We do use hundreds of different environments across disciplines, and it seems to line up with subjective experience better than other benchmarks. We test coding, agentic coding, and non-coding reasoning in the environments.