不合格品处置里的检测复验规则

结论：复验必须有规则，没有规则的复验等于挑数据

检验出现不合格，能不能再测一次？这个问题不能简单回答能或不能，而是「在什么条件下能，测几次，结果怎么用」。

没有事先规则的复验是危险的：测到不合格就再测，直到出现一个合格结果为止。这在统计上必然会发生，但得到的结论没有意义。审核和监管对这一点都很敏感。

什么情况下可以复验

前四行的共同点是：有明确的、可查证的原因说明原数据无效。 这类复验实际上是重新做一次有效的试验，而不是对同一件事给第二次机会。

第五行需要事先定义。有些方法本身允许在特定条件下复验并给出判定规则，这时候按方法执行；方法没有规定的，企业自己的规则要写进文件并一致执行。

规则设定的原则

制定复验规则时，几条原则值得遵守：

一是事前定义。 规则写在文件里，在出现不合格之前就存在。事后制定的规则没有说服力。

二是次数有限。 明确最多复验几次，不能无限测下去。

三是结果的处理方式明确。 复验合格了，原不合格数据怎么处理——是作废还是保留并说明。通常应当保留并在记录中说明原因，而不是简单删除。

四是触发条件客观。 判断「设备异常」要有客观依据，不能凭操作人员说一句设备不对劲就复验。

五是复验的样品来源明确。 用原样品重测，还是取新样品，两者的含义不同，规则要说清楚。

原样品与新样品的区别

这一点常被混淆：

用原样品重测，前提是原样品状态未被前次试验改变。适用于非破坏性项目。它回答的是「上次测量是否准确」。

取新样品重测，适用于破坏性项目，或原样品已经改变的情况。它回答的是「这批产品是否合格」——注意这是另一个问题。

混淆两者会导致结论错误。 如果原因是测量出错，应当重测原样品；如果怀疑是批次个体差异，那么取新样品测出合格，也不能说明原来那个不合格样品没问题。

记录要求

复验的记录应当包含：原不合格数据；判断需要复验的原因及依据；复验的授权人；复验用的样品（原样品还是新样品，标识）；复验数据；最终判定及理由。

原不合格数据必须保留。 删除或覆盖原数据是严重的记录问题，比不合格本身严重得多。审核中如果发现记录有被抹去的痕迹，整个数据体系的可信度都会受影响。

复验被滥用的表现

几种常见的滥用形式：

复验次数明显偏多，某些项目经常出现两三次复验；复验原因写得笼统，多份记录都写「操作失误」而无具体说明；只有复验后的合格数据进入报告，原数据不见踪影；复验的授权人就是操作人员本人，没有独立审核；复验规则在不同批次上执行不一致。

这几种表现在数据统计上很容易看出来。 把一段时间内的复验记录汇总，复验率和复验原因的分布一目了然。建议企业自己定期做这个统计，早发现早纠正。

复验与批次处置的关系

复验的结论会影响整批产品的处置。需要明确的是：一个样品复验合格，不等于整批合格。

如果原不合格是因为测量问题，复验合格说明该样品确实合格，批次判定回到正常流程。如果原不合格反映的是批次质量波动，那么即使某个样品复验合格，仍需要扩大抽样确认批次状态。

这个判断要在规则里写清楚，否则实际操作中容易简化为「复验合格就放行」。

复验与统计抽样的关系

如果检验采用抽样方案，复验规则要与抽样方案配套。抽样方案本身规定了判定规则（多少个不合格判定批次不合格），随意增加复验会破坏这个方案的统计意义。

常见的错误是：抽样判定批次不合格之后，再抽一批测，合格就放行。 这实际上改变了抽样方案的接收概率，使方案失去了设计时的保护水平。正确做法是按抽样方案规定的加严检验或者批次处置流程走。

复验数据的统计价值

把一段时间内的复验记录汇总分析，能看出不少东西：哪些项目复验率高（可能是方法稳健性不足或产品该项余量小）；复验集中在哪些时段或班次（可能是人员或设备因素）；复验后合格率有多高（异常高说明可能存在挑数据）。

这类分析成本很低但发现问题的效率高，建议纳入日常的质量数据分析。

与供应商之间的复验争议

进货检验不合格时，供应商常常要求复验，理由是「我们出厂时是合格的」。这类争议的处理需要事先在质量协议里约定：由谁复验、用什么方法、在哪里做、结果以谁为准、费用怎么分担。

