Chapter Assessment: Evaluating Agents

This review tests how to evaluate the scheduling agent: the controlled environment, final state checks, trajectory checks, repeated runs, simulated users, and the cost of each task. It holds 20 questions, and each attempt draws ten. Most questions describe a run of the agent and ask how a grader should treat it, which rule belongs on the path, or which number a report should lead with. Several ask you to compute pass@k and pass^k, a token limit, or a share of runs