Not as many as a guru tells you, not as few as your designer wants. A simple way to size your creative pipeline.没有"大师"说的那么多，也没有你的设计师希望的那么少。一个给素材产能定量的简单方法。
The honest answer is: enough to keep one fresh winner entering the account before your current winner fatigues. That is the whole goal of a testing cadence — never to be caught with a tired creative and nothing to replace it.老实的答案是：多到足以在你当前的赢家疲劳之前，让一条新的赢家进场。这就是"测试节奏"的全部目的——永远不要落到"素材累了、却没东西替补"的境地。
Work backwards from your win rate. If roughly one in eight concepts becomes a scalable winner, and a winner lasts about three to four weeks before fatigue, you need to be producing somewhere around six to ten testable concepts a week to keep the pipeline full. Spend more and you can test more in parallel; spend less and the cadence relaxes.从你的"中签率"倒推。如果大约每八个概念能跑出一个可放量的赢家，而一个赢家在疲劳前能撑三到四周，那么你大概需要每周产出六到十个可测试概念，才能让管线不断档。预算更高就能并行测更多；预算更低，节奏就相应放缓。
Volume without structure is just noise, though. Each concept should be a clean hypothesis — a new hook, a new angle, a new format — not the same ad in five colors. Variations of a proven winner are for scaling, not for testing; they belong in a different budget so they do not pollute your read on what is genuinely new.但没有结构的产量只是噪音。每个概念都该是一个干净的假设——一个新钩子、新角度、新形式——而不是同一条广告换五种颜色。"已验证赢家的变体"是用来放量的，不是用来测试的；它们该放在另一笔预算里，免得污染你对"什么才是真正新东西"的判断。
Finally, give tests room to breathe. Killing a concept after a few hundred impressions tells you nothing but variance. Set a minimum spend per test based on your CPA, judge on leading indicators like hook rate and cost per result, and resist the urge to declare a winner before the data has earned it.最后，给测试留出呼吸空间。一条概念跑了几百次曝光就砍掉，你读到的只是方差、什么都说明不了。按你的 CPA 给每个测试设一个最低消耗门槛，用完播率、单次结果成本这类前置指标来判断，并克制住在数据还没"配得上结论"之前就宣布赢家的冲动。