The sample is wrong. Should I change factory?

Not on the first sample, and not before you know which of the three causes it was. A sample that fails because the specification was ambiguous is a specification problem, and moving to another factory reproduces it with a new supplier. Change factory when the diagnosis says the factory cannot hold a tolerance you have written down clearly.

How many samples should I expect to approve?

More than one is normal, and the number is a function of how detailed the specification was rather than of the factory's quality. The more of the specification that was written before the first sample, the fewer rounds it takes.

Who pays for a second sample?

This is worth agreeing before the first one, in the same conversation as the tooling and the packing. A sample that fails against a clearly written specification is a factory cost; a sample that fails against an ambiguous one is a specification cost, and the honest answer is often that both sides contributed.

The color is slightly off. Does that matter?

It matters if you cannot say what "off" means. color is the most common disagreement in this business and the one with the clearest fix: state the standard and the light it is checked under. Without those two, two acceptable batches will be rejected against each other — which is what the glossary entry on colorway tolerance describes.

The sample is perfect. Am I safe?

No, and this is the specific trap the sample sets. A sample confirms what it tested and nothing else. It does not confirm a fall rate, a salt-spray standard, a strand that survives a carton, or a tolerance that will hold on the two-thousandth piece. Those are all specification lines, and the sample is not evidence about any of them.

What should I write down after a failed sample?

Three things: what the specification said, what the sample did, and which of the two was wrong. That last one is the judgement that matters, and writing it down turns a rejection into a change to the specification. Without it, the second sample is a guess.

What to Do When a Sample Fails

First: a failed sample is the cheap failure

Nothing has been committed. The tooling has not run, the deposit has not moved, and the factory still has every reason to make the second attempt work. Compare that with the alternative — discovering the same problem on a production run — and the sample round looks like what it is: the cheapest inspection in the order.

So the productive question is not accept or reject. It is what did this tell us.

Diagnose before you decide

A sample fails for one of three reasons, and they have three different responses.

Cause	What it looks like	What to do
The specification was ambiguous	Two people read it differently and both are right	Fix the specification. This is the most common cause and the only one worth fixing before anything else.
The sample is outside the specification	The specification is clear and the sample misses it	Reject against the named field. This is the case where a factory has to re-run it.
Neither — the specification was wrong	The sample is correct but the product is not what you wanted	Change the specification, not the factory. This is a design change and it is normal.

Only the second row is a factory problem. Moving factory in the first or third row reproduces the same failure with a new supplier and a new lead time.

The three questions that turn a rejection into a fix

What did the specification say? Read the wording, not your memory of the intention. If the field was not written down, it was not specified.
What did the sample do? Measure it against the tolerance, not against a feeling. A "slightly off" color is not a finding; a color outside a stated standard under a stated light is.
Which of the two is wrong? Write the answer down. That sentence is the change to the specification, and it is what the second sample is built from.

What a sample cannot tell you

This is the part that catches buyers who approve quickly, and it is the reason our own records carry a field list rather than a photograph.

A sample confirms only what it tested. It does not confirm:

a fall rate at a stated water temperature
a salt-spray standard on the hardware
whether a strand or a skirt survives a full carton
whether the tolerance holds on the two-thousandth piece
whether the pack survives the same carton

Those are all specification lines, and every one of them is checkable — just not by looking at a sample. The sample-to-production gap is the term for exactly this, and it is on this site because it is the failure mode that survives a normal quality conversation.

The field guide for the second round

Before the second sample, add these to the specification if they are not already there:

[ ] The tolerance band for every dimension, in mm or grams
[ ] The color standard and the light it is checked under
[ ] The test method for anything behavioural: fall rate, tracking, buoyancy
[ ] The salt-spray standard and hours for every metal part
[ ] The pack, and what the product looks like after a full carton

A second sample made against those is a different document from the first one. That is the whole point of a failure.

Where to go next

Where programmes go wrong — the four failure modes and the control for each
How to compare two factory quotations
What a QC record should contain — the fields to ask for before the next run
From sketch to tooling — why a change after the mould is cut resets the programme
A first order from China, step by step
The specification builder
Send a specification

What to do when a sample fails