Blind head-to-head testing platform for AI computer-use agents, with early-stage community discussion on methodology and model rankings.