Provides aggregated, privacy-preserving cluster outputs from three external research teams' analyses of ~250k Claude/Claude Code conversations; includes per-team CSVs (Stanford, Oxford, METR) for studying human–AI collaboration and model behavior without raw conversations.