Chapter Assessment: A Help Center RAG Bot

This review follows the help center bot design from its first reading of traces to its one page eval plan, with stops at the test set, the graders, the release gates and monitoring. The bank holds 20 questions, and each attempt draws ten, so repeated attempts cover different parts of the design. Most questions place you inside a decision the team faced, and nearly half ask you to work out a number. Keep four things close: where each of the six failure modes starts, the two rates of the faithfulness judge, why version B failed a gate it seemed to pass, and why an article update counts as drift