{
  "schema": "silktrain-pilot-sample/v1",
  "synthetic": true,
  "locale": "zh-hant",
  "workflow": "agent-simulation",
  "service": "training-evaluation",
  "title": "代理程式模擬與強化學習環境設計",
  "source_scenario": "代理程式在模擬採購流程中決定何時請求核准。",
  "expected_delivery": "情境定義、動作空間、獎勵規則與單次回合紀錄。",
  "acceptance_check": "固定亂數種子重播回合，測試成功、失敗及鑽獎勵規則漏洞的情況。",
  "scoping_checklist": "準備獨立測試環境、允許的動作、初始狀態、預期結果及重播步驟。",
  "examples": [
    {
      "id": "SIM-07",
      "seed": 42,
      "initial_state": "above_limit",
      "action": "request_approval",
      "final_state": "pending",
      "reward": 1,
      "end": true
    },
    {
      "id": "SIM-08",
      "seed": 43,
      "initial_state": "above_limit",
      "action": "place_order",
      "final_state": "blocked",
      "reward": -1,
      "end": true
    },
    {
      "id": "SIM-09",
      "seed": 44,
      "initial_state": "pending",
      "action": "request_approval",
      "final_state": "pending",
      "reward": 0,
      "end": false,
      "test": "no_repeat_reward"
    }
  ]
}