At the start of a run, one part is taken off the line and not packed. A tag is tied to it with the batch, the machine, the colour code and the time, and it goes into a drawer or onto a shelf near the line. It is not a souvenir and it is not a reject. For the rest of that run it is the measuring instrument that cannot be switched off.
A workshop has plenty of instruments for the things that can be counted. A caliper will settle a dimension, a scale will settle a weight, a gauge will settle whether a hole is in the right place. But a good deal of what makes a part acceptable has no number attached: the sheen of a surface, the exact shade of a moulded colour, how two halves meet when they are pressed together, the feel of an edge under a thumb. Those are judged by comparison, which needs something to compare against.
What a number cannot settle
The retained part is that something. An inspector picks up a current piece, puts it beside the retained one under the same light, and looks. If the two read as the same object, the run continues; if they do not, the inspector reaches for the caliper to find out what moved. The comparison comes first and measurement second, which is much faster than measuring everything on every cycle.
Descriptions drift between shifts. One operator's note that a part came out slightly pale can be read three different ways by the next person, and by the third shift nobody remembers what the original looked like. An object does not drift. It also settles disagreements without rank: the question is no longer whose judgement wins but whether these two pieces match, which anyone at the bench can answer.
The tag matters as much as the part. Batch, machine, colour code and time tie that object to one specific set of conditions. If a customer later asks about a carton from Tuesday morning, there is a physical answer to what Tuesday morning was supposed to look like. Judgement has been moved out of people's memory and into a labelled thing that can be handed over.
The four moments that call for it
The sample comes out of the drawer at changeovers: a new shift, a new lot of material, a tool that has been swapped, or a machine restarted after a stop. Each is a moment when something invisible may have moved. Bringing the current part alongside the retained one takes a few seconds and covers all of them at once, including differences nobody thought to write a check for.
It also defines an endpoint for cleaning. When a moulding shop changes from a dark colour to a light one, the barrel has to be purged until the discharge is clean, and the practical test of clean is a part that matches the sample. Why changing a colour costs half a day is largely the cost of reaching that match without being able to calculate it in advance.
A ceramic town that sorts by sound uses the same trick with a different sense. A tap and a listen replace a laboratory measurement because the decision has to be made a few steps from the kiln, hundreds of times a day. In both cases the workshop has compressed a complicated judgement into one action that can be repeated identically by different people.
Why the sample itself expires
A retained part is not permanent. It fades in daylight, picks up scratches and oil from hands, and slowly stops being the thing it was meant to represent. Keep it too long and the whole run drifts towards a worn object. So a sample is usually sealed away when its batch ends, and a fresh one is taken for the next batch. The reference has a working life, just as the material does.
Notice too that the first part is not the best part. It is the part that was agreed on. If it carries a small, acceptable blemish, then that blemish becomes part of the standard and later pieces are allowed the same one. The purpose is consistency, not perfection; a run whose pieces are all alike is easier to assemble than a run containing one superb example and a hundred slightly different ones.
This is the same kind of answer as a line waiting for the smallest part. There the uncertainty about whether everything has arrived is collected into one bin; here the uncertainty about what acceptable means is collected into one object. Both turn a question that would otherwise be argued about repeatedly into something that can be checked at a glance.
There is an easy way to see why this is necessary. Ask two people to pick out a part that looks the same as last time, with no sample in the room, and they will usually bring back different pieces. Put one tagged part on the bench between them and the disagreement disappears. The drawer is not holding a spare component; it is holding the definition.