Sampling Inspection of a Jewelry Batch: Checking a Delivery Without Opening Everything

A repeat order lands and the cartons are heavier than the last one. Checking every piece would take the better part of a week, and the listings are already live. So the box gets opened, a few pieces come out of the top layer, they look fine, and the delivery is booked in. That is sampling inspection of a jewelry batch done badly — not because too few pieces were checked, but because the pieces were chosen by whatever the packer happened to put on top.

Below we cover how to pull a sample that represents the batch, how to size it against the risk in front of you, when a partial check stops being defensible, and what to record so the check means something later.

What a sample can and cannot tell you

Sampling answers one question: does this batch behave the way the process that made it behaves? A sample catches faults that repeat — a mould that shifted, a plating run that came out cold, a clasp supplier who changed something. Those faults touch many pieces, so a modest sample finds them.

Sampling is weak against faults that occur once. A single dented pendant in a drawer of clean ones can pass every sample you pull and still reach a customer. Accepting that is part of using the method honestly, and it is why sampling pairs with a returns policy rather than replacing one.

Random has to mean random

The sample must come from the whole batch, not from the accessible part of it. Open more than one carton. Reach past the top layer. Take pieces from the bottom of pouch bundles, from the middle of the run, and from the last carton packed, because the end of a run is where fatigue shows.

Number the packages and use a list of random numbers to choose which ones you open. It sounds fussy until you notice how consistently a human hand reaches for the same positions. If a supplier knows which pieces you tend to check, those are the pieces that get the extra minute at the polishing wheel.

Sizing the sample against the risk

Sample size follows risk, not carton size. A model you have run many times, from a workshop whose last several deliveries passed cleanly, justifies a light check. A new model, a new alloy, a new finish or a first order from anyone justifies a heavy one.

Raise the sample when anything upstream changed. A new mould, a different stone supplier, a change in chain source, a gap of months since the last run. Each of those resets the assumption that this batch resembles the last one. The trade-off between sampling and full inspection comes down to which of those conditions apply.

Split the batch before you sample it

One delivery is often several populations wearing the same packing list. Two metals, three sizes, pieces cast in different sessions, a portion plated on a later day. Pulling one sample across all of them hides problems inside the averages.

Sample each population separately. If the packing list does not tell you where the divisions are, ask before you start; a workshop that runs its own bench knows which pieces were finished together. Our quality control and packaging page describes how that grouping is kept visible through to dispatch.

When a partial check is not enough

Some deliveries do not suit sampling at all. Personalised pieces are the clearest case: every item carries different text, so every item is unique and a sample tells you nothing about the piece with the misspelling. Check those individually, without exception.

Full checks also earn their time on a first production run of a new model, on any batch where the previous delivery failed, and on pieces where a fault is unrecoverable after sale. A first run has a second purpose too, since it shows where the sample and the production run diverge.

Recording what was actually sampled

An unrecorded check is a memory, and memories lose arguments. Write down the date, the batch reference, which cartons were opened, how many pieces came out of each, and the identifier of every piece inspected. Mark each sampled piece so it can be pulled back out if the batch is questioned.

Record the passes as well as the failures. A log of clean checks is what lets you defend a lighter sample next time, and it is the only evidence that a fault arriving in the third delivery was absent from the first two.

What a failure in the sample means

One failure is a signal, not a verdict. The response depends on the type. A dimensional or weight failure usually points at the process, so widen the check immediately and measure across the whole population; the same reasoning drives the weight and dimension check done before dispatch.

A cosmetic failure on one piece may genuinely be isolated. Pull a second sample from different cartons before deciding. Two failures of the same type in two independent samples is no longer chance, and the batch moves from sampling into a sort.

Sampling at the workshop instead of on arrival

A sample pulled before dispatch is worth more than the same sample pulled after, because pieces that fail have not yet crossed a border. Ask for the check to happen at the bench, with photographs of the selected pieces against a rule, and specify that you choose the identifiers rather than the packer.

That arrangement does not remove your own incoming check. It changes what that check is for: confirming the delivery matches the pieces you were shown, rather than discovering the batch for the first time.

Set your sampling level from the answer to one question: what changed since the last delivery that passed? Nothing changed and the history is clean, so a light sample across several cartons is enough. A new model, a new finish, a new supplier or a fault in the previous batch, and the sample goes up until you have seen enough pieces to trust the process again. Write that rule down once and apply it mechanically, so the size of your check never depends on how busy the week is.