"A layered playbook for interpreting AI agent evaluation scores, classifying failures by root cause (eval setup, agent config, or platform limitation), and mapping each to specific remediation actions. - View it on GitHub