{
  "id": "E04",
  "slug": "amazon-search-terms-utf8-budget",
  "group": "ecommerce",
  "title": "Qwen3.8-Flash 实测：亚马逊后台词：AI 选词，程序核对 UTF-8 字节预算",
  "query": "亚马逊 search terms 250 bytes AI",
  "goal": "从给定候选词挑选相关词，去重后实际编码计数；演示中文字符与英文字符的字节差异。",
  "input": {
    "sample_notice": "虚构业务样本；实际调用 apizt API。不是商店、客户或游戏生产数据。",
    "records": [
      {
        "id": "K1",
        "facts": "500 mL不锈钢旅行杯",
        "allowed": [
          "travel",
          "cup",
          "camping",
          "stainless",
          "便携",
          "水杯"
        ]
      },
      {
        "id": "K2",
        "facts": "可调节桌面手机支架",
        "allowed": [
          "desk",
          "phone",
          "stand",
          "adjustable",
          "手机",
          "支架"
        ]
      }
    ],
    "knowledge": [],
    "rules": {
      "byte_budget": 80
    }
  },
  "instruction": "每行输出id、terms（从本行allowed选取的字符串数组，无重复）、source_ids。全部词必须来自allowed，不能添加品牌名、竞品或性能。程序用UTF-8计算空格连接后的字节数；本例预算80bytes。 source_ids只能选这些来源ID：K1,K2；不能用resource、字段名或虚构的编号。",
  "columns": [
    "id",
    "terms",
    "source_ids"
  ],
  "expected": [],
  "steps": [
    "把候选词放进 allowed；先剔除无来源的品牌和不相关词。",
    "模型只给 terms 数组，程序按空格拼接并执行 len(text.encode('utf-8'))，不让模型猜字节数。",
    "查看 byte-budget.json；再把一个词替换成长中文字符串，确认离线负例触发预算错误。实际站点字段限制仍需运营核对。"
  ],
  "trap": "字符数不是字节数；本例80bytes是演示预算，不宣称提交后台后已收录。",
  "sources": [
    {
      "label": "Amazon 商品详情页规则",
      "href": "https://sellercentral.amazon.com/help/hub/reference/external/G200390640?locale=en-US"
    },
    {
      "label": "Amazon Seller University：后台Search Terms字节要求",
      "href": "https://m.media-amazon.com/images/G/28/AS/AGS/SU/CN_GS_Listing_Optimization_1.1_General_Guidance_EN.pdf"
    },
    {
      "label": "Qwen3.8-Flash 官方模型说明",
      "href": "https://docs.qwencloud.com/developer-guides/getting-started/latest-model"
    }
  ],
  "base_title": "亚马逊后台词：AI 选词，程序核对 UTF-8 字节预算",
  "model": "qwen3.8-flash",
  "model_label": "Qwen3.8-Flash",
  "request_options": {
    "enable_thinking": false
  },
  "omit_temperature": false
}