CARGO Framework Reduces Evaluation Error in Agentic AI

CARGO addresses 'reference-instance divergence' in agent evaluation by treating retrieved references as procedural exemplars rather than literal targets. It gates evaluation based on retrieval confidence and focuses on penalizing contradictions rather than identifier mismatches.