UDI Traceability in Production: A Manufacturer's Guide

UDI traceability in production is the work of connecting the identifier intended for a device with the label actually applied, the components actually consumed, the packaging relationships actually created and the distribution events supported by records. A reference identifier alone cannot answer those questions. The useful manufacturing record preserves both the connections and the places where evidence is incomplete.

This guide covers US operations for devices subject to the selected provisions of 21 CFR Part 801 Subpart B and Part 830 in the reviewed October 21, 2026 section texts. It is not a review of the complete Parts. Selected QMSR record and labeling provisions are included where directly relevant. Regulatory statements, FDA website explanations and proposed operating practices are distinguished below. Issuing-system syntax, check digits and technical standards are outside this guide.

For QA/RA, labeling and manufacturing engineering teams, the following buyer questions provide a practical reading map. The section names identify where to examine evidence, not promises that a particular system supplies it.

Start with applicability, not the printer. 21 CFR 801.20 establishes the general UDI requirement for device labels and every device package, with exceptions. Section 801.30 contains important exceptions: for example, the specified class I GMP-exempt devices are excepted, while class I UDIs are not required to include a PI. Class I is therefore not shorthand for one universal labeling outcome.

Under 21 CFR 830.3, the DI is the mandatory fixed portion identifying the labeler and specific version or model. The PI is conditional: its possible elements are lot or batch, serial number, expiration date, manufacturing date and, for an HCT/P regulated as a device, the distinct identification code. These elements depend on what appears on the label; they are not a universal instruction to serialize every unit.

Subject to applicable exceptions, 801.40 requires easily readable plain text and AIDC presentation, and requires the PI segment to convey the listed production information appearing on the label. 830.20(a) requires issuance under a system operated by FDA or an FDA-accredited issuing agency. The examples below illustrate record relationships, not compliant identifier construction.

One Fictional Packet Throughout This Guide

For an operating baseline, retain the device and package descriptions, assigned identifiers, PI field selection, approved label content, source revisions and the people responsible for resolving differences. Preserve the baseline used for each event rather than reconstructing yesterday's label from today's master data. This is recommended evidence design, not a requirement to place every source in one application.

In this packet, internal unit keys allow unit-level reconciliation even though the label carries lot-level PI. That distinction matters: a lookup using the printed lot identifies a population, whereas an internal unit key identifies one manufacturing record. Do not silently represent the internal key as an externally readable serial number.

The labeler must provide device-identification information for each version or model required to bear a UDI under 830.300(a). However, the reference record is not a production-event ledger. 830.310(b)(10) calls for the type of PI appearing on the label. FDA's GUDID webpage explains that GUDID contains the DI portion and PI flags, not actual PI values.

For the fictional baseline, the reference-data concept is therefore a lot-number presence flag, not a database entry containing TRN-24A. Actual lot values, application events, component consumption and shipment evidence belong in the manufacturer's operational reconstruction. A matching catalog entry cannot supply those missing events.

These timing and change rules come from 830.330 and 830.50. Relabeling has a separate rule discussed below. For the packet, the reference-data owner must explain the original and successor DI relationship and the applicable update assessment; an internal task marked complete is not evidence of the external outcome.

830.360 requires records of UDIs used and the version or model associated with each DI, retained for three years after marketing that version or model ceases and submitted to FDA upon specific request. That period is not a blanket retention period for all manufacturing evidence: the section expressly preserves other FDA recordkeeping obligations.

A print job documents an instruction or output, not necessarily what reached a particular device. For each chosen control point, connect the expected identifier to the observed label and its disposition. Useful evidence includes the baseline revision, unit or batch reference, label-instance record, application time, inspection result, performer and any exception. An unrecorded result should remain missing rather than being inferred from successful printing.

The selected QMSR provisions establish relevant duties without prescribing this guide's event model. 820.35(c) and 820.45 require recording UDI for each device or batch, documented labeling and packaging procedures, examination for accuracy before release or storage where applicable, documented release of labeling for use, and controls against mixups with documented inspection results.

Separate Reading, Matching And Acceptance

• Reading: what information was obtained from the actual label? Keep unreadable and unsuccessful attempts visible.
• Matching: did the observed DI, applicable PI and package level agree with the intended record and label revision? A readable wrong label still fails this comparison.
• Acceptance: did the inspection meet the operation's approved criteria, and who reviewed the result or exception? Do not treat data capture as disposition.

Scan locations, inspection sampling, serial allocation methods and duplicate-detection design are process choices to justify for the operation, not universal requirements established by these selected UDI sections. Ask what failures a proposed method can detect and what it leaves unexamined. For example, a sample result should identify its sample scope rather than becoming an undocumented claim that every label was checked.

