quiz.json

{
  "lesson": "15-prompt-caching",
  "title": "Prompt 缓存与上下文缓存",
  "questions": [
    {
      "stage": "post",
      "question": "相比基础输入费率，Anthropic 对缓存读取（cache read）给予多少折扣？",
      "options": [
        "打 7.5 折（便宜 25%）",
        "打 5 折（便宜 50%）",
        "打 2.5 折（便宜 75%）",
        "打 1 折（便宜 90%）"
      ],
      "correct": 3,
      "explanation": ""
    },
    {
      "stage": "post",
      "question": "为什么动态时间戳必须放在缓存断点（cache breakpoint）之下，而不是之上？",
      "options": [
        "缓存只在前缀逐字节相同时才命中；一个变化的时间戳会破坏它之后所有内容的匹配",
        "时间戳会让 tokenizer 困惑",
        "它们比静态文本花费更多 token",
        "Anthropic 明确拒绝缓存块中的时间戳"
      ],
      "correct": 0,
      "explanation": ""
    },
    {
      "stage": "post",
      "question": "OpenAI 的 prompt 缓存是如何配置的？",
      "options": [
        "显式的 cache_control 标记",
        "你创建并引用的 CachedContent API",
        "无需配置的自动前缀匹配",
        "你为每个项目切换的系统级开关"
      ],
      "correct": 2,
      "explanation": ""
    },
    {
      "stage": "post",
      "question": "对 Anthropic 而言，相比 5 分钟的默认值，1 小时的延长 TTL 会带来多少写入溢价（write premium）？",
      "options": [
        "相同",
        "2 倍写入溢价（比基线高 50%）",
        "4 倍写入溢价",
        "无写入溢价"
      ],
      "correct": 1,
      "explanation": ""
    },
    {
      "stage": "post",
      "question": "要在 Anthropic 的 25% 写入溢价上达到收支平衡，需要多少次复用？",
      "options": [
        "1 次",
        "2 次",
        "5 次",
        "10 次"
      ],
      "correct": 1,
      "explanation": ""
    }
  ]
}