Rework and Reprocessing: One Word Changes the File

A batch misses assay by four tenths of a per cent. There is material available, the customer is waiting, and somebody suggests running it back through the crystalliser. Whether that is a routine correction or a change to the validated process depends on one question: was that step already in the file, in that position?

The answer determines the documentation, the notification and, in some markets, whether the batch can be released at all.

Two words, one boundary

Reprocessing returns material to an earlier step of the approved process and repeats it. Rework subjects material to a step the approved process does not contain.

The boundary is positional. If the process says crystallise, centrifuge, dry, mill, then repeating the crystallisation on dried material is not the same act as crystallising material that had not yet been dried. The first repeats a step; the second creates a new sequence. Chemically the operations may look identical. In the file they are different events.

	Reprocessing	Rework
Position in the process	Repeats an existing step	Introduces a step or a new sequence
Validation status	Covered if the step was validated for repetition	Not covered by the existing validation
Effect on impurity profile	Usually small and anticipated	Potentially significant and unstudied
Customer notification	Often summarised	Usually requires agreement in advance
Regulatory treatment	Described in the process	Typically a defined, approved operation

Why the crystalliser is the classic case

Recrystallisation is the most common recovery step in a small-molecule plant because it addresses several failure modes at once: purity, colour, crystal form, particle size. It is also the step with the largest hidden effect, because it changes the solid-state history of the material.

A batch that has been dried, milled and then redissolved has a different thermal and solvent history from one that was crystallised once. That affects residual solvent, particle size distribution and, in some cases, polymorphic form. None of these necessarily appears on a certificate unless the specification includes them, and a specification that includes them is the exception rather than the rule.

The impurity profile is the other place the effect shows. Recrystallisation removes some impurities and concentrates others in the mother liquor, which is then discarded. A second crystallisation can reduce one related substance and increase another, and the net effect on total impurities may be close to zero while the individual numbers move.

The stability clock restarts

Stability data attaches to a process, and the material produced after a rework step has a different process history. Whether the existing stability package describes it is a question, not an assumption. In practice, a reworked batch is often placed on stability alongside the regular batches, and the batch used for the customer's file should be one that has been.

This is the part that surprises buyers. Two certificates that look identical can describe material with different stability evidence, and the difference is invisible unless somebody asks which batches went into the stability package.

Blending is not a third option

Adding a failing lot to two good ones until the composite meets specification is a different act again, and in the pharmaceutical context it is generally treated as a prohibited practice rather than a recovery technique. It does not correct the non-conformity; it dilutes it. The distinction from legitimate blending is intent and disclosure, and the burden of showing that intent is on the site.

Where blending is a genuine process step, it is validated, it has a uniformity check, and it is described. Where it appears only when a batch is in trouble, it is none of those things. This is treated separately in the note on batch blending, because the uniformity question is technical rather than documentary.

What the record should show

Four items. The reason the batch failed, in terms of the attribute and the limit. The step performed, with parameters, referenced to a written instruction. The results before and after, including the tests that were not repeated and why. And the disposition decision, with the name of the person who made it.

A record that shows only the before and after results, with no written instruction behind the step, is describing an improvisation that happened to work. That is a legitimate thing to have done once, in an emergency, in a plant that then writes the instruction; it is not a legitimate way to run a process.

What a buyer should ask

Three questions, all answerable in a sentence. Has any batch supplied to us been reprocessed or reworked? If so, which, and what was the basis for release? And does the quality agreement require notification before the fact?

The third question is the one that changes behaviour, because it converts a discretionary decision made under delivery pressure into a contractual one made before the pressure arrives. It belongs in the change notification categories rather than in a general clause about compliance.

Where this leaves a supply-chain partner

We describe what was done to a batch in the terms above, distinguish repetition from rework, and treat rework as a notification event rather than an internal recovery. Where a customer's file requires stability data on the released material, we say which batches support that file rather than assuming the class of process is enough.

One question worth putting to any supplier: in the last two years, how many batches were recovered by a step that is not in the process description? A supplier with the number has a system that counts. A supplier without one has a system that remembers, and memory is selective under delivery pressure. The batch consistency record is where that memory becomes visible as a trend.

