Traceability systems: capturing genealogy where material changes identity

Capture points decide which questions you can answer

Genealogy is built at transformation events: material issued to an order, a component consumed into a parent, a lot decanted into a smaller container, a unit passing an operation that changes what it is. Any event you do not capture becomes an inference later, and inferences widen recalls. Mapping those points before choosing a system is the work that determines everything downstream — how many scan stations exist, where the network must reach, which operations gain an extra step. Plants that buy first and map afterwards find the chain has a hole in exactly the process where scanning was inconvenient.

Choosing a granularity the process can sustain

Lot, batch and serial are not a ranking. They describe how much precision the physical process can carry. Mixing and continuous processes destroy unit identity by their nature, so lot-level genealogy is honest and serial-level is fiction with a database behind it. Discrete assembly can carry serial identity but pays in scan time and in exception handling whenever a unit is dismantled. The real test is whether the granularity survives a bad night: jammed printer, partially consumed pallet, short-staffed shift. Precision you cannot hold under pressure leaves gaps that are worse than a coarser scheme applied without exception.

Splits, merges and rework break the simple model

The tidy parent-and-child picture holds until material moves sideways. A lot is split across containers and the remainder used much later. Rejected units are stripped for good components that re-enter stock. A batch is reprocessed and now carries two production records. Surplus returned from the line goes into a bin already holding a different lot. Each needs an explicitly modelled event, and candidate systems differ sharply in whether they provide one. Where the model has no event, operators improvise — usually by writing material off and re-receiving it, which severs the link precisely where an investigator would want it whole.

The recall query is the acceptance test

Before signing anything, write out the questions you will one day answer under pressure, and make each candidate answer them using your data. Backwards: for this shipped unit, every input lot, operator, machine and inspection result. Forwards: for this suspect raw lot, every finished unit and every customer who received one. Then look at how wide the answer is, because a system that returns a month of production has technically responded and practically failed. Load sample data and run it during evaluation instead of accepting a spoken description of the capability.

The record fails at the terminal, not in the database

Nearly every genealogy gap traces to a human decision at a screen. A damaged label will not read, the queue is building, and the operator picks the default lot from a list. A scanner is flat, so the shift is booked afterwards from memory. A supervisor overrides a block to keep the line moving. Hard stops work only where exceptions have a fast, supported path; otherwise they are simply routed around. Log every override with a reason, review who is using them, and read a rising override rate as a process defect rather than an individual discipline failure.

Frequently asked questions

Our planning system already stores lot numbers. Is that traceability?

Storing lot numbers and maintaining genealogy are different capabilities. A record that a lot was received and later issued to an order tells you material went somewhere. It does not tell you which finished unit contains it once orders are split, reworked or partially consumed. The practical test is whether you can move in both directions across every transformation without opening a spreadsheet. If any link depends on somebody remembering, you have storage rather than traceability.

What do we do about material that arrives with no lot identity?

Assign your own identity at receipt and record whatever supplier reference exists — delivery note, container marking, heat number if it is stamped. This is normal for fasteners, consumables and commodity material, and it keeps the internal chain complete even though it terminates at your gate. State the limitation in the traceability plan so nobody assumes coverage that does not exist, and reserve supplier lot discipline for materials whose failure would genuinely trigger a recall.

How long should genealogy data be retained?

The driver is product life plus any contractual or sector retention obligation, not what the database can comfortably hold. Decide it explicitly and design the archive at the same time, because the practical failure is rarely deletion. It is data that still exists in a system nobody can query after a migration. Keeping an exported, readable genealogy extract beside the live system costs far less than keeping an obsolete application running purely to satisfy a retention rule.