Systems and records
Food labeling correction dataset: tracing a label revision to lots in transit
By SourceX Editorial · Reviewed by Noah Loul ·
9 sections
Short answer
Start from the revision, not the warehouse. Record what changed on the label and why, the first lot printed with the new version, and every lot made under the old one. Then follow each old-version lot through holds, shipments and receipts to its current location, assign the action the quality team approved for that location, and keep the traceability records that link any relabeled stock back to its source lots.
System export
Food labeling correction
Key takeaways
- A label revision has an effective lot; every earlier lot carries the old label wherever it now sits.
- For foods on FDA's traceability list, relabeling is transformation, so a new lot code and links to input lots are required.
- The reason for the revision, from undeclared allergen to marketing refresh, decides the action on shipped stock.
- A replayable task reconciles quantities, times, links and identifiers, and records unrecovered units as a number with a reason.
- Recall, withdrawal and sell-through decisions stay with food safety and regulatory staff; the dataset teaches the trace, not the call.
How do you trace the effects of a label revision on inventory already in transit?
Treat a label revision as an event with a reach. The revision has an effective point (a production date, a lot or a packaging run), and every lot made before that point carries the old label wherever it now sits: in finished goods, on a truck, at a distribution center or on a customer's shelf. Tracing the impact means listing those lots, locating each one at the moment of the decision, and recording what the business decided to do with it.
A food labeling correction dataset captures that reasoning as tasks. Each task gives a model the revision record, the lot and shipment history, and the quality decisions that applied, and asks it to identify the affected lots, their state and the approved action. The value is not in the label text but in the links: revision to lot, lot to shipment, shipment to receiver, decision to outcome.
This page describes the task specification, a worked example and the limits of the method. It does not tell a business whether to recall a product; that is a regulatory decision made by qualified people, often with the FDA.
Why does a label change become a traceability event?
For many foods, fixing a label is not only an artwork change; it changes the traceability record. Under FDA's food traceability rule in 21 CFR Part 1, Subpart S, transformation includes changing a food by relabeling, and a traceability lot code is assigned when a food is transformed. When a covered food is relabeled, the business keeps transformation records that list the new lot code and, for each input lot, the quantity used.
The rule applies to foods on FDA's Food Traceability List. FDA has said Congress directed it not to enforce the rule before July 20, 2028, and that it intends to follow that directive. Even for foods outside the list, the same structure (old lot in, new lot out, quantity linked) is what makes a correction auditable.
Product identifiers can change too. GS1's GTIN management rules say any increase or decrease in the legally required declared net content printed on the pack requires a new GTIN, so a net content correction can split one item into two identifiers in the order, shipping and receiving records.
Which kind of revision decides what happens to lots already shipped?
The reason for the revision determines the action on stock already in motion. A missing allergen declaration and a corrected marketing claim may look alike in the artwork system, but they lead to different decisions and different records.
FDA's recall regulation, 21 CFR 7.3, defines a correction as repair, modification, adjustment, relabeling, destruction or inspection without physically removing the product, separates a market withdrawal (a minor violation or none) from a recall, and ranks recalls in three classes by health hazard. FDA says labeling errors cause most of its food allergen recalls. The table shows typical paths; the classification of a real case is made by the business's food safety and regulatory staff and, for recalls, by FDA.
| Revision type | Example | Typical path for shipped lots | Record consequence |
|---|---|---|---|
| Undeclared major allergen | Sesame added to a recipe but missing from the label | Hold and retrieve or correct at every location; regulatory review | Lot list, holds, notices, relabel or destruction records |
| Declared net content | Pack weight corrected on the label | Correct or retrieve depending on direction and amount; new GTIN | Old and new GTIN mapping, relabel transformation records |
| Required statement error | Wrong name or address of the responsible firm | Often corrected in place or at the next run; regulatory review | Correction records by location |
| Voluntary or marketing change | New claim, artwork or design refresh | Sell through existing stock; new label from a set lot | Effective lot only; no action on shipped stock |
| Ingredient or supplier change | Substitute ingredient approved for a run | Label must match the formula used for each lot | Formula version linked to each lot |
Which fields does the revision and affected lot task specification need?
