{
  "kind": "illustration",
  "alt": {
    "en": "A teal glass agent cube with a glowing core sends a dashed orange line with two packets to a gray server rack in the center. A dotted violet arrow runs from the rack to a violet checker block with a turnstile mark, which a gray dotted line joins to a violet gauge.",
    "zh": "左侧发光的青色智能体立方体沿橙色虚线向中央的灰色服务器机架发送两个数据包；一条紫色点线箭头从机架指向带推导符号的紫色检查方块，检查方块又以灰色点线连到紫色仪表盘。"
  },
  "caption": {
    "en": "Illustration: a coding agent builds an LLM serving system that a checker examines before its score is read, standing for how the ServerBench authors describe evaluating agent-designed systems. It does not depict ServerBench, its verifier or a tested deployment, and it does not show that the checks work as described.",
    "zh": "示意图：编程智能体构建大模型服务系统，检查器先审查该系统，再读取评分，用以表示 ServerBench 作者描述的智能体设计系统评估方式。图中并非 ServerBench、其验证器或任何经过测试的部署，也不表明这些检查如所述般有效。"
  },
  "scene": {
    "version": 1,
    "focal": "rack",
    "items": [
      {"prop": "agent", "id": "agent", "at": [-1.3, -1.9], "tint": "agents", "glow": true},
      {"prop": "server_rack", "id": "rack", "at": [0, 0], "tint": "neutral", "units": 6},
      {"prop": "math_block", "id": "checker", "at": [1.9, 0.7], "tint": "evaluations", "mark": "turnstile"},
      {"prop": "gauge", "id": "score", "at": [0.9, 2.5], "tint": "evaluations", "value": 0.7}
    ],
    "links": [
      {"from": "agent", "to": "rack", "style": "dashed", "color": "data", "packets": [0.4, 0.7]},
      {"from": "rack", "to": "checker", "style": "dotted", "color": "evaluations", "arrow": true},
      {"from": "checker", "to": "score", "style": "dotted", "color": "mute"}
    ]
  },
  "svg_sha256": "15ece95652a9e557cde63b9da6e4aa0b2053a5de0b12d17f65adca20b44e5fb7"
}
