BenchMIRT analyzes what LLM benchmarks are actually measuring

BenchMIRT evaluates the fidelity of LLM benchmarks to determine if current metrics accurately reflect model capabilities or merely capture dataset contamination and superficial pattern matching.