A Benchmark Dataset for Commit-Level Directed Fuzzing of RISC-V Processors

It is constructed from ten publicly reported functional bugs in the open-source Rocket RISC-V processor, referred to as the Observable-10 benchmark. Each bug instance corresponds to a specific processor behavior change introduced by a code commit and contains bug descriptions, semantic keywords, and target behavior information for evaluation.The dataset supports research on processor fuzzing, regression verification, hardware vulnerability detection, and commit-aware testing techniques.

It enables researchers to evaluate whether fuzzing approaches can efficiently reach commit-relevant architectural behaviors and identify functional mismatches between the processor implementation and reference models.