LPR+: Diverse Transformations for LLM-Aided Program Reduction

Program reduction helps compiler and language-tool developers turn large failure-inducing programs into small, shareable bug reports. LPR (Large Language Models-Aided Program Reduction) demonstrated that large language models (LLMs) can complement syntax-guided reducers by proposing language-specific transformations once a generic reducer has made the input small enough for a model. However, LPR's prompt catalog contains only five broad transformation categories, leaving much of the space explored by conventional reducers implicit. We present LPR+, which makes this transformation space explicit. LPR+ keeps LPR's oracle-validated discipline, in which every candidate must still be smaller and preserve the interesting behavior, and adds thirty refined, reducer-inspired rules derived from observations of C-Reduce, Perses, and Vulcan and provided as prompts rather than trusted rewrites. On the LPR benchmark across C, Rust, and JavaScript with gpt-5.4-mini, LPR+ lowers the mean final size from 189.7 to 180.9 tokens on the cases completed by both protocols, at the cost of additional model queries. These results suggest that an explicit, oracle-validated transformation catalog provides an extension point for LLM-aided program reduction. The demonstration video is available at https://youtu.be/kNULghJH-2E.