A recovery scenario, step by step

A batch of material fails the assay limit by three tenths of a per cent after drying. The process description includes a crystallisation followed by centrifugation, drying and milling. Three routes are available, and they are not equivalent.

Route one: re-dissolve and crystallise. The material has already been dried and milled, so returning it to the crystalliser is not a repetition of the approved sequence. It is a new sequence, and it is rework. It needs a written instruction, a rationale, an assessment of what the second crystallisation does to residual solvent and particle size, and usually a conversation with the customer.

Route two: re-dry. If the assay failure traces to residual moisture and the drying step has a validated range that permits a further period, extending the drying within that range is a repetition of an approved step. It is reprocessing, documented as an event, and it does not change the process description.

Route three: blend with a compliant lot. This corrects nothing. It dilutes, and in the pharmaceutical context it is generally treated as a prohibited practice rather than a technique. The honest version of the same commercial need is a validated blending step that is part of the process and applied to every batch, not a rescue applied to one.

What the customer's file needs

Material from route one is not the material described in a file written around the original process. The impulse at that point is to note that the certificate is identical, which is true and irrelevant: a certificate describes tests, and the file describes a process. Where a customer's regulatory submission names the process, a reworked batch may require a variation or at least a documented assessment.

The practical requirement is disclosure before the fact, and it belongs in the change notification categories so that it is discussed under calm conditions rather than at the end of a quarter. The batch consistency record then shows whether the recovery was an isolated event or a habit.

The framing is public and consistent across markets: the ICH Q7 guidance addresses reprocessing and rework for active substances, the EudraLex volume 4 text sets the European expectations for the process description and for changes to it, and the USP general chapters define the tests that would have to be repeated afterwards. A supplier whose recovery route cannot be described in those terms has improvised.

The question worth asking is small and revealing: how many batches in the last two years were recovered by a step that does not appear in the process description? A number, even a small one, is a system. A pause is not.

Second crops and what they inherit

A second crop is the material that crystallises from the mother liquor after the first crop has been removed. It is a legitimate recovery of yield, it appears in many processes, and it has a defining property: it is enriched in whatever stayed in the liquor. The related substances that did not fit into the first crop's lattice are more concentrated in the second.

That makes the second crop a different material from the first, produced by the same process description. Where the process anticipates it, the second crop is combined or segregated by a rule, and the rule is part of the validation. Where it is not anticipated, a second crop used to recover a shortfall is rework by the definition above, with the added complication that its impurity profile is predictably worse.

The certificate cannot show the difference unless the specification's limits are tight enough to separate the two materials. That is the quiet risk: a second crop that meets a wide specification is indistinguishable on paper from a first crop, and the only record of the difference is the batch record.

The practical question for a buyer is whether the process anticipates second crops, and under what rule they are combined. A supplier who can answer has thought about it. A supplier who describes the practice as "sometimes, when yield is short" has described a variable process that the certificate cannot represent.

Frequently asked questions

What is the difference between rework and reprocessing?

Reprocessing takes material back to an earlier step of the process that was already approved, and repeats it. Rework subjects material to a step that is not part of the approved process at all. The first is inside the file; the second changes it.

Is reprocessing always allowed?

Where the process description permits it, the step has been validated for that purpose, and the specification is still met afterwards. It should also be recorded as an event rather than performed silently, because the batch record has to show that the material went back a step and why.

Why is rework treated more seriously?

Because a step that is not in the approved process has not been validated, and its effect on the impurity profile, on crystal form and on stability has not been studied. Regulators generally expect rework to be a defined, approved and controlled operation rather than an improvised recovery.

Does recrystallisation count as rework?

If crystallisation is already a step in the process, repeating it is reprocessing. If the material had already been dried and milled and is now being dissolved and crystallised again to fix a failing result, that is usually rework, because the sequence has departed from the approved route.

Must a customer be told?

Under most quality agreements, yes, and the agreement should say so before the situation arises. A reworked batch is a different batch from the one the customer's file describes, even when the certificate looks identical.

What about blending a failed batch into a good one?

That is a separate and generally prohibited practice in the pharmaceutical context, because it disguises a failure rather than correcting it. The distinction is intent: correcting a defined attribute within an approved step is one thing, diluting a non-conforming lot is another.