2026-10-19 Quality Seller's notes

Sample and Bulk: How They Are Supposed to Be the Same Thing

Peter Yang's own column. He sells from a trading company and wrote this from inside one, after fifteen years on the other side of the table. The view is his. SUNWARD INDUSTRY sets the commercial terms, and a price comes as a written quotation.

A sample taken from ordinary production goods is not a softer version of the batch you receive - it is a piece of that batch, and the way to make a difference checkable rather than arguable is a recorded sample: take the sample one to three times, and each time make a watermark record with it, carrying the time and the place the sample was taken from. The bulk matching the sample is the normal case, and a mismatch is the exception that needs explaining. Nobody has ever looked at a watermark and concluded that a shipment was safe, and I would not claim more than that.

A sample taken from ordinary production goods is not a softer version of the batch you receive. It is a piece of that batch. People expect a sample to be a promise and the bulk to be the reality, and the normal case in this trade is the other way round: the reason is dull and practical rather than reassuring, because to keep it simple a sample is normally taken from ordinary production goods.

Why is the sample not the same as the bulk order?

The premise of the question is usually backwards, and that is worth saying first because it changes what you are looking for. A sample is normally drawn out of an ordinary production run, so the bulk matching the sample is the normal case; a mismatch is the exception that needs explaining, not the other way round. What one sample cannot settle is whether the bulk will match it, and the way to settle that is not a better sample but a recorded one. Take the sample more than once, and each time make a watermark record with it, carrying the time and the place the sample was taken from — that is the whole of it, a few marks that cannot be put there after the fact.

Why the sample is normally the same goods

That is the ordinary practice. Rather than make something special for the purpose, people draw the sample out of a normal run — and that is exactly what makes a sample worth anything. A sample produced separately, or drawn from some other run, is a fact about that sample and not about what you are going to receive. It tells you what somebody could do when the whole purpose of the exercise was to do it.

And if the production is normal, then generally there is not going to be a large quality difference between the sample and the bulk order. I say that to buyers, I believe it, and it is also the position I would take with my own customer if the question were turned around.

Which makes the expectation that the bulk is where the truth finally arrives worth examining. Not because suppliers are dishonest — that is not the mechanism, and I have no evidence that it is the mechanism here — but because the premise itself does not hold.

None of that makes one sample a guarantee of everything that follows, and I would not want to be quoted as saying it does. Production varies. Raw material deliveries differ, one run is longer or shorter than the last, and the shift that runs the machine is not always the same shift. That is ordinary variation rather than anything sinister, and it is why nobody sensible treats a single sample as the specification for every future batch. What a properly taken sample gives you is a fair picture of how this supplier's ordinary production behaves. That is a smaller claim than people hope for, and it is the one I would stand behind.

What a repeated, recorded sample is for

The case I use when I want the question settled rather than agreed is the recorded one, and its purpose is not to catch anybody and not a trick to be turned on somebody. It changes what kind of question is being asked. Without a record, the question of whether the bulk will be the same as the sample has no answer except an opinion, and both sides end up relying on how much they trust each other's character. With several recorded samples, a claim that the bulk is different has something to be checked against. The disagreement moves from what somebody believes to what somebody can look at, and that is the entire difference between the two.

The record itself is simple. I take the picture with a camera that stamps the time into the image, so the photograph carries the moment it was taken. It also carries the surroundings - the line, the drum, the place the sample came out of - and the sample in the frame with them. One picture holds the time, the place and the goods together, and that is the point of it. A photograph of a sample on its own can be made anywhere. A photograph of a sample with a clock and a floor in it cannot be moved anywhere without losing something.

Video would be better, and I have thought about it, because it carries the sound and the movement and the people. The reason I do not use it is that I am not certain a video gives me the timestamp in a form I could rely on later, and the whole value of the record is that it is fixed at the moment it is made. I would rather use the thing I know holds.

How many samples is a practical question rather than a principled one. I take one to three. If the result needs more than that, I take more — tested separately, or pooled into one sample and tested once, which is the ordinary way to get a single figure for a batch. The part I am careful about is that every one of them is recorded and watermarked, because a sample nobody can place is a sample that proves nothing, however good the number on the report is.

And the honest reason for taking more than one is not to hunt for differences that are not there. Standard product off one line inside one batch is very nearly the same material. The variation that matters shows up between batches — a different raw material delivery, a run that came out longer or shorter than the last, a different shift on the machine. So a second sample is not ammunition. It is a second record that either agrees with the first or does not, and the second case is the one worth noticing.

