What is a split batch? Dividing a Batch or Lot While Keeping Traceability

A split batch is an existing manufacturing batch or inventory lot divided into two or more separately tracked portions that retain a traceable link to the original batch or lot. Batch split, lot split and split lot are common aliases, not universally standardised legal categories. Operationally, the defining relationship is one-to-many: one identified source becomes several distinguishable portions, rather than unrelated stock.

A physical portion, a container identifier and a separately numbered child lot are different things. A portion is material; a container identifier identifies its holding unit; a child lot identifier establishes a separately tracked inventory identity. A physical subdivision can retain the original manufacturing batch number while adding container or internal child identifiers. The identifier convention alone does not establish a new manufacturing batch. This distinction matters when designing batch identifiers and bin-level lot genealogy: a new tracking reference need not mean new manufacture.

These are operational comparisons, not regulatory classifications. An operation can involve more than one activity: splitting during repacking changes both the portion structure and the packaging. Conversely, dispatching fewer units than are available does not necessarily create child lots. Describe what happened to the material before choosing the transaction name; the actual activity and applicable framework determine its regulatory classification.

For APIs, the ICH Q7 glossary (section 20), which is GMP guidance, defines a batch (or lot) as a specific quantity of material produced in a process or series of processes so that it is expected to be homogeneous within specified limits, and a batch number (or lot number) as a unique combination of numbers, letters and/or symbols that identifies a batch and from which the production and distribution history can be determined. Dividing a homogeneous batch into portions does not change that production history, which is why child portions normally keep a traceable link to the parent number.

Sites split batches to match material availability to equipment, packaging and distribution needs. Representative operating reasons include:

• Supplements: allocate bulk capsules to separate bottle-count packaging runs.
• Food: divide a bulk mixture into portions that fit smaller vessels; equipment-capacity limits belong in constraint-based scheduling.
• APIs: divide an intermediate for staged processing, with one portion proceeding while another waits.
• Devices: allocate portions of a component lot to different customer orders or assembly destinations.
• Chemicals: route portions to separate storage locations and set aside retained material for later examination.

These examples describe operating choices, not interchangeable regulatory requirements. A planned split assigns portions deliberately; an unplanned remainder or interrupted transfer leaves material whose allocation still needs resolution. Record the reason and intended destination for each portion so a later reviewer can understand why it exists. Describe an interruption as an interruption, rather than retrospectively calling it a planned split. Subdivision changes how material is managed; it does not by itself improve quality or homogeneity.

The following seven-step workflow is recommended operational practice, not a universal regulatory procedure. Set its responsibilities, transaction boundaries and completion criteria in the site procedure before routine execution.

1. Select the source. Identify the exact source lot, physical containers, storage location, available quantity and units. Check current restrictions and specify the intended destinations before material moves; a lot number alone does not identify which container to open.
2. Assign responsibility. Name the person executing the split and the person reviewing its record. Identify who can resolve discrepancies and who makes the applicable quality-status decisions, rather than leaving those decisions to the operator by default.
3. Reserve the quantity. Where picking or dispensing can occur concurrently, reserve the proposed amount against the selected stock. Use allocation arbitration to prevent competing requests from allocating the same quantity twice.
4. Capture actual transfers. Record what physically moves, not a copy of the planned quantities. Associate each portion with its receiving container, operator, time and relevant equipment through contemporaneous recording.
5. Post one logical event. Link the source debit and portion credits under one split-event reference. Account for any quantity retained in the parent containers without crediting that remainder twice. Keep partially recorded transfers identifiable as incomplete.
6. Review the result. Reconcile quantities and resolve label-to-container associations. Route discrepancies to the assigned reviewer, and apply the status decisions described later before allowing the affected portions into their next activity.
7. Close and track onward movement. Record storage or shipment movements against the affected portions. Define completion as reconciled quantities, resolved identification, recorded status decisions and finished review. Until then, keep incomplete transfers visible with an owner, outstanding action and controlled availability.

Model genealogy as related records: source lot → split event → child allocations, with quality decisions and downstream movements linked to the affected portions. This is a practical data design, not a mandated database schema. Preserve manufacturer_batch_number separately from internal lot identifiers so receipt, subdivision and relocation do not overwrite the original manufacturing identity.

Use split_event_id to connect the source quantity and every resulting allocation. A full split allocates the entire source quantity; a partial split leaves an explicit residual quantity under the parent identity. Retain the parent record in both cases. Link later splits to their immediate sources so descendants remain discoverable across multiple generations.

child_lot_id can be optional when portions retain one lot number, but every tracked portion still needs an unambiguous reference. Keep traceability_lot_code as a separate food-specific field, rather than assuming it equals an internal child identifier. For location-level tracking, connect these references to bin-level lot genealogy.