For direct marking, 801.45 permits plain text, AIDC or both; do not import the ordinary label's two-form requirement unchanged. The direct-mark UDI may be identical to the label UDI or different to distinguish the unpackaged device. Exceptions cover interference with safety or effectiveness, technological infeasibility, specified additional processing of a single-use device, and a device previously marked under the section. A labeler using an exception must document its basis in the required design and development files. Size alone is not a separately listed exception.

The software row is limited to the display provision in 801.50(b); it does not establish requirements for every software component. Neither special presentation situation applies to our fictional single-use physical device. Their purpose here is to prevent a buyer from accepting one universal label-verification workflow for unlike products.

Build genealogy from actual consumption, not the planned bill of material. In the fictional packet, each device consumes one primary component. Inventory records show 90 components issued from C-TRAINING-A and 40 from C-TRAINING-B; ten unused A components were returned to quarantine. Thus 130 issued equals 120 net consumed plus ten returned. That arithmetic does not fill missing unit-level consumption records.

The eight unknown links remain a separate category. Do not relabel them confirmed A merely to make a genealogy report look complete. Equally, do not exclude them from an investigation of A. Record both the inventory-level inference and its evidentiary limitation, then seek the missing station records or other justified evidence.

Packaging adds another relationship. 830.3 defines a device package as a fixed quantity of a particular version or model, while a shipping container has contents that may vary between shipments. 801.30(c) does not require UDI on a shipping container. Calling something a carton does not resolve which definition fits.

P-TRAINING-01 contains units 001–010, P-TRAINING-02 contains 011–020, and the same sequence continues through P-TRAINING-10 containing 091–100. All ten package-membership records are present. Units 101–116 remain individually packaged outside these ten-device packages; units 117–120 are scrapped. These are mutually exclusive physical groupings at the cutoff.

If a package is opened later, preserve its former membership and record the removal and new membership events. Do not overwrite an earlier shipped configuration with today's contents. In this packet no ten-device package contains a relabeled unit, so the successor individual-package DI is not silently substituted into an existing package relationship.

The investigation begins with two fictional signals: a supplier query concerning C-TRAINING-A and discovery of obsolete R0 labels on units 101–112. Use separate branches for the material question and the labeling defect while retaining one reconciled population. A replacement label is an event involving an existing device, not another manufactured device.

830.60 requires a new DI when a device required to bear UDI is relabeled, plus a record relating the prior and new DIs. Do not declare every printer retry to be relabeling, or exempt every correction by calling it a reprint. Have the responsible reviewer classify the actual activity and record the basis.

Reconcile Label Instances Separately From Devices

The unit-label ledger contains 120 original instances plus 16 replacements: 12 for relabeling and four for pre-application spoil. Total unit-label output is therefore 136. At cutoff, 116 instances are last-recorded current labels on nonscrapped devices. The other 20 are inactive: 12 removed and voided, four spoiled before application and four destroyed with scrapped devices. Thus 136 equals 116 plus 20. The ten outer-package labels have a separate ledger and are not included in this calculation.

Current does not mean accepted, physically located or releasable. Two current replacement labels lack verification records, and ten devices have unresolved location. A report that counts only active label instances would conceal both problems unless it also presents inspection and location states.

For unit U-TRAINING-111, label instance L-TRAINING-111-A carried DI-TRAINING-0001 and is recorded removed and voided. Replacement L-TRAINING-111-B carries DI-TRAINING-0002 and the unchanged lot. Application is recorded, but its verification result is missing. The old identifier remains searchable through the replacement relationship; it is not deleted from history. This unit remains on HOLD.

For U-TRAINING-113, L-TRAINING-113-A was spoiled before application and L-TRAINING-113-B was applied and inspected. Both refer to DI-TRAINING-0001. That example is deliberately narrower than permission to replace any applied label without a new-DI assessment. Rework routing, identifier classification and acceptance remain separate decisions.

Reconstruct distribution from the population outward, using departure evidence rather than the sales order alone. The packet has planned allocations of 70 devices to Distributor-TRAINING-A, 20 to Distributor-TRAINING-B and ten to Distributor-TRAINING-C. Those 100 allocated devices are not 100 proven shipments.

This table reconciles the whole packet without counting the 12 relabeled devices twice. Its unresolved category is neither a confirmed shipment nor verified on-site inventory. The missing carrier record is a genuine gap, not a reason to force those ten units into whichever category makes the report appear finished.

Continue Beyond The First Recipient Without Inventing Reach

The 80 confirmed departures have a second, separate distribution breakdown. Distributor A reports 40 devices still in its warehouse and 20 forwarded to Clinic-TRAINING-X. Distributor B has not supplied a downstream inventory response for its 20. Thus 40 plus 20 plus 20 equals the 80 confirmed departures. These are subcategories of shipped devices, not additional units added to the 120-device denominator.

