A Result on the Limit: Trend, Method Precision, or Reject?

A result that passes by a thin margin creates an awkward situation for a reviewer. Nothing is wrong: the number is inside the limit, the batch conforms, and the certificate can be signed. Something is nevertheless unsatisfying, and the instinct is to do something about it. The instinct is right; the impulses it usually produces are not.

This note sets out the three inputs that actually decide whether a narrow pass is acceptable, why a re-test is normally the least useful response, and what the margin is really telling you about.

What a narrow pass is a statement about

The margin between a result and its limit is not a statement about the material in that drum. It is a statement about the process that produced it and about the method that measured it. A batch that passes narrowly has been produced by a process with little remaining room, and it has been judged by a method whose own variability may be of the same order as the margin.

Keeping those two things separate is the whole of the analysis. If the process has room and the method is precise, a narrow pass is a coincidence and needs no action. If the process has no room, the next batch may fail regardless of how precise the method is. If the method is imprecise at that concentration, the number itself is less informative than it appears, and the correct response is analytical rather than commercial.

The first input: the trend across batches

One result cannot establish a trend, and a trend is what turns a single number into a decision. The same value that looks comfortable against a stable series looks alarming against a series that has been climbing. This is why batch-to-batch consistency is the right framework for reading a narrow pass, and why a buyer who keeps the last five or ten results has far more information than one who reads certificates one at a time.

The pattern to look for is straightforward. A flat series with one point close to the limit is a single-batch question. A series that has been moving towards the limit over several batches is a process question, and the process question is the one that needs an answer before the next order rather than after it. Related substances limits explains how the individual limits are built, which is useful context for judging how much room a given line normally has.

The second input: the method's precision at that level

A limit is a single number and a measurement is a distribution. At concentrations near a limit, repeatability matters more than accuracy, because what the buyer needs to know is how often the method will produce a value on the wrong side of the line for material that is genuinely acceptable, and how often it will produce a comfortable-looking value for material that is not.

Where the method's repeatability at the relevant concentration is a meaningful fraction of the margin, the margin should be discounted accordingly. A specification whose typical result sits half a repeatability unit away from its limit is not really specifying control; it is specifying that the result will usually pass. The practical test is simple: if two independent determinations on the same sample could straddle the limit, the margin is too small to carry the weight being placed on it.

There is a related point about the limit of quantitation. Where the limit sits close to the level at which the method can no longer quantify reliably, results near it carry more uncertainty than the figure suggests. In those cases the honest response to a narrow pass is to say that the measurement near that limit is uncertain, which is a different and more useful statement than a re-test.

The third input: the margin in your own specification

Buyers often hold an in-house specification tighter than the released one, and the margin that matters to them is the margin against their own limit rather than against ours. A batch that passes our limit with room to spare may pass their tightened limit only narrowly, and that difference changes the conversation.

Where those two limits differ, the useful thing to establish is not which one applies but what the buyer intends to do with the material. A tightened limit usually encodes a downstream requirement, and the response to a narrow pass against it should be about whether that requirement is at risk, not about whether the certificate is valid.

The decision, as a table

Situation	What it indicates	Useful response
Narrow pass, flat series, precise method	A coincidence in a stable process	Record it; no action
Narrow pass, climbing series	The process is losing margin	Raise it as a process question before the next order
Narrow pass, method repeatability of the same order as the margin	The measurement cannot resolve the question	Ask for precision data, not a second result
Narrow pass against your own tighter limit	A downstream requirement may be at risk	Discuss the requirement rather than the certificate
Result outside the limit	Out of specification	Follow the specification's own rejection path

Only the last row is a rejection case. The other four are all conversations of different kinds, and the value of separating them is that the right conversation happens instead of a re-test.

Why a re-test usually makes it worse

When a result is close to a limit, the most common request is for another determination. The request is understandable and usually counterproductive. If the method is precise, the second result will agree with the first and settle nothing. If the method is imprecise, the second result will differ, and the buyer now has two valid numbers and no rule for choosing between them, which is a worse position than having one.

What the re-test displaces is the trend. The value of the previous batches is that it was generated independently of this question, so it carries information the re-test cannot. Asking for the last ten results is a better use of the same week than asking for a second determination of the same sample.

What we do when a result sits at an edge

We report it. An edge result that is disclosed with its trace and its context is more useful to a reviewer than a result that happens to be comfortable, and disclosing it costs nothing when the batch conforms. Alongside the result we state the method and its repeatability at that concentration, and we supply the previous results for the same impurity so that the trend can be read without a round of correspondence.

Where the trend shows that a line has been creeping towards its limit, we say so before the buyer asks. A specification is a set of limits, but a supply relationship is a set of expectations about notice. Retest period versus shelf life makes a similar point about time-dependent information: the number matters, and what matters more is whether it arrives while there is still time to act on it.