ICH Q7 6.51, API GMP guidance, recommends unique batch or identification numbering for the records it addresses. That recommendation is not a rule requiring a new manufacturing batch number for every split container.

Test the model with three retrieval questions:

1. Which original material supplied this portion? Traverse parent relationships backwards to the original receipt and manufacturer batch identity.
2. Where did every portion from this split go? Retrieve allocations, recursive descendants and movement history, including remaining stock, consumption and shipments. This supports recall readiness.
3. Which evidence and decisions apply to this portion? Retrieve linked tests, deviations and scoped quality decisions through evidence_ref. Reference a controlled source document and its applicable version rather than creating conflicting copies for each child.

For a simple subdivision, use a conservation-based balance with a defined start and finish. This practice model assumes no material addition or transformation during the event. Express every term in one common unit, using mutually exclusive quantity categories:

Unexplained variance = source quantity before the event − child allocations − parent remainder − samples removed − documented loss.

Fix the measurement boundary before posting movements. If samples are already included in child allocations or another deducted category, do not subtract them again. Keep the source reading and the closing quantities tied to that same event boundary.

Accounted quantity is the sum of child allocations, parent remainder, samples removed and documented loss. Accounted percentage divides that total by the source quantity and multiplies by 100. Usable recovery instead divides usable child portions plus usable parent remainder by the source quantity, then multiplies by 100. A fully accounted event can therefore have reduced usable recovery. Neither percentage is a chemical reaction yield or manufacturing yield without an appropriate process definition and denominator; keep those measures distinct from theoretical versus actual yield.

For net-versus-gross reconciliation, distinguish material mass from container mass and record tare. Identify whether quantities were counted or weighed; any count-to-mass conversion needs a stated unit-mass basis. Volume-to-mass conversions need a density basis, including applicable temperature. Keep wet and dry quantities on a consistent basis, documenting moisture corrections where used. Preserve original readings, units, conversion factors, measurement resolution and rounding rules rather than retaining only converted totals.

A perfectly balanced calculation can still be wrong: systems may exchange identical numbers under different unit labels, or both may treat a tare-inclusive reading as net material. Arithmetic agreement does not establish measurement correctness.

The site's in-process reconciliation procedure should define investigation thresholds, responsibilities and handling of unexplained differences using process risk and measurement capability—not a universal acceptable-loss percentage. Correct measurement or transaction errors with a traceable explanation; document an actual spill or other material loss as a physical event. Do not relabel unexplained variance as loss merely to close the balance.

Use a practical label-and-record association for each portion: material identity, original lot reference, portion or container identifier, quantity and unit, location, and relevant status information. Keep the physical identifier readable and linked to the current record so operators can distinguish otherwise identical containers. These are recommended controls, not a universal label specification; the EU starting-material label guidance is addressed separately below.

Separate inherited provenance from child-specific history. The original material identity and lot reference establish origin; exposure during transfer, damaged packaging or additional processing describe what happened to that portion. Do not reset manufacture, expiry or retest dates merely because an identifier changed. Any justified date change should follow the applicable product procedure and retain its basis.

A parent test result is not automatically applicable to every later condition. Assess the sampling basis, material homogeneity, subsequent processing, storage and handling before relying on existing results for a portion. For APIs, ICH Q7 6.60 says laboratory raw data should be identified to show the specific material and batch tested. That identification anchors the evidence; it does not itself establish representativeness after later events.

Authorised people review the relevant evidence and make disposition or release decisions; creating a child record is not approval. A child-specific hold can restrict a portion with damaged packaging while its extent is assessed. A source-level concern, such as suspect parent test data, calls for assessment of potentially affected siblings rather than an isolated container restriction. Record the decision scope, rationale, reviewer and time, linking quarantine status to the applicable decision and retaining any release approval in the electronic release record.

Use the framework that applies to the product and activity, not the label “split batch” alone. ICH Q7 and EU GMP Chapter 5 are guidance sources in this comparison; 21 CFR 1.1320 is binding US regulation within its applicable scope. The table separates provision-specific expectations from their practical implications.

Record design is a separate implementation choice

As recommended industry practice, distinguish internal portion identifiers from the source batch number and any regulated food traceability lot code. A lot pedigree can retain those relationships without treating each identifier as legally interchangeable. This is a record-design recommendation, not a prescribed schema from these provisions.

For FSMA 204, check current Subpart S applicability, exemptions, the meaning of transformation and implementation dates. For supplements, devices and industrial chemicals, check the applicable product and jurisdiction rules rather than extending the API, medicinal-product or food provisions by analogy.

