Studying the efficiency gap between human and machine reasoning. We build benchmarks, tools, and architectures to close it.