Creative Test Ledger — Real Counts, Clear Comparisons

Mode: Download · one-time US$9.99

English description

Already tested several ad versions, but unsure which numbers can fairly sit side by side? Creative Test Ledger turns your own CSV or pasted counts into a descriptive record with every numerator and denominator visible, context comparison limits, and one variable to test next.

It separates different periods, placements, audiences and experiments; preserves zero-denominator N/A; and calculates pooled rates with total counts rather than averaging percentages. A higher observed rate stays an observation—never a guaranteed winner, statistical significance claim or causal promise.

Includes SKILL.md, English and Traditional Chinese buyer guides, a CSV contract, a synthetic worked example and optional Python rate helper with example input/output. Ordinary notes are accepted; your agent prepares the CSV. No advertising-account access, API key, platform connection, upload or budget change is included or required by this package.

Requires your own compatible agent/model. Python 3.10+ standard library is needed only for the optional helper; your model/provider fees are separate. Helper rates all use impressions as denominator and require compatible exposure-level counts and non-overlapping rows. It does not verify source accuracy, assignment or attribution definitions, run inferential statistics, forecast revenue or optimize spend. Human review remains necessary before decisions.

繁體中文介紹

廣告素材測完幾版，卻不確定哪些數字能放在一起比？Creative Test Ledger 將你的 CSV 或貼上計數整理成描述性紀錄，每個比率都保留分子與分母，標示不可比較的情境，並提出下一輪只改一項的測試方向。

不同期間、版位、受眾及實驗分開處理；零分母保留 N/A；合計使用總分子／總分母，不直接平均百分比。「觀察比率較高」不會被寫成保證勝出、統計顯著或因果。

內含 SKILL.md、繁中／英文買家指南、CSV 規格、虛構完整範例，以及選用的 Python 比率程式與輸入輸出示範。一般文字也能貼，由 Agent 整理 CSV。本套件不登入廣告帳號、不需 API 金鑰或平台連線、不上傳、不改預算。

需自備相容 Agent／模型；選用程式需 Python 3.10+，僅用標準函式庫。模型／供應商費用另計。程式統一以曝光為分母，資料需符合逐曝光計數定義且各列不重疊。不驗證資料真偽、分派或歸因定義，不做推論統計、收入預測或支出最佳化。決策前請人工確認。

Suggested invocation / 使用提示

“Use Creative Test Ledger to compare these first-party creative counts. Keep numerators and denominators visible, separate incomparable contexts, and recommend one next-test variable without declaring a guaranteed winner.”

「使用 Creative Test Ledger 整理這些自有素材計數，保留分子與分母、分開不可比情境，提出下一輪只改一項的建議，不保證贏家。」

Reproducible synthetic demonstration / 可重跑合成示範

Synthetic A has 100 qualifying clicks / 1,000 impressions = 10%; B has 30 / 200 = 15%. Show the observed difference with denominators and comparison limits, not a guaranteed winner. Separate rows of 1/10 and 90/100 pool to 91/110 = 82.73%, not 50%.

合成A為100/1000＝10%，B為30/200＝15%，保留分母與可比限制，不保證勝負。另兩列1/10與90/100合計為91/110＝82.73%，不是50%。