Batch-to-Batch Consistency: Reading Three Consecutive COAs
Share
A single certificate of analysis is a statement about one batch. It is a good document and it answers a real question, but the question is narrow: did this batch meet the specification? It says nothing about whether the next batch will, and nothing about whether the process that made it is steady. Consistency is a property of a series, not of a batch, and it is invisible on a single sheet of paper.
That is why buyers who take supply risk seriously ask for three consecutive certificates rather than one. The habit has a real basis in the guidelines, and it also has a real limit. Both are worth knowing before you build a qualification decision on it.
Where the three-batch convention actually comes from
The convention is borrowed from process validation, and the borrowing is often careless. The texts that establish it are specific, and they are more cautious than the way the number gets used in purchasing.
| Text | What it actually says |
|---|---|
| ICH Q7, Section 12.5 (Process Validation) | For prospective and concurrent validation, three consecutive successful production batches are to be used as a guide. |
| EU GMP Annex 15, 5.19 and 5.20 | The producer must determine and justify the number of batches; a minimum of three consecutive batches under routine conditions is generally considered acceptable. |
| FDA process validation guidance (2011) | No fixed number. The guidance asks for enough samples to give statistical confidence of quality both within a batch and between batches. |
Two things follow from reading those three lines together. The first is that three is a defensible default rather than a legal minimum, and a supplier who treats it as a magic number has not read the guideline. The second is that the same three-batch set is being asked to do two different jobs: it is evidence that the process has been validated, and it is the data a buyer uses to judge consistency. The jobs are related but not identical, and the second one needs a different kind of reading.
Annex 15 also states, plainly, that retrospective validation is no longer an acceptable approach. That removes a shortcut that used to exist: you cannot validate a process by assembling a set of historical batches after the fact. The series has to have been made prospectively, which is precisely why a real three-batch set is worth asking for and a curated one is not.
What a single certificate cannot show
Three parameters decide whether a batch is healthy, and all three are invisible on one sheet.
- Position within the limit. An assay of 99.4 per cent against a 99.0 to 101.0 per cent window sits comfortably inside. The same figure means something different if the three previous batches were all at 99.9. A result is only good or bad relative to the distribution it comes from.
- Direction. A drying-loss figure of 0.4 per cent against a 0.5 per cent maximum is a pass. Four successive batches moving 0.1, 0.2, 0.3, 0.4 is a trend, and the fifth batch is the one that will fail.
- Spread. Two suppliers can both sit in the middle of the window, one with a batch-to-batch range of 0.2 per cent on assay and the other with 1.6 per cent. The second is closer to a failure even though today's number is identical.
None of this is exotic statistics. It is the same reasoning a production manager applies to any measured process, and it is available to anyone holding three consecutive certificates and ten minutes.
How to read three certificates together
Lay the three out side by side and read down each line rather than across each sheet. The order matters: reading certificate by certificate invites you to approve each one on its own, which is exactly the habit that hides a trend.
Keep one thing straight while you compare. All three certificates are the producing site's own reported results, measured in its own laboratory. A series tells you whether that one process is steady; it adds no independent evidence to any single value. A flat line is reassuring, and equally consistent with a test method that has never been challenged — which is the argument for independent verification alongside the series, not instead of it.
| What to compare | What a steady process looks like | What asks a question |
|---|---|---|
| Assay | Three values clustered in a narrow band, not pressed against either end | A wide spread, or three values marching in one direction |
| Impurity total | Flat, with individual impurities wandering without a pattern | A total that creeps upward while each individual result stays legal |
| Drying loss or water | Stable, and consistent with the packaging and the season | A slow rise, which points to storage or to a changed drying step |
| Particle size, where specified | Inside the window with a stable spread | Sitting at the edge, or a sudden tightening that signals a harder milling regime |
| Specific rotation | Near the middle of the window | Drifting towards an edge, which is a stereochemical signal |
The single most useful pattern in that table is an impurity total that rises while the assay stands still. Our note on whether an assay method can see degradation sets out the underlying reasoning. It belongs in this comparison because a batch-by-batch reading will never surface it: every certificate passes, and the series is telling you something.
Within specification is not the same as in control
The phrase to keep in mind is process capability. A process that produces results comfortably centred in the specification window is capable; a process that produces legal results scattered towards the limits is not, even while every certificate passes.
Producers put a number on this. The capability index compares the distance from the process mean to the nearest specification limit against the spread of the process, and an index of about 1.33 — four standard deviations of room on each side — is a common internal target for a critical attribute. A supplier who monitors capability this way keeps a control chart. A supplier who does not keeps three certificates and an assurance. Both may be selling the same material; only one of them can tell you what the fourth batch will look like.
There is a second reason this matters, and it is regulatory rather than statistical. A batch record review confirms that a batch was made as intended. It does not confirm that the process is stable over time. Those are separate obligations, and satisfying the first while ignoring the second produces exactly the gap that appears as a slow drift in a customer's incoming data rather than as a deviation in the producer's own records.