An opinion about quality is what a plain sample gives you. It settles nothing between two people who have already decided to get along, and it settles nothing at all between two who have not.
A fact about the batch is what several recorded samples give you. It does not make a shipment safe, but it does mean a disagreement about what actually arrived has something in it to be argued from.

I want to be accurate about what that is worth, because overselling a method is precisely the thing I criticise elsewhere. It is one process detail out of several. Nobody has ever looked at a watermark and concluded that a shipment was safe.

Why no single step makes a shipment safe

There are a good many other process details that let both sides settle. What a certificate actually covers, and what it quietly does not. What happens when a batch does not match, and how fast. Who decides, and with what standing. Those things exist, they matter, and I am not going to set them out here as a list. Partly because a list handed to a buyer is a tool for confirming a supplier they have already chosen, and it does nothing at all about the ones they have not. Partly because a method like that is aimed at my business as much as at theirs, and I would rather not hand it over.

What I can tell you is what the work is. And I am not an expert in how to make a customer feel safe. I am a person who works in a service business, and the honest description of my position is that I am more useful inside a transaction than standing in front of one.

There is one thing I would add about all of it, process details included. What a supplier says when you ask is a different thing from what they do when you assume. The second is the ordinary course of the business and nobody is performing in it. The first is work somebody has to do, and how much work they will do is itself information. That is not a trick to be played on anyone, and it is not a line to be handed to a buyer. It is only that the cheapest quality control in any transaction is finding out whether the other side will do work when nobody is making them.

So if you want to know what any of this is worth, the answer is not something I write down for you. Do business with me and you will understand it in a way that no article manages to explain. And if what you want is my judgement on a particular factory rather than on the general case, you can buy my time for it — a factory inspection, for instance — and take the answer from there rather than from my word about it.

Questions this note answers

Why is the sample not the same as the bulk order?

Because the premise of the question is usually backwards: a sample is normally drawn out of an ordinary production run, so the bulk matching the sample is the normal case and a mismatch is the exception that needs explaining, not the other way round. What one sample cannot settle is whether the bulk will match it, because production varies - raw material deliveries differ, one run is longer or shorter than the last, and the shift that runs the machine is not always the same shift. The way to settle it is not a better sample but a recorded one: take the sample one to three times, and each time make a watermark record with it, carrying the time and the place the sample was taken from.

How can I make sure the bulk order matches the sample?

Not with a better sample, but with a recorded one. Take the sample one to three times, and each time make a watermark record with it, carrying the time and the place the sample was taken from - a few marks that cannot be put there after the fact. Without a record, the question of whether the bulk will be the same as the sample has no answer except an opinion, and both sides end up relying on how much they trust each other's character. With several recorded samples, a claim that the bulk is different has something to be checked against, and the disagreement moves from what somebody believes to what somebody can look at.

Does a sample guarantee the bulk order will be right?

No, and no single step does, including a recorded sample. A watermark record gives a claim about a difference something to be checked against, which is a narrower thing than a guarantee - nobody has ever looked at a watermark and concluded that a shipment was safe. There are other process details around it, and I am not going to set them out here as a list: partly because a list handed to a buyer is a tool for confirming a supplier they have already chosen, and partly because a method like that is aimed at my business as much as at theirs. The way to find out what any of this is worth is to trade with me, and if you want my judgement on a specific factory, you can buy my time for it.

How do I prove a sample came from the batch I will receive?

By taking the picture with a camera that stamps the time into the image, so the record carries the moment it was made, and by letting the frame carry the surroundings as well - the line, the drum, the place the sample came out of - with the sample in the same frame. One picture then holds the time, the place and the goods together, and a photograph of a sample on its own can be made anywhere. Video would carry the sound and the movement as well, but I am not certain a video gives me a timestamp I could rely on later, and the whole value of the record is that it is fixed at the moment it is made.

How many samples should I ask for?

One to three is what I take, and if the result needs more than that, more - tested separately, or pooled into one sample and tested once, which is the ordinary way to get a single figure for a batch. Every one of them is recorded and watermarked, because a sample nobody can place proves nothing however good the number on the report is. And taking more than one is not about hunting for differences: standard product off one line inside one batch is very nearly the same material. The variation that matters appears between batches - a different raw material delivery, a run that came out longer or shorter, a different shift on the machine - so a second sample is a second record that either agrees with the first or does not.