A restaurant group ordered 480 dining chairs and 240 oak tables for nine kitchens. Three of the eighty chairs the inspector load tested had an armrest that moved under the buyer's own load requirement, which was inside the AQL acceptance number, so the lot passed and the faulty chairs would still have been loaded. The report recommended holding a fifth of the balance until four conditions closed. The buyer was ready to pay it in full that afternoon.

Nine kitchens, 480 chairs, and a request for the money on Wednesday

A restaurant group was refitting nine kitchens. The order was 480 upholstered dining chairs and 240 solid oak dining tables, 720 units in one forty-foot high-cube container. The order value was $184,800. Thirty per cent was paid as a deposit when the order was placed, and the remaining seventy per cent, $129,360, was due when the factory said the goods were finished.

On the Wednesday the invoice for the balance arrived with a set of photographs attached and a note that production was complete and the container was booked for Friday. The buyer forwarded the email to us with one line above it: should I pay the balance China supplier. It was a real question and it did not have a yes or a no answer, because the answer depends entirely on what has been checked and what has been written down.

The photographs were good. That was the problem. They showed a dozen stacks of finished chairs under decent light, and a table top with a finish you could see the grain in. A stack of finished goods photographed on the factory floor proves the goods exist. It does not prove the arms are tight, that the colour is the colour, that the cartons will survive four to eight weeks of ocean transit, or that the chairs are the height the kitchens need. None of that is visible in a photograph of a stack.

What the buyer sent, and what was not in the file

Before anything can be judged, the documents the judgement will be made against have to exist. This order came with three: a purchase invoice listing model numbers, quantities and unit prices, a specification sheet for the chairs and the tables, and a photograph of the approved colour.

Three things were missing, and all three were things only the buyer could supply. There was no tolerance schedule. The spec sheet gave seat height, table height and overall dimensions as single numbers, with no plus or minus and no drawing to fall back on. There was no physical approved sample: the colour existed as a picture on a phone, and the picture had been through at least two rounds of message compression. And there was no packaging specification, so there was no drop test to run and no protocol to record a result against.

None of those three gaps is the factory's fault, and none of them stopped the goods being made. They only matter later, when somebody has to decide whether what arrived is what was ordered, and whether the money should go.

The five checks, and what each one produced

The inspection was booked as an Order plus Quality check, two man-days, with a loading supervision slot on the Friday. One inspector on the Thursday, one on the Friday morning with the container. The lot was 720 units, which at general inspection level two is a sample of 80, drawn from across the stacks and not from the pallet nearest the door.

Each of the five confirmations produced a result. Two were closed cleanly, two were closed with a qualification, and one could not be closed at all.

What the result was worth

Four minor surface marks and three armrest failures. Everything else on the order was as ordered. That is a good result on the goods, and it is not the same thing as a recommendation to release $129,360.

Three chairs out of eighty, and a lot that still passed

The armrest finding is the one worth explaining, because it is where most buyers' expectations and the sampling standard come apart. The buyer's specification required an armrest that would not move under a static outward load of 700 N. The inspector applied the same load to the armrest fixings of every sampled chair, pushing outward at the top of the arm.

On three of the eighty units the arm moved at the fixing. The joint did not fail outright. It shifted, and the shift was measurable and repeatable on the same three chairs.

Under the grading used for this order, an armrest that moves under a load the buyer's own specification requires is a major defect. At general inspection level two, on a sample of 80, the acceptance number for major defects at AQL 2.5 is five and the rejection number is six. Three is below five. The lot passes.

The third chair is the reason the armrests are a whole-lot problem rather than three chairs. The factory knows about three. The report's projection is about 27 across the container. Nobody can act on a projection without checking all 480 chairs, and that check costs the factory a day and a half, which is a reasonable thing to ask for and a hard thing to refuse.

AQL is a tool for deciding whether to accept a lot, not a guarantee that every unit in it is sound. It is deliberately tolerant of the occasional bad unit in a large batch, because stopping a shipment over three chairs in eight hundred usually costs more than replacing them. That is a sensible rule for a buyer who has already paid, and a bad rule for a buyer who has not.

Why the report said conditional release

The recommendation section gave the buyer three words it had to interpret: conditional release. That is not a softer version of release. It means the report is asking for something before the money moves, and the something has to be named, or it is not a condition, it is a mood.

The report named four, and put a number against them. Release $103,488 of the balance and retain $25,872 until: the affected chairs are reworked and the count confirmed at the loading check; the buyer issues a written tolerance schedule and the inspector re-grades the 80 sampled units against it within 24 hours, while the goods are still on site; the buyer supplies a physical approved sample and the inspector confirms the colour against it before loading; and loading supervision is booked for the Friday with the seal number, the stowage and the carton condition photographed.

None of those four requires the factory to make anything different. Three are paperwork the buyer can produce in an afternoon, and one is a supervisor standing in a warehouse. That is what a conditional release usually looks like when the goods are sound and the file is thin.

On the Friday the conditions closed. The factory tested all 480 chairs rather than the three in the sample, found 31 with the same fixing, reworked them before the container was sealed, and the inspector confirmed the count and the colour against the physical sample at loading. The retained $25,872 was released on the Monday.

