What AI says about CASTLE

CASTLE Benchmark is a modern micro-benchmarking dataset and scoring method for evaluating static analyzers and language models on vulnerability detection. It provides a hand-crafted set of 250 C programs covering 25 CWE categories, plus the CASTLE Score metric to measure true positives, false positives, and the ability to detect common issues. The dataset enables fair comparison of individual tools and tool combinations, and the project (with a peer-reviewed paper) has a preprint available on arXiv.