The statistical view, without the theory

Three quantities determine whether a narrow pass is meaningful, and none of them requires statistics to use. The first is the distance between the result and the limit. The second is the repeatability of the method at that concentration, which is the size of the difference two determinations on the same sample can reasonably show. The third is the spread of previous results for the same substance.

The comparison is a simple one. If the distance to the limit is several times larger than the repeatability, the result is informative and the margin is real. If the distance is comparable with the repeatability, the result is close to a coin toss and the specification's control over that property is weaker than the limit suggests. If the distance is smaller than the repeatability, the measurement cannot distinguish material that conforms from material that does not, and no number of repeat determinations will change that.

The spread of previous results does the rest of the work. A result close to the limit in a series with a wide spread is less remarkable than the same result in a series that has been flat for a year. The value of keeping the history is that it converts a single number into a position within a distribution, and positions are what decisions are made on.

What a paper trail should show

When a result sits at an edge, the documentation matters as much as the number. A reviewer should be able to see the original determination, the conditions it was run under, the standard used, and the result in context with the previous batches. Where those four items are available, the question is usually settled by reading rather than by testing.

The general chapters of the European Pharmacopoeia describe the techniques and the suitability requirements that support the number, and the ICH quality guidelines describe the validation expectations that establish how much confidence a method's results can carry near a limit. Both are general; neither produces the specific repeatability figure a buyer needs, which has to come from the laboratory that ran the method.

That specific figure is worth requesting once and keeping. A laboratory that reports the repeatability at the relevant concentration alongside the result has answered the question before it was asked, and the practice costs very little.

How a margin should be agreed

Where a buyer needs a specific margin, the place to establish it is the specification rather than the correspondence. A limit that is set at the pharmacopoeial value with no margin is a limit that will be met by producing material that sits near it, which is what a supplier optimising against a limit will do. A limit set tighter, with the reasoning recorded, produces material with room.

Two practical forms are in use. The first is an internal action limit, set below the specification limit, which triggers an investigation without making the batch non-conforming. The second is a trend rule, under which a result outside the established pattern is investigated regardless of its distance from the limit. Both are described in the regulatory guidance documents that deal with investigation and data integrity, and both are more effective than a re-test.

Where a buyer and a supplier agree an action limit, it should be recorded where both can find it. An action limit that lives in a set of emails is an expectation rather than a commitment, and it will be discovered to be either too tight or too loose at the least convenient moment.

What a narrow pass says about the specification itself

A specification whose results habitually sit close to its limits is not necessarily a bad specification, but it is a specification that is not doing the work it appears to do. A limit is meant to separate acceptable material from unacceptable material, and if the process can only produce material near the line, then the line is close to the process capability and the separation is thin.

Two readings of that situation are legitimate and they point in different directions. The limit may be set beyond what the process can comfortably achieve, in which case the limit is the problem and the specification should be revisited against the process capability and the downstream requirement. Or the process may have drifted towards the limit over time, in which case the specification is fine and the process needs attention.

Where a buyer and a supplier disagree about which reading applies, the disagreement is usually resolvable by looking at the results from the period when the material was first qualified. Material that was comfortable then and is marginal now describes a drifting process. Material that has been marginal throughout describes a limit that was never comfortable.

Frequently asked questions

Is a result close to the limit an out-of-specification result?

No. Out of specification means outside the limit. A result inside the limit, however close, conforms. The question it raises is different: whether the margin is large enough to be relied on, given the method's precision and the trend across previous batches.

Should I ask for a re-test when a result is close to the limit?

A re-test is usually the least informative response. It cannot change the original result, and where the method's precision is wider than the margin, the second value will not settle the question either. Trend data and method precision settle it better.

What is the difference between out of trend and out of specification?

Compliance is judged against the limit; trend is judged against the previous results for the same material. A result inside the limit but well away from the established pattern is out of trend, and out-of-trend results are investigated without the batch necessarily being rejected.

How much margin is enough?

There is no universal number. A workable rule is that the margin should be comfortably larger than the method's repeatability at that level, and it should be stable across batches. Where the margin is inside the method's own variability, the result is effectively a coin toss repeated each batch.

Does a thin margin mean the batch should be rejected?

Not on its own. It means the batch conforms with little room, and that the next batch carries real risk. The decision belongs with the buyer's quality function and depends on what the material is used for and how much notice they need of a supply interruption.

What data should I request to judge it properly?

Three items: the results for the same impurity across the last five to ten batches, the method's repeatability at the concentration in question, and the limit set in your own in-house specification if it differs from ours. Those three usually decide the case.