# IL 201 — Evaluation Decision Brief 提交檢查表 / Submission Checklist

## 中文

在提交 Evaluation Decision Brief 前，請逐項確認：

- [ ] 解決方案假設具體到「可能被資料推翻」，包含情境、AI 做的事、預期結果與推翻條件四個欄位。
- [ ] 評估設計包含現行流程、非 AI 改善與 AI 輔助三方比較；若某一方不適用，已說明原因。
- [ ] 開發集與驗收集有穩定識別碼，驗收集在原型版本凍結後才使用，且未被用來調整原型。
- [ ] 量測指標逐案例類別分別報告，沒有用單一整體分數掩蓋弱項或重大錯誤。
- [ ] 停止條件在測試前就已定義，不是看到結果後才決定。
- [ ] 原型執行有完整 run log（版本、時間、模型／工具、設定、輸入版本、人工判定者）。
- [ ] 失敗分析單獨列出所有重大錯誤，並誠實揭露樣本不足的類別，沒有假裝已有定論。
- [ ] 成本區分一次性與經常性項目，且每個數字都標記「估計」或「已驗證」。
- [ ] 建議決定是 Proceed to Readiness／Iterate／Hold／No-Go 四者之一，並附至少三項具體證據支持。
- [ ] 若為合成練習，已標明「合成練習，不等同正式專案」；若涉及真實企業資料，已依組織資料政策取得授權。

## English

Before submitting an Evaluation Decision Brief, confirm each of the following:

- [ ] The solution hypothesis is specific enough to be disproved by data, with a scenario, the AI's role, an expected outcome, and a falsification condition.
- [ ] The evaluation design includes a three-way comparison of the current process, a non-AI improvement, and the AI-assisted approach; if one arm does not apply, the reason is stated.
- [ ] The dev and acceptance sets have stable ids; the acceptance set was used only after the prototype version was frozen, and was never used to tune the prototype.
- [ ] Measurement metrics are reported per case category, with no single overall score hiding a weak spot or a critical error.
- [ ] Stop criteria were defined before testing began, not decided after seeing the results.
- [ ] The prototype run has a complete run log (version, timestamp, model/tool, configuration, input version, human rater).
- [ ] The failure analysis lists every critical error separately, and honestly discloses under-sampled categories instead of treating them as settled.
- [ ] Costs separate one-time and recurring items, and every number is tagged "estimate" or "verified."
- [ ] The recommended decision is one of Proceed to Readiness / Iterate / Hold / No-Go, backed by at least three concrete pieces of evidence.
- [ ] If this is a synthetic exercise, it states "synthetic exercise, not equivalent to a formal project"; if real company data is involved, it has been authorized per the organization's data policy.