Batch Numbering: What the Batch Number Can and Cannot Prove

A batch number is the most quoted string on a certificate and the least examined. Buyers use it as shorthand for "this drum and that document belong together", which is correct, and then extend it to mean "everything in this drum is the same", which the number was never designed to establish.

The distinction matters because the number is what an investigation reaches for first. If the number is a reliable index, the investigation has a route. If it is a label applied at packing, the route ends at a warehouse shelf.

The four jobs a batch number performs

In the pharmaceutical framework, the batch is the unit of disposition. That single decision produces four consequences, and each one has a different evidence requirement.

Job	What the number does	What has to exist behind it
Link	Connects container to record	A batch record retrievable by that number
Release	Defines what is accepted or rejected together	A disposition decision covering the whole unit
Recall	Defines what would be withdrawn together	A distribution list keyed to the same number
Stability	Defines what the stability data describes	Material from that unit, stored under the stated condition

Notice what is absent from the table: purity, homogeneity, age. None of those are jobs of the number. They are claims made in other documents, and the number only tells you which documents to read.

Three things the number cannot prove

It cannot prove homogeneity. A single number asserts one process description and one disposition. Whether the twenty-fifth kilogram of a two-hundred kilogram lot matches the first is a question about mixing and segregation, answered by a uniformity study and by what the blending step was designed to do. A drum can carry a valid number and a stratified particle size distribution at the same time.

It cannot prove chronology. Numbering by campaign, by customer order or by packing date all produce sequences that look chronological and are not. A number issued in November can describe material synthesised in March and held as bulk. That is not a defect. It is a reason to read the manufacturing date rather than infer it from the code.

It cannot prove the unit is closed. This is the one that fails quietly. A lot is packed into forty drums, three are held back for retesting, two are repacked into smaller packs for a sample order, and one is relabelled for a different market. Each of those actions creates a portion with its own history, and if all of them carry the same number with no sub-lot identifier, the number now describes five different things.

Anatomy of a numbering scheme

Most bile-acid sites use a compound code: a year or year-month element, a sequence, and often a line or campaign letter. The scheme is arbitrary. What matters is whether it is written down, whether it is applied without exception, and whether the record can be retrieved from it years later.

A code that encodes the month of packing is common and mildly misleading, because the certificate's manufacturing date and the code's month can differ by the hold time of the bulk. A code that encodes the campaign is more informative for investigations, because it groups material made under one process state. Neither is wrong. A scheme that is not documented is wrong, because it cannot be audited.

Where traceability actually breaks

Not at the reactor. Traceability breaks at the boundaries between organisations of work: sampling, blending, repacking, relabelling, and splitting for a smaller order.

Sampling is the mildest case. A sample drawn from a drum and sent to a laboratory acquires a laboratory number, and the link between the two numbers lives in a logbook. If that logbook is paper and the batch record is electronic, the link is the weakest point in the chain.

Blending is the sharpest case. Two lots combined into one produce a unit that is genuinely new, and the question of whether the components remain visible in the file is a documentation decision, not a chemical one.

Repacking is the most common case and the least discussed. Material transferred from a fibre drum into five aluminium packs for a trial order is still the same material, but it is now five containers with five opening dates and five potential contamination events. The original number should travel with each pack, together with the repacking record and the new container identity. A repack that issues a fresh number with no cross-reference looks tidy and destroys the route back.

The repack boundary, stated plainly

A workable rule: a new number is required when the material changes state, composition or process step. A sub-lot identifier is required when a portion of one batch becomes independently releasable, independently storable for a different period, or independently shippable to a different customer. A repack that does neither, and only subdivides for handling convenience within one site, can keep the original number.

The test is not administrative tidiness. It is whether two portions can later be in different conditions and still be described by one record. If they can, the number is doing a job it cannot do.

What this means for a buyer

Three questions separate a numbering system from a label. Where in the process is the number assigned? Does repacking create a sub-lot identifier, and is it printed on the pack? Can the batch record behind a given number be produced during an audit, without a week of searching?

The third question is the only one that tests the system. The first two can be answered from a procedure. The third requires the site to have actually kept the record, in a form it can still read, for as long as the material can be on the market.