Each task in the specification is one revision applied to one product's lots, with the decision trail that followed. Version the specification (for example v1.0, dated 2026-10-18) and stamp each task with it.
The fields below are what a reviewer needs to replay the task without access to the plant's systems. Use internal codes for customers and carriers; the task needs locations and states, not names.
Keep the decision log as it was written, including decisions that were later reversed. A hold released after testing, or a retrieval canceled once a customer confirmed the stock was never opened, is part of the trace. Tasks built only from final states hide the reasoning that a quality team actually went through, and a model trained on them cannot explain why one lot was released while its neighbor was destroyed.
| Field | Content | Source record | Check before use |
|---|---|---|---|
| revision | Label version ids before and after, reason code, changed elements, approval date | Label or artwork management and change control | Reason code matches the change; approval precedes the effective lot |
| effective_point | First lot, run or production date printed with the new version | Production and packaging records | No lot falls on both sides |
| affected_lots | Every lot made under the old version, with quantity and unit | Production lot records | Totals reconcile to production |
| lot_states | For each lot at decision time: on hand, on hold, in transit, received by customer, consumed | Inventory, shipping and receiving records | Timestamped no later than the decision |
| identifiers | Item GTIN, traceability lot code and any new codes after relabel | Item master and traceability records | Old to new mapping is complete |
| quality_decisions | Action per lot or location: release, hold, correct in place, retrieve, destroy, sell through | Quality and regulatory decision log | Each decision names the approving role and time |
| notifications | Who was notified, when, and acknowledgment state, coded | Customer notice and recall communication logs | Coded receivers resolve to shipment records |
| outcome | Quantity corrected, retrieved, destroyed or unaccounted for, per lot | Returns, relabel and disposal records | Outcome quantities reconcile to affected quantities |
How do you replay each affected lot from the plant to its last known location?
Match every state change in the task to an event in the records. Replay means walking each affected lot from production to its last known location and confirming that the decision recorded for it was possible given its state at that time.
At a glance
How do you replay each affected lot from the plant to its last known location
GS1's Core Business Vocabulary, published with the EPCIS standard, gives a shared set of terms that make replay easier across companies. Its business steps include shipping, receiving, holding, inspecting and repackaging, and its dispositions include in_transit, recalled, non_conformant, non_sellable_other and sellable_not_accessible. Mapping each internal status to one of these terms lets a reviewer read a lot's history the same way regardless of which warehouse system produced it.
- Quantity check: the old-version quantity produced equals the sum of quantities on hand, in transit, received, corrected, retrieved, destroyed and consumed.
- Time check: no lot is marked corrected or retrieved before the decision that ordered it.
- Link check: every relabeled output lot points back to its input lots with quantities, as the transformation record requires.
- Identifier check: every shipment after a declared net content change uses the GTIN that matches the label on the stock.
- Gap check: unaccounted quantity is recorded as a number with a reason, not dropped.
How is an illustrative missing sesame declaration traced across lots in three places?
This example is illustrative; the product, lot codes and quantities are invented. A bakery adds a sesame topping to a bread line and updates the recipe, but the first production runs ship with the previous label, which does not declare sesame. Sesame has been a major food allergen under the FASTER Act since January 1, 2023, so the omission is an undeclared allergen.
Three lots carry the old label. Lot A is on hold in the plant's finished goods area. Lot B left the plant the previous evening and is in transit to a distribution center. Lot C was received by two retail customers the day before. The task gives the revision record (reason code: allergen omitted), the effective point (first lot with the new label is Lot D), the lot states at the decision time and the decision log.
The correct answer lists Lots A, B and C as affected and Lot D as unaffected, assigns the actions the quality team recorded (destroy A, intercept B on receipt and hold, retrieve C from both customers), and links the notifications sent. The outcome field shows the retrieved quantity from one customer was smaller than the quantity shipped, with the gap recorded and explained as sold before the notice. That gap is the part a model most needs to learn: a trace is complete when every unit is accounted for, including units that can no longer be recovered.
Can the same revision call for two different answers?
The same label change can call for different actions depending on timing. When FDA added sesame as the ninth major allergen, it said products already in interstate commerce or on store shelves before January 1, 2023 did not have to be removed or relabeled. Stock in motion at the time of a regulatory change can therefore be compliant under one rule while new production is not.
