EvoOtter: Evolutionary Generator of Reproduction Tests

In modern software development, reproducing bugs reported in issues is a critical step before implementing any fix. However, generating a test that faithfully reproduces a failure is not trivial: descriptions are often informal and ambiguous, and validating whether a candidate test actually fails for the described reason requires costly executions and unreliable feedback. Recent research proposes combining evolutionary programming with large language models (LLMs) to address this challenge, leading to approaches like EvoOtter, which optimize both test quality and computational costs. This method uses successive elimination techniques to control execution costs, batches mutations into a single LLM call, and employs a novel fitness system specific to reproduction tests. The result is high-quality test generation at a fraction of the cost of previous inference-scaling-based methods. At Q2BSTUDIO, as a company specialized in AI for businesses, we see a direct parallel with our optimization philosophy: we combine artificial intelligence with development of custom applications to create efficient and scalable solutions. Our AI for businesses services integrate AI agents that automate testing and debugging tasks, while our offering in cloud services aws and azure provides the necessary infrastructure to run massive tests cost-effectively. Additionally, cybersecurity is a pillar in all our custom software developments, and our business intelligence services solutions with power bi help companies visualize code quality metrics. Thus, we combine evolutionary techniques and machine learning to deliver more accurate tests and more agile development processes, always with a practical and results-oriented approach.