StatEval Benchmark for LLM Statistical Reasoning and Proofs

StatEval provides 100,000 problems spanning undergraduate curricula and research-level proofs extracted from statistical journals. The benchmark uses the TRACE multi-agent pipeline to convert unstructured academic text into theorem-level reasoning tasks, testing LLM proficiency in complex statistical practice.