We are investigating underlying quantitative and qualitative metrics around notions of quality in legal work output. Senior attorneys at large firms undergo a blind evaluation and will rank, based on their preferences, output completed by enterprise legal AI tools (e.g. CoCounsel, Lexis+, Harvey, etc.), off-the-shelf large language models (e.g. GPT-5, Claude-4, etc.), and human mid-level associates (3-5 years of experience).