同一注册单元怎么判：型号规格覆盖的逻辑

结论：单元划得宽，检测省；划得窄，论证省。取舍点在差异大小

注册单元的划分会直接影响检测工作量。单元划得宽，多个型号合在一起申报，检测可以做覆盖论证只测部分型号；单元划得窄，每个单元独立，检测量增加但每个单元内部差异小，论证简单。

所以这是个取舍：把差异大的型号硬凑在一个单元里，检测省下来了，覆盖论证会很吃力，还可能被质疑；拆得太碎，论证轻松但检测和申报成本都上去了。

划分的判断维度

前两行是硬性的，不一致就不可能同单元。后面几行有一定弹性，取决于差异的性质和是否能论证。

覆盖论证的基本逻辑

论证要回答的是：为什么测了 A 型号，可以认为 B、C、D 型号也符合要求。

逻辑链条是三步：第一步说明分组依据——这几个型号在哪些特征上一致，差异在哪里。第二步说明检测对象的选取理由——为什么 A 是这一组里的不利型号。第三步说明其余型号不会更差——基于什么原理或数据得出这个结论。

三步里，第二步最容易出问题。很多论证写的是「选取典型型号」，而审评关心的是不利型号，不是典型型号。典型是使用频率高的，不利是性能余量小的，两者常常不是同一个。

不利型号要分项选

这是最容易被忽略的一点：不同检测项目的不利型号可能不是同一个型号。

举例来说，同一系列产品里：结构强度的不利型号可能是尺寸大、承重高的那一档；温升的不利型号可能是功率大的那一档；EMC 的不利型号可能是带无线功能的那一档；生物学的不利型号可能是接触面积大或接触时间长的那一档。

如果只选一个型号测全部项目，那么它只在某些项目上是不利的，其余项目的覆盖论证就站不住。

正确做法是分项列表：每个项目分别说明不利型号是哪个、为什么。这样做检测量会增加一些（因为可能要测多个型号），但论证是扎实的。

划分过宽的代价

把差异较大的型号放进一个单元，可能出现这些问题：覆盖论证写不圆，反复被追问；某个型号的特殊结构在论证中无法被覆盖，最后还是要单独测；后续某个型号变更时，可能影响整个单元。

最后一点常被忽略。 同一单元内的型号是绑定的，其中一个发生变更，处理上可能牵动整个单元，灵活性下降。

划分过窄的代价

拆得太细则是另一套成本：每个单元独立检测，项目重复做；申报资料成倍增加；后续维护（变更、延续）的工作量按单元数量放大。

对于型号多的产品线，这个成本相当可观。

怎么找平衡点

实务上可以这样判断：把所有型号列出来，逐个标注它们之间的差异；差异只体现在规格参数（尺寸、容量、功率档位）上的，倾向于同一单元；差异体现在结构形式、材料、原理上的，倾向于分开。