The reported 40 are not yet confirmed segregated. The forwarded 20 lack clinic receipt and use records. The remaining 20 lack downstream status entirely. No patient-level exposure conclusion is supported. Even a customer acknowledgment would answer only the question documented in that acknowledgment, not automatically establish clinical use, non-use or successful containment.

The responsible reviewer should retain the source and timestamp of each statement: carrier evidence, distributor report, warehouse count or unresolved posting. If a later response changes a location, update the time-specific state while preserving the earlier uncertainty. Do not convert a current location into an invented history of where the device was during the interval.

Forward Query From A Component Lot

Start with C-TRAINING-A and follow the 72 documented consumption links. Units 001–060 went to Distributor A; units 061–072 went to Distributor B. Then add the eight unresolved consumption records, units 073–080, to the conservative review population rather than treating them as confirmed A consumption. All eight have confirmed departure to B. The packet therefore identifies 72 documented A-linked devices and eight additional devices requiring investigation.

The inventory calculation—90 A components issued minus ten returned equals 80 net—supports investigating those eight records, but it is not a substitute for them. Keep the review population of 80 distinct from a finding that all 80 definitely consumed A. Also assess whether the supplier query or record weakness extends beyond this bounded packet; the example's denominator does not establish the universe of potentially affected production.

Reverse Query From A Replacement Label

Start with the successor DI on U-TRAINING-111. The replacement record leads to its original label instance and DI, the unchanged lot, recorded C-TRAINING-B consumption and the rework location. It also exposes the missing verification result. Searching only the current DI would miss the original label history; searching only the original DI would miss the replacement unless the relationship is traversed.

Neither query proves that B-linked units are unaffected. They establish recorded associations and gaps, not a complete hazard assessment. Nor does the ten-device package DI identify a particular package instance: P-TRAINING-01 and P-TRAINING-10 share that configuration DI but have different contents and distribution histories. Preserve the distinction between configuration identity and a physical instance.

For a buyer demonstration, require both queries to return original evidence references, unknown relationships and the relevant cutoff time. A clean list without the eight missing consumption records, two missing label-verification records and unresolved ten-device departure is an incomplete reconstruction, even if every displayed row is internally consistent.

The fictional QA decision owner records a temporary lot-level HOLD because the label investigation's scope is not closed, two replacement-verification records are missing and the material-query review remains open. This is a conservative case decision, not a universal rule that any traceability gap mandates an identical hold. A later narrowing decision would need its own evidence and rationale.

1. Manufacturing record owner: investigate the eight missing consumption links. Preserve original evidence and identify any reconstructed entry as such; do not fabricate contemporaneous records.
2. Labeling owner: investigate verification for units 111 and 112 and confirm the scope of obsolete R0 use. Any new examination should be identified as new work, not backdated as the missing inspection.
3. Warehouse and distribution owners: locate P-TRAINING-09, reconcile the dispatch posting with carrier evidence and obtain physical confirmation or document continuing uncertainty.
4. Distribution owner: request segregation confirmation for A's reported 40, receipt and subsequent-status evidence for the forwarded 20, and downstream status for B's 20.
5. Regulatory and reference-data owner: document the relabeling DI relationship and applicable submission/change assessment without treating a catalog record as product disposition.
6. QA decision owner: review returned evidence and the separate risk assessment, determine further containment or field-action evaluation, and record the next decision and its basis.

For this exercise, all owners report progress by 16:00 on October 23, with unresolved items escalated then rather than silently closed. That checkpoint is an internal scenario choice, not a regulatory reporting deadline. The decision record identifies the population, evidence reviewed, uncertainties, action owners and next review time. It grants no release authority and makes no claim that recipients or patients have all been reached.

Evaluate a proposed labeling or manufacturing-record workflow using this packet, not a feature checklist alone. Ask the evaluator to preserve the same denominator, historical relationships and missing records through each demonstration. The tests below are practical acceptance questions; they do not assert capabilities of any product, plan or application.

Ask who owns the DI baseline, label revision, PI source, inspection result, package membership, distribution evidence and reference-data change. Where evidence crosses a system boundary, require an explanation of how a missing or misbound result becomes visible and who investigates it. This is an ownership test, not a claim that one application must perform every task.

This guide does not duplicate rollout planning or record-format comparisons. Use the eDHR implementation guide for that implementation intent and DHR vs eDHR: What FDA Requires for the separate format question. Here, the evaluation stays focused on reconstructing what happened to identifiable devices and documenting what remains unknown.