{
  "analytic": "agent-chain",
  "kind": "declared_analytical_or_constructed_example",
  "experimental_model_result": false,
  "panels": [
    {
      "kind": "line",
      "x": [
        1,
        2,
        3,
        4,
        5,
        6,
        7,
        8,
        9,
        10,
        11,
        12,
        13,
        14,
        15,
        16,
        17,
        18,
        19,
        20
      ],
      "series": [
        {
          "label": "Each conditional success=0.9",
          "y": [
            0.9,
            0.81,
            0.7290000000000001,
            0.6561,
            0.5904900000000001,
            0.531441,
            0.4782969000000001,
            0.4304672100000001,
            0.3874204890000001,
            0.3486784401000001,
            0.31381059609000006,
            0.2824295364810001,
            0.2541865828329001,
            0.2287679245496101,
            0.20589113209464907,
            0.18530201888518416,
            0.16677181699666577,
            0.15009463529699918,
            0.13508517176729928,
            0.12157665459056935
          ]
        },
        {
          "label": "Each conditional success=0.97",
          "y": [
            0.97,
            0.9409,
            0.912673,
            0.8852928099999999,
            0.8587340256999999,
            0.8329720049289998,
            0.8079828447811298,
            0.7837433594376959,
            0.7602310586545651,
            0.7374241268949281,
            0.7153014030880802,
            0.6938423609954377,
            0.6730270901655746,
            0.6528362774606074,
            0.6332511891367891,
            0.6142536534626855,
            0.5958260438588049,
            0.5779512625430407,
            0.5606127246667495,
            0.543794342926747
          ]
        },
        {
          "label": "Each conditional success=0.99",
          "y": [
            0.99,
            0.9801,
            0.970299,
            0.96059601,
            0.9509900498999999,
            0.941480149401,
            0.9320653479069899,
            0.9227446944279201,
            0.9135172474836408,
            0.9043820750088044,
            0.8953382542587164,
            0.8863848717161292,
            0.8775210229989678,
            0.8687458127689782,
            0.8600583546412884,
            0.8514577710948755,
            0.8429431933839268,
            0.8345137614500875,
            0.8261686238355866,
            0.8179069375972308
          ]
        }
      ],
      "xlabel": "Required stages",
      "ylabel": "Whole-chain success",
      "logx": false,
      "logy": false
    }
  ],
  "title": "Small conditional failure rates accumulate across required stages",
  "assumptions": "Each stage has the labeled success probability CONDITIONAL on all prior successes. No independence of marginal events is asserted.",
  "worked": "若每一步在此前成功条件下成功率均0.97，20步全链成功率约0.544。恢复、重试与外部状态可改变这些条件概率，不能用此数估计真实Agent产品。",
  "source_ids": [
    "react",
    "toolformer",
    "webarena",
    "arxiv-2211.17192"
  ],
  "formula_source": "Article equations and this versioned script",
  "diagram_nodes": [
    "任务目标 / 权限 / 版本",
    "候选行动与参数检查",
    "带意图标识的正常调用",
    "待确认 / 成功 / 拒绝",
    "未知结果查询与显式重放",
    "外部目标和保护状态核验"
  ]
}