Multi-agent AI evaluation platform. Structured contests between agent configurations with observable traces, scoring, failure tagging, and cost visibility.