Why a release series is not stability data
The three certificates you receive are release data. They were generated on fresh material, tested when the batch was made, and they describe the batch at time zero. They are not stability data, and no number of release certificates will substitute for one stability series.
Stability data answers a different question: how do these results move as the material ages, under defined storage conditions, over the re-test period? It is generated on the same batch over time, not across three batches at one time. A supplier can have a perfect release series and no stability series at all, and the two absences look very different from the outside.
The practical point for a buyer is that a stable release series is evidence about the process, and a stability series is evidence about the material's future. When you ask for one, be clear which one you mean, because suppliers answer the easier question by default. Our note on what a re-test date actually requires describes what the second document should contain.
What consistency actually buys you
There is a commercial answer, and it is worth stating because it is the reason a supplier should want to provide the series rather than resist it. A source whose results have been stable across a series of batches is a lower-risk source, and lower risk is the justification for reducing the testing you do on receipt.
That reduction is not a shortcut invented by buyers. It is the normal consequence of supplier qualification: a buyer evaluates the source, confirms that the quality system is in control, and then reduces incoming testing on the basis of that performance. The series is what makes the decision defensible. Without it, every batch has to be tested as though it were the first.
For the supplier, that means the series is not a burden imposed by the customer. It is the evidence that lets the customer stop testing everything, which is a cost saving on both sides. A supplier who cannot produce it is asking the buyer to carry the entire risk of the relationship on every shipment.
The limit of the comparison
Three certificates are comparable only if the conditions around them were comparable. A set that spans a scale-up, a change of equipment, a change of analytical method or a revision of the specification is not a single series, and variation inside it may reflect the change rather than the process.
Two cautions follow. The first is to ask whether the three batches were made under the same conditions before reading them as a series. The second is to check whether the specification itself moved between them: a limit that changed can make a steady process look inconsistent, or hide one that is not steady. The revision cycle that drives those changes is regular and predictable, and it is the subject of a separate note on what a pharmacopoeial revision does to stock already released.
What to ask for
- Three consecutive certificates, with the batch numbers visible. Consecutive matters. Three certificates picked from across a year are not a series.
- A statement of whether the batches were made on the same equipment and at the same scale. A set spanning a scale-up is not a consistency set.
- The trend of the impurity total, not only the individual results. If the supplier tracks it, this is a one-line answer. If not, the question itself is informative.
- Whether the specification changed between the three batches. A moved limit can make a steady series look inconsistent, or hide one that is not steady.
- Whether any batch was reworked, re-milled or re-tested before release. Each of those leaves a mark on the numbers that consistency reading depends on.
None of these questions requires the supplier to disclose anything commercially sensitive. All five are answerable from records the quality unit already keeps.
Where this leaves you
Three certificates are the minimum honest sample of a process, and they are useful only when read as a series. Look for position, direction and spread before you look for pass or fail, and treat an impurity total that rises while the assay holds flat as the finding rather than as background noise.
What we release against, and how a batch series is documented, is set out on our ursodeoxycholic acid page. If you want a batch series read against your own acceptance criteria before you place an order, you can raise it through a quotation request. For what each individual line on those certificates proves, see our breakdown of the certificate of analysis, and for the arithmetic behind the impurity line, how the related-substances limits add up.
This article describes documentation, statistics and pharmacopoeial testing for quality, regulatory and procurement professionals. It states no therapeutic use, dose or indication for any substance, and no order quantity or lead time. Confirm all requirements against the current published version of the referenced guidance.
Frequently asked questions
Why do buyers ask for three consecutive certificates?
Because consistency is a property of a series rather than of a batch. One certificate tells you whether a single batch met its specification; three consecutive ones let you see position within the limit, the direction of travel and the spread, which together describe the process rather than the batch.
Is three batches a regulatory requirement?
It is a widely used default rather than a fixed legal minimum. ICH Q7 states that three consecutive successful production batches are to be used as a guide for prospective and concurrent validation, and EU GMP Annex 15 says a minimum of three consecutive batches is generally considered acceptable once the number has been justified. The FDA process validation guidance sets no fixed number and asks instead for statistical confidence within and between batches.
What is the difference between a batch being within specification and the process being in control?
A batch is within specification when its results sit inside the limits. A process is in control when the results are centred and their variation is small relative to those limits. A process can be out of control and still produce legal batches, which is why a series read is more informative than a batch read.
Which line should I look at first when comparing certificates?
The impurity total, read against the assay. An impurity total that creeps upward while the assay stays flat is the pattern that batch-by-batch reading cannot surface, and it points to a method or process question rather than to a single bad batch.
Can a certificate set from different equipment still be compared?
Only with care. Three batches made across a scale-up, on different equipment or after a specification change are not a single series, and variation between them may reflect the change rather than the process. Ask whether the batches were made under the same conditions before comparing the numbers.
What does a consistent batch series buy me commercially?
It is the evidence that supports reduced incoming testing, because a supplier whose results have been stable over a series of batches is a lower-risk source than one that cannot show a series. It also gives you a reference point for judging a future batch that looks different.