We treat the batch number as an index into a file, and we say so when a customer asks what it guarantees. The document pack and the batch history carry the claims. The number carries the address. The general expectations for batch records and for traceability are set out in the ICH quality guidelines, the European GMP text and the PIC/S guides, and none of them asks a number to do more than point.

The question worth putting to any supplier is narrower than "is it traceable": if we picked one drum at random and asked what happened to it between synthesis and loading, how long would the answer take, and would it be a record or a recollection?

What happens to a numbering scheme when the site grows

A scheme that works for one reactor and one packing line starts to fail at the first structural change, and structural change is normal. A second site is qualified. A tolling arrangement is added for a step that the main site cannot run. A company is acquired and its stock arrives with numbers from a different convention. None of these are unusual, and each of them breaks the assumption that a number is unique within the supplier's world.

The failure is quiet. Two sequences collide, one from each site, and the collision is discovered when a customer quotes a number and the answer contains two records. From that point the number needs a site prefix, and every downstream label, certificate and quality agreement has to change, which is why the prefix should have been there from the beginning.

A workable convention encodes four things: the site, the year, the campaign or sequence, and optionally the packing run. The site element is the one that gets omitted because it seems redundant while there is only one site. It stops being redundant at the worst possible moment, which is always after the second site has shipped.

Where contract steps sit in the chain

Many bile acid supply chains include a step performed elsewhere: a purification, a milling operation, a drying step. The material that leaves for that step and returns is not obviously a new batch, and whether it becomes one is a decision that should be written down.

Two defensible conventions exist. The batch number is retained and the external step is recorded in the batch record as an intermediate operation, with the receiving inspection and the return quantity. Or a new number is assigned at return, with a cross-reference to the outgoing number. The first keeps the chain linear. The second keeps the site's own numbering clean but requires the cross-reference to be reliable, because without it the customer's number leads to a record that begins with material appearing from nowhere.

What is not defensible is the third pattern, which is common: the material returns, is combined with fresh material, and the resulting lot takes the number of whichever portion was larger. That is a blend under a name that suggests a single history, and it is the same problem as blending described as a routine step, arrived at by a different route.

A five-minute test a buyer can run

Take three certificates from different years, ideally spanning a change of site or process. Read the batch numbers and ask what each element encodes. Then ask whether the same number could appear twice in the supplier's system under any circumstances.

A supplier who can answer both questions from memory has a scheme. A supplier who has to check with the plant has a convention, which is weaker but workable. A supplier who says the numbers are generated by the system has described a database, not a traceability design, and the next question is whether the system prevents duplicates across sites.

None of this is about the certificate. It is about whether the certificate can be followed back, and the follow-back is what an investigation needs. The batch history is only as good as the index that reaches it.

Frequently asked questions

What is a batch number actually for?

Four jobs. It links a container to a production record, it defines the unit that is released or rejected as a whole, it defines the unit that would be recalled, and it defines the unit that stability data attaches to. Everything else people read into it is inference.

Does a sequential number prove the batch was made after the previous one?

No. Sequencing is a convention, not a rule. A site may number by campaign, by customer order or by packing date, and a number that looks later can belong to material synthesised earlier. Only the batch record establishes the order.

Can a batch be split without a new number?

It can be packed into many containers under one number, and that is normal. What is not normal is splitting material into packs that are then released, labelled or stored as separate units without a sub-lot identifier. The moment two portions can be dispositioned differently, the single number no longer describes the unit.

What happens to traceability when material is repacked?

The original number should travel with the new pack, together with the repacking record, the new pack date and the new container identity. A repack that produces a new number with no cross-reference severs the link to the original batch record, and that link is what an investigation needs.

Is a batch number evidence of homogeneity?

Not by itself. One number asserts that the material was produced under one process description and released as a unit. Whether it is uniform inside the container is a separate claim, supported by blending validation and sampling, not by the label.

What should a buyer ask about numbering?

Where in the process the number is assigned, whether repacking creates a sub-lot identifier, and whether the batch record behind a given number can be produced on request. The third question is the only one that tests the system rather than the label.