器测试与验证方法，提出基于保障契约和证据的审计框架，通过失败案例和实例区分方法适用性，强调回归架构需受控评估。
AI 中文摘要
量子编译器保障需要一种语义关系，该关系须与预期用途相匹配，并能提供可在交付接口处暴露违规行为的观测手段。本综述作为一项关键的整合性调研，比较了验证转换、等价性检查、差分测试与蜕变测试、基于属性的测试，以及跨修订版本的保障证据。一个源代码目录和聚焦提取矩阵区分了形式化保障、已报告的实证发现、分析性推论和提议的实践。记录的覆盖更新和源代码级选择决策使该综述在其声明范围内可审计。四个记录的失败案例将条件适用性、参数关联、终止和相位约定与方法选择联系起来。一个涉及布局元数据、参数绑定和辅助量子比特的实例演示了为何在重构解释下验证电路可能使交付接口未受检查。随后一个说明性修订场景将故障检测与变更归因区分开来。所得指南规定了适用性、互补观测以及不确定结果的处理方式。已发布的缺陷计数、变异测试结果和处理基准支持不同的主张，并未建立共同的效能排名。所提议的回归架构作为完整系统尚未评估；其实用价值需要在独立裁决的编译器变更事件上进行受控比较。
英文摘要
Quantum compiler assurance requires a semantic relation that matches the intended use and observations that can expose violations at the delivered interface. This critical integrative survey compares verified transformations, equivalence checking, differential and metamorphic testing, property-based testing, and evidence for assurance across revisions. A source catalogue and focused extraction matrix distinguish formal guarantees, reported empirical findings, analytical deductions, and proposed practice. A recorded coverage update and source-level selection decisions make the review auditable within its declared scope. Four documented failure cases connect conditional applicability, parameter association, termination, and phase conventions to method selection. A worked example involving layout metadata, parameter bindings, and ancillary qubits demonstrates why validating a circuit under a reconstructed interpretation can leave the delivered interface unchecked. An illustrative revision scenario then separates fault detection from change attribution. The resulting guidance specifies applicability, complementary observations, and treatment of inconclusive outcomes. Published bug counts, mutation results, and processing benchmarks support different claims and do not establish a common effectiveness ranking. The proposed regression architecture remains unevaluated as a complete system; its practical value requires controlled comparisons on independently adjudicated compiler-change events.
Comments38 pages, 3 figures