然后做个粗算：同单元方案下覆盖论证的难度与可能的追问成本，对比分单元方案下增加的检测与申报成本。两边的成本都能大致估出来，比凭感觉决定要可靠。

与检测计划的衔接

单元划分应当在排检测计划之前定下来。顺序反了会出问题：先按某个假设做了检测，后来单元划分调整，已完成的检测可能覆盖不到新的范围。

划分定下来之后，检测计划的结构就清楚了：每个单元一份项目清单，清单里标明各项目的检测对象型号及选取理由。这份清单本身可以直接转化成覆盖论证的骨架。

型号差异说明怎么写

这份材料是覆盖论证的基础，建议做成表格：每一行一个型号，每一列一项特征，把差异用颜色或标记突出。

列要覆盖：外形尺寸、承重或容量、功率或输出、关键材料、结构形式、软件版本、可选配件。列不全，差异就看不全，不利型号也就选不准。

表做出来之后通常会发现一件事：型号之间的差异比想象的少，有些标注为不同型号的产品实际只差一个颜色或包装。这类可以合并处理，进一步减少工作量。

后续变更时的影响

单元划分定下来之后，要考虑它对未来变更的影响。同一单元内的型号是绑定的，新增型号如果落在单元定义范围内，处理相对简便；落在范围外则可能要新建单元。

所以划分时可以适当考虑产品规划：已经确定要出的后续型号，如果特征相近，可以在定义单元时把范围留出来，避免每出一个型号就新建一个单元。

单元内新增型号的处理

产品线扩充时，新型号能否纳入现有单元，判断依据与最初划分时相同：预期用途和工作原理是否一致、差异是否仅在规格参数上。

如果能纳入，要重新审视覆盖论证——新型号可能在某个项目上比原来的检测对象更不利，这时候需要补测该型号的该项目，而不是简单声明被覆盖。

这一步常被省略，新型号直接挂进单元而不重新评估覆盖，等于覆盖论证出现了缺口。

我们的做法

我们排检测计划时会要求提供完整的型号清单及各型号之间的差异说明。有这份材料，才能判断每个项目该测哪个型号。

实际中常遇到的情况是，委托方给了一份型号清单但没有差异说明，我们只能按参数猜。猜错的后果是测了非不利型号，论证时才发现覆盖不住。 所以这份差异说明虽然要花点时间准备，但它决定了后面所有工作的有效性。

如果你的产品系列型号较多、单元划分还没定，可以把型号清单和差异情况发过来一起理，或者直接联系：132 4819 8029。能力范围见服务介绍，送检要求见送检要求，流程见检测流程。

English version

Conclusion. How registration units are drawn directly determines the testing workload. Broader units let several models be submitted together, so testing can rely on a coverage argument covering only some models. Narrower units mean each is independent, increasing testing but reducing internal variation and simplifying the argument. Forcing dissimilar models into one unit saves testing but makes the coverage argument strained and vulnerable to challenge; splitting too finely makes argument easy but raises testing and submission costs.

Dimensions for the judgement. Intended use and operating principle must be identical. Main structural composition should be broadly identical, allowing specification differences. Patient-contacting materials must be identical or demonstrably equivalent. Energy form and delivery must be identical, as must the sterilisation method. The first two are absolute; the rest allow some latitude depending on the nature of the difference and whether it can be argued.

The logic of a coverage argument. It must answer why testing model A supports the conformity of B, C and D. Three steps: state the grouping basis, identifying which characteristics are shared and where the differences lie; justify the choice of test article, explaining why A is the least favourable of the group; and show the others cannot perform worse, on a stated principle or data. The second step causes most problems. Arguments often say a typical model was selected, whereas reviewers care about the least favourable one. Typical means commonly sold; least favourable means smallest margin, and they are frequently different models.

The least favourable model differs by item. This is the point most often missed. Within one series, the least favourable model for structural strength may be the largest and highest-capacity variant; for temperature rise, the highest-power variant; for EMC, the variant with wireless capability; for biological evaluation, the variant with the greatest contact area or duration. Testing a single model across all items means it is least favourable for only some of them, leaving the rest of the coverage argument unsupported. The correct approach is an item-by-item table stating which model is least favourable for each and why. This increases testing somewhat but produces a sound argument.

The cost of drawing units too broadly. Coverage arguments that cannot be completed convincingly and attract repeated questions; a model with distinctive structure that cannot be covered and has to be tested separately anyway; and reduced flexibility later, since models within a unit are bound together and a change to one may affect the whole.

The cost of drawing them too narrowly. Independent testing for each unit with repeated items, dossier volume multiplied, and maintenance work for changes and renewals scaled by the number of units. For a broad product line this is substantial.

Finding the balance. List every model and annotate the differences between them. Where differences are confined to specification parameters such as size, capacity or power rating, lean towards a single unit. Where they involve structural form, materials or operating principle, lean towards separation. Then estimate both sides: the difficulty and likely challenge cost of the coverage argument under a combined unit, against the additional testing and submission cost under separate units. Both are roughly estimable, which is more reliable than intuition.

Sequencing with the test plan. Settle unit structure before planning testing. Reversing the order risks testing against an assumption that later changes, leaving completed work outside the revised scope. Once settled, each unit gets an item list stating the test article for each item and the reason for its selection, and that list converts directly into the skeleton of the coverage argument.

How we handle it. We ask for a complete model list with a statement of the differences between them, because without it we can only infer from parameters, and inferring wrongly means testing a model that is not the least favourable and discovering it when the argument is written.

Send us the model list and difference statement and we will work through it. Phone or WeChat: +86 132 4819 8029.