Batch ID Generation

A batch ID (often called lot number in regulatory text) is the unique identifier assigned to one production run at the moment a control recipe is created from its master. It is what 21 CFR 211.188 means when it requires the batch production record to be 'uniquely identified'. The ID propagates to the eBR, to every sample and IPC, to packaged units (via the GS1-128 lot field), and into the genealogy graph for forward and backward traceability.

• Generated automatically at WO release or at first control-recipe instantiation.
• Issued atomically — no two batches receive the same ID even under concurrent load.
• Immutable once issued — corrections never overwrite, they reference a new ID.
• Embeds the metadata needed by humans to read it at a glance (site, line, year, sequence).
• Used as the foreign key in every system the batch touches: MES, ERP, LIMS, WMS, QMS.

Schemes vary by industry and regulator preference. Common patterns:

Pick one scheme per site and resist exceptions. Mixed schemes within a site break sort orders, search assumptions and downstream parsers.

• ID issuance is a transactional operation in the system of record — never derived from a clock or a parsed display string.
• Sequence counters are namespaced by site/line so two sites cannot collide.
• Reservation precedes use — if a WO is cancelled before any record is created, the ID stays reserved and is not recycled.
• Concurrent WO releases pull from a serialised sequence; no two batches share an ID even microseconds apart.
• External system reconciliation uses the canonical ID, not formatted display variants.

A batch ID, once issued, is permanent — even if the batch is aborted, lost, rejected, or never started. Corrections to batch records reference the ID; they do not reassign it. Common discipline:

• Aborted batches keep their ID with status Aborted; the ID is not reused.
• Batches that fail QA disposition retain their ID with status Rejected.
• Reprocessed batches receive a new ID linked to the original via parent–child genealogy.
• Test or 'engineering' batches use a clearly distinguishable ID prefix or namespace.
• ID list with status is a queryable inventory — auditors ask 'show every B26-04500 to B26-04600' and expect a complete answer.

• Genealogy — backward (what materials went in?) and forward (what products contain this batch?) traceability rests on the batch ID.
• Labels and packaging — GS1-128 AI(10) lot field carries the batch ID to distribution; FDA recall calls reference this.
• Sample IDs — derived from batch ID + sample point + sequence; CoA links back via batch ID.
• Complaint and CAPA linkage — complaints reference batch IDs, enabling pattern analysis.
• Recall execution — recall scope expressed as 'all units of batches B26-04500 through B26-04520'.

• Pharma — Year+sequence, with prefix flagging product family; aligned with FDA submission and EU MAH lot conventions.
• Biopharma — Lot ID + sub-lot ID (per-vial fill) traceability into the patient-administered unit.
• Medical device — Lot or Serial number per UDI rules; FDA GUDID requires this on the label and in submissions.
• Food — Julian date + line tradition retained for compatibility with retailer expectations.
• Cosmetics — MoCRA-aligned lot codes carrying the manufacturing-date encoding for cosmetic recall events.
• Chemicals — REACH-compliant batch numbering carried to SDS, eSDS and labels.

• Manual typing of batch IDs into downstream systems — drift inevitable; reconciliation manual; complaint linkage broken.
• ID recycling after long gaps — collisions decades later when archives are queried.
• Display formatting (with hyphens, spaces) used as the canonical ID — comparisons fail across systems.
• Two systems independently generating IDs that 'should match' — they diverge under load.
• Engineering or test batches indistinguishable from production — accidental inclusion in regulatory queries.
• ID embedded with information that changes (e.g. customer code) — outdated after customer is renamed.
• No status field on IDs — aborted vs released vs in-progress not queryable.