STAB: Specification-Driven Testing for Algorithmic Bottlenecks

A novel technique known as STAB creates test cases aimed at revealing algorithmic limitations based solely on a natural-language problem description. It divides the process into two components: constraint-bound maximization and the injection of adversarial structures. The constraint saturator identifies constraints and determines size assignments through rule-based saturation and CP-SAT optimization. Meanwhile, the adversarial scenario injector draws on implementation-level adversarial construction strategies from a carefully curated scenario catalog. This strategy tackles structural input conditions that influence the worst-case performance of algorithms, setting it apart from earlier methods that merely focus on enlarging input size or producing code-specific inputs.