Hunter Leary | Searcharxiv

arXiv2026

TestMap: Evidence Infrastructure for Foundation-Model-Assisted Test Generation

Foundation models (FMs) can generate plausible unit tests, but determining whether those tests are correct, useful, maintainable, and worth integrating remains difficult. Generated tests must be mapped to the code they target, inserted into real projects, built, executed, measured against the baseline suite, repaired when necessary, and compared across models and generation strategies. This validation process is fragmented across build systems, test runners, coverage tools, mutation tools, static analyzers, and experiment scripts. The problem is especially important because generated tests are both code artifacts and validation artifacts: they must themselves be validated before they can be trusted as evidence about the system under test. This paper presents TestMap, an open-source infrastructure prototype that automates evidence-backed foundation-model-assisted test generation for C#/.NET repositories. TestMap supports repository analysis, source-test mapping, baseline execution, code metric collection, test smell detection, coverage measurement, mutation testing, model-guided test generation, validation, repair, and repository-specific experiment tracking. Rather than reporting only final passing tests, TestMap records the lifecycle of each generated candidate, including failed, repaired, low-impact, and evidence positive outcomes. These intermediate outcomes can reveal model limitations, missing context, repair cost, toolchain inefficiencies, or possible faults in the system under test. Using TestMap as a design case, we describe the architecture and evidence model needed to make generated tests observable, repeatable, and comparable across repositories, models, prompts, and generation strategies. We conclude with lessons learned and open challenges, including oracle and assertion quality, metric attribution, test maintainability, flakiness, execution cost, and developer acceptance.

cs.SE↗

arXiv2026

Beyond Code Snippets: Benchmarking LLMs on Repository-Level Question Answering

Large Language Models (LLMs) have shown impressive capabilities across software engineering tasks, including question answering (QA). However, most studies and benchmarks focus on isolated functions or single-file snippets, overlooking the challenges of real-world program comprehension, which often spans multiple files and system-level dependencies. In this work, we introduce StackRepoQA, the first multi-project, repository-level question answering dataset constructed from 1,318 real developer questions and accepted answers across 134 open-source Java projects. Using this dataset, we systematically evaluate two widely used LLMs (Claude 3.5 Sonnet and GPT-4o) under both direct prompting and agentic configurations. We compare baseline performance with retrieval-augmented generation methods that leverage file-level retrieval and graph-based representations of structural dependencies. Our results show that LLMs achieve moderate accuracy at baseline, with performance improving when structural signals are incorporated. Nonetheless, overall accuracy remains limited for repository-scale comprehension. The analysis reveals that high scores often result from verbatim reproduction of Stack Overflow answers rather than genuine reasoning. To our knowledge, this is the first empirical study to provide such evidence in repository-level QA. We release StackRepoQA to encourage further research into benchmarks, evaluation protocols, and augmentation strategies that disentangle memorization from reasoning, advancing LLMs as reliable tool for repository-scale program comprehension.

cs.SE↗