Include tasks like this in the dataset, labeled with the condition that changes the answer: the effective date of the requirement, the date each lot entered commerce, and the guidance the business relied on. Without them, a model learns that every old-label lot must be pulled, which is wrong and costly.
The limit of the method is the decision itself. The dataset teaches a model to assemble the facts and replay recorded decisions; it should not be used to make the recall or withdrawal call.
Which decisions must a food safety or regulatory lead make?
Stop and hand the question to a person when the answer depends on hazard, law or a regulator rather than on the records. A model can gather the lot list in minutes; it cannot weigh the risk to a consumer with an allergy.
- Whether an issue is a recall, a market withdrawal or a correction, and which recall class applies.
- Whether stock in commerce may sell through after a requirement or formula change.
- Whether relabeling at a distribution center or customer site is acceptable, and who may perform it.
- How to treat quantities that cannot be accounted for after notices go out.
- Any case involving infant formula, medical foods or products regulated by USDA rather than FDA, which follow their own rules.
- For dataset builders: narrow to one product family and one revision type if decision logs are incomplete for others, and have a food safety reviewer confirm every quality decision label.
Can label revision and lot histories be licensed?
Label change records, lot genealogy and decision logs can be useful for training traceability and quality assistants, but they often mix the company's records with co-manufacturer, customer and supplier data. Access to the systems, privacy permissions, ownership of the records and contract terms with brand owners or co-packers are separate questions, and de-identification alone does not give a right to license.
A food manufacturer that sorts out those questions can license revision, lot and decision records through SourceX, which handles the rights review, contract, data preparation and delivery. The manufacturer first takes a short company assessment that needs no lot data uploaded, and no record is delivered before it has approved the scope and terms and signed.
Frequently asked questions
What is a food labeling correction dataset?
It is a set of tasks built from real label revisions and the lot, shipment and quality records that followed them. Each task asks which lots were affected, where they were, what action was approved for each and what the verified outcome was. It teaches traceability reasoning, not label design.
Does relabeling a food create a new traceability lot code?
For foods on FDA's Food Traceability List, relabeling is a form of transformation under 21 CFR Part 1, Subpart S, and a traceability lot code is assigned at transformation. The transformation record links the new lot to the input lots and quantities used.
Is a label correction always a recall?
No. FDA's regulation distinguishes corrections, market withdrawals and recalls, and classifies recalls by health hazard. Which one applies depends on the violation and its risk, and the call is made by the business's regulatory staff and, for recalls, FDA.
Do I need a new GTIN when a label changes?
Not for every change. GS1's rules require a new GTIN when the legally required declared net content printed on the pack changes, up or down. Other changes follow other GS1 rules, and local regulations can require more frequent changes.
What records make a label revision traceable?
The revision approval with its reason and effective lot, production lot records, inventory and hold states, shipping and receiving events, the quality decision log, notification records and the final disposition of every unit.
Sources
- FDA's traceability rule defines transformation to include relabeling and assigns a traceability lot code when a food is transformed. Source
- Transformation records under the traceability rule list the new lot code and, for each input lot, the quantity used. Source
- FDA's food traceability rule covers foods on the Food Traceability List and requires key data elements for critical tracking events such as shipping and receiving. Source
- FDA says Congress directed it not to enforce the food traceability rule before July 20, 2028, and that it intends to comply. Source
- FDA's recall regulation defines correction (including relabeling), market withdrawal and recall, and classifies recalls into three classes by health hazard. Source
- FDA says labeling errors cause most food allergen recalls and lists the nine major food allergens. Source
- The FASTER Act made sesame the ninth major food allergen from January 1, 2023, and products already in commerce before then did not have to be removed or relabeled. Source
- GS1's GTIN management rules require a new GTIN for any change to the legally required declared net content printed on the pack. Source
- GS1's Core Business Vocabulary business steps include shipping, receiving, holding, inspecting and repackaging. Source
- GS1's Core Business Vocabulary dispositions include in_transit, recalled, non_conformant, non_sellable_other and sellable_not_accessible. Source
See if your company qualifies
A short company assessment. No data uploads are needed.