Case 1 — Supplement powder allocation

Assume 1,000.0 kg of net source powder, no added material and a common mass basis. The split transaction transfers 400.0 kg to child SUP-01-A and 350.0 kg to child SUP-01-B, debiting parent SUP-01 by 750.0 kg. A separate sampling transaction debits 2.0 kg; a documented handling-loss transaction debits 8.0 kg. Each references the same split event.

The reconciliation is 400.0 + 350.0 + 240.0 + 2.0 + 8.0 = 1,000.0 kg. Unexplained variance is 0.0 kg. Usable material is 400.0 + 350.0 + 240.0 = 990.0 kg; usable recovery is 990.0 ÷ 1,000.0 × 100 = 99.0%.

The parent inventory balance holds the remaining 240.0 kg: it has not moved and receives no additional debit. The reconciliation references the sampling transaction rather than posting its 2.0 kg again. This illustrates net-mass reconciliation, not an acceptable-loss limit or a manufacturing-yield target.

Case 2 — API drums with different handling histories

Assume six sealed drums from batch API-01, each containing 20.0 kg, with no loss. Container references D01–D04 move to location A: 4 × 20.0 = 80.0 kg. D05–D06 move to location B: 2 × 20.0 = 40.0 kg. Total inventory remains 120.0 kg.

A recorded handling excursion affects D05 and D06. Associate the excursion report, time interval, monitoring data and movement records with those container references and API-01. Place their 40.0 kg on a scoped hold, retaining the location and status history.

Quality reviewers then assess whether the cause also reaches A—for example, through shared transport before separation. Neither rejection of all six drums nor automatic clearance of D01–D04 follows from the location split. Record the assessed scope, supporting evidence and human disposition decision.

Case 3 — Food shipment allocation

Assume 100 already packed, sealed cases of covered food, each containing 10.0 kg. This is shipping only; no other event or applicable exception changes the code decision. Assume the food-code decision covered above establishes that existing traceability_lot_code FOOD-01 carries through both shipments.

Allocation AL-01 links 30 cases to shipment SH-01 for customer A: 300.0 kg. AL-02 links 70 cases to SH-02 for customer B: 700.0 kg. Both allocations record FOOD-01; 300.0 + 700.0 = 1,000.0 kg.

For recall readiness, connect the original lot and code-source reference to allocation lines, picked case or pallet references, quantities and units, shipping dates, dispatch location and each receiving destination. Keep shipment and allocation identifiers separate from the food lot-code field.

Assign one authoritative record for each data domain. For example, ERP may own material identity and inventory balances; warehouse software owns container locations and movements; production owns split execution; laboratory software owns test evidence; and a quality-controlled record owns approved status. Label-printing software consumes authorised data rather than deciding disposition. Document how warehouse movements update inventory without creating competing balances. These are implementation choices, not capabilities implied for any named product.

Pass stable split-event, parent and portion identifiers, alongside units and record versions. Map these to shared material master data; do not match transactions only by label text or timestamps. Record message state in a process event log.

• Handle duplicate messages idempotently: replaying one event must not deduct inventory twice. Track partial acknowledgements by destination and route unresolved failures to an exception queue.
• For offline execution, retain identifiers and pending transactions for reconciliation before downstream use. After interrupted printing, establish which labels physically printed before authorising replacements.
• Reject stale status updates using version checks; block dependent movements when current approval status cannot be established.

For a wrong-parent or wrong-unit event, preserve the original transaction, record an authorised correction or reversal, reconcile quantities across affected systems, and resolve affected labels before resuming use. Test recovery paths under the site's validated operating procedures.

Use these practical checks to test split records and transactions. Apply the quantity and disposition principles covered above rather than treating reconciliation alone as proof of control.

Two investigation traps need separate tests. An incomplete forward trace may find immediate children but miss later descendants. Test the actual genealogy query against a known multigeneration split, including shipped portions, to challenge recall readiness.

A child-specific incident may also be expanded to every sibling—or narrowed to one container—without checking cause. Test that decision scope against handling history and causal evidence. Neither a visually plausible label nor a balanced total establishes which material is affected.

Use these practical checks in a site acceptance test; record pass/fail results and supporting record IDs.

• Lot and container identifiers distinguish material lineage from physical container identity.
• A partial split leaves the remainder identifiable, located and available at the correct quantity.
• Source, portions, remainder and recorded losses reconcile in a documented unit.
• Evidence references open the intended records.
• Each disposition identifies an accountable human reviewer and decision.
• Each physical portion matches its label and recorded identity.
• A source query retrieves all downstream destinations.
• Corrections retain original values, reasons, authors and timestamps in the process event log.

Challenge exercise: select one real portion and trace backwards; select one source and trace forwards. In a test environment, simulate a duplicate message and a wrong-parent correction. Confirm no duplicate depletion and a recoverable correction history. Assign an owner for every unresolved result before routine use.