Benchmark

Compare agent configurations with blind evaluation