没有约定时，双方各测各的，数据不一致又无从判断谁对。 比较可行的约定是：分歧时由双方认可的第三方实验室仲裁，方法由双方事先确认，费用由判定错误的一方承担。

我们的做法

作为外部实验室，我们遇到不合格时的处理是：先检查试验过程有没有问题（设备、条件、操作），有问题则重做；没有问题则如实出具不合格结论，并把当时的条件记录完整。

委托方提出复验时，我们会记录复验的请求与理由，并在报告中体现试验次数。 这样做有时会让报告看起来不那么干净，但它保证了数据的真实性。如果报告只呈现合格的那一次而隐去前面的，这份报告在后续任何追溯中都是隐患。

如果你在制定复验规则，或者对某次复验的处理方式不确定，可以把具体情况发过来一起看，或者直接联系：132 4819 8029。能力范围见服务介绍，流程见检测流程，其他问题见常见问题。

English version

Conclusion. When an inspection gives a nonconforming result, the question of whether it can be retested is not answered by yes or no but by under what conditions, how many times, and how the result is used. Retesting without a prior rule is dangerous: test again whenever a result fails, until a conforming one appears. Statistically that will eventually happen, and the conclusion means nothing. Auditors and regulators are alert to this.

When retesting is legitimate. Where an operating error is identified, the original data are invalid. Where equipment malfunction is confirmed, the data source is unreliable. Where sample preparation was faulty, the sample did not represent the product. Where environmental conditions fell outside specification, the test conditions were not met. Where a result sits near a limit, a predefined rule may apply. Where the only reason is dissatisfaction with the outcome, retesting is not legitimate.

The first four share a characteristic: a specific, verifiable reason why the original data are invalid. Retesting in those cases is performing a valid test for the first time, not giving the same question a second chance.

Principles for the rule. Define it in advance, in a document that exists before any nonconformity arises; a rule written afterwards carries no weight. Limit the number of retests explicitly. State how results are handled, including what becomes of the original nonconforming data, which should normally be retained with an explanation rather than deleted. Make the trigger objective, since a judgement that equipment malfunctioned needs evidence rather than an operator's impression. And specify the sample source, because retesting the original sample and testing a fresh one mean different things.

Original sample versus fresh sample. Retesting the original sample presupposes its condition was not altered by the first test, so it suits non-destructive items, and it answers whether the earlier measurement was accurate. Testing a fresh sample suits destructive items or altered samples, and it answers whether the batch conforms, which is a different question. Confusing the two produces wrong conclusions: if the cause was a measurement error, retest the original; if batch variability is suspected, a conforming result on a fresh sample does not establish that the original nonconforming unit was acceptable.

Records. A retest record should contain the original nonconforming data, the reason and evidence for retesting, who authorised it, which sample was used and how it was identified, the retest data, and the final judgement with reasons. The original data must be retained. Deleting or overwriting them is a serious record integrity problem, considerably more serious than the nonconformity itself; signs of erased data undermine the credibility of the entire dataset.

Signs of abuse. A retest rate that is noticeably high, with some items routinely retested two or three times. Reasons recorded in general terms, with multiple records citing operator error without specifics. Only the conforming retest appearing in the report while the original data are absent. Retests authorised by the operator who performed them, with no independent review. And inconsistent application of the rule between batches. These patterns are easy to see statistically: summarising retest records over a period makes the rate and the distribution of reasons immediately apparent, and manufacturers would do well to run that summary themselves periodically.

Retesting and batch disposition. The conclusion affects disposition of the whole batch, and a conforming retest on one sample does not make a batch conforming. If the original failure stemmed from a measurement problem, a conforming retest establishes that the sample conforms and batch judgement returns to the normal route. If the original failure reflected batch variability, wider sampling is needed even after a conforming retest. State this in the rule, or practice will simplify it to releasing on a conforming retest.

How we handle it. When a result is nonconforming, we first check whether the test process itself was sound, covering equipment, conditions and execution. Where a problem is found, we repeat the test. Where none is found, we issue the nonconforming result as measured and record the conditions completely. When a client requests a retest, we record the request and the reason, and the report reflects how many times the test was run. That sometimes makes a report look less tidy, and it preserves the integrity of the data. A report presenting only the conforming run and omitting what preceded it is a liability in any later traceability exercise.

Send us the situation and we will work through it. Phone or WeChat: +86 132 4819 8029.