It was closable because the goods were sound and the file was thin. Neither of those is guaranteed. The value of naming conditions before the money moves is that it turns an argument about the goods into a list of things to do by Friday.

The three sections of the report that decide your money

A verification report has eight sections, and they are not equally useful to a buyer deciding about a payment. The counting and comparing sections are where the evidence comes from. The three that carry the decision are the reference documents section at the front, the findings and recommendation section near the back, and the section on items that could not be verified at the very end.

The reference documents section decides whether the rest of the report means anything. If the report does not name the documents it judged against, then a finding of conform is an assertion rather than a comparison, and a finding of non-conform cannot be defended by either side. In this case that section recorded a purchase invoice, a specification sheet with no tolerances, and a photograph standing in for a physical sample. The report was still valid. It was valid against a thinner set of references than the buyer assumed.

The findings and recommendation section is where the graded list and the conclusion live. It is also the section most buyers read first and least carefully. The conclusion here said conditional release and then listed the conditions, which is the most useful sentence in the document, because it converts a judgement into a list of actions with an end point.

The section on items that could not be verified is the one that decides the payment, and it is the one most buyers skip. It is also the one most inspection companies write reluctantly, because it is the part that says what the inspection did not cover. If it is empty and the report is short, you have bought a light check. If it is full, the recommendation above it is conditional by definition, whatever words the recommendation uses.

What counts as evidence, and what only feels like evidence

The distinction that decides this kind of argument is between evidence and reassurance. Evidence is something with a number, a limit and a method attached. Reassurance is something that makes you feel the situation is under control.

A factory photograph is reassurance. So is a production-complete email, a packing list the factory generated for itself, a certificate with no batch number on it, a sample approved as a JPEG, and a report that names no reference documents. All of these can be entirely true, and none of them can be argued with.

• A dimension with a tolerance next to it. 470 mm plus or minus 2, measured on the finished unit. Two numbers can be discussed. One number can only be believed.
• A comparison against a physical object. Cloth against the swatch, the finished chair against the approved chair, the carton label against the packing list. A screen has a colour and a scale that the goods do not have.
• A test with its method written down. The load, the direction, the standard clause, the reading. A chair tested at 700 N outward with an arm that moved at 640 N is a finding. A chair described as having passed the test is not.
• A count performed rather than reported. Cartons opened, units counted, against a list the buyer supplied rather than the factory generated.
• A document that names what it judged against. Without that, the report is an opinion with photographs attached.

Deduct, withhold, or reject

Three responses are available, and buyers pick the wrong one constantly because the middle one feels like mistrust. The four minor surface marks on the hidden faces of four chairs are not worth a percentage of a balance. They are cosmetic, and they are worth a few dollars a chair at most. Deducting four times a small number against the invoice is the right move. Holding ten per cent of the balance until the factory polished them is not, and it is expensive: the container is booked, the sailing date is fixed, and holding goods on the far side of the world runs into five figures.

The three armrest failures are a different matter. They are a function failure on goods going into commercial kitchens, where the failure will be found by staff and reported by customers rather than by a laboratory. That argues for fixing them before the container is sealed, not for paying and fixing later. The factory had to find them, and finding them means testing all 480 chairs, not the three in the sample.

Withholding money is for the items you cannot resolve without the factory's involvement, and it should be written with an end point. Deducting is for items that are already quantified. Rejecting is for a critical finding, and it should be almost never: on this order there were none.

Write the holdback as an amount, not a percentage. The balance is already a percentage of the order, and holding twenty per cent of the balance is a number two people have to compute under time pressure. $25,872 is a number that can be written on an invoice and argued about once.

Six lines that would have made this decision simple

None of this needed an argument if the order had said six things. All of them belong in the purchase order or the specification sheet, at the point where they cost nothing to write.

• Name the tolerances. A table of every dimension on the order, with a figure and a limit, and a statement that a measurement outside the limit is a major defect. Twenty lines here removes the entire argument about whether 470 mm is right.
• Send the sample to the factory and keep one. A physical approved unit, sent to production before cutting, with the inspection company holding a copy so the comparison can be made on site against an object rather than a picture.
• Name the packaging standard and the test. Which carton, which drop height, which protocol, recorded in the report. A test nobody specified cannot be performed, and a test that cannot be performed cannot be reported.
• State the load requirements for the parts people lean on. Armrest, backrest and handle loads with numbers, and the direction of each load. This is where a furniture specification is normally silent, and where the failure usually is.
• Say which sampling standard and which defect classes apply, and what happens at each one. General level two, critical at zero acceptance, and the arithmetic of what a lot pass does and does not mean for an individual chair.
• Tie the balance to a milestone, not to a date. The balance is released when the report is issued and the loading check is passed and the named conditions are closed. A payment date is an instruction to rush a factory. A milestone is an instruction to finish.

The part that is easy to leave out

The last line of that list is the one buyers argue about when it is too late, because at invoice time it reads as a refusal to pay rather than as a plan. Written at order time it reads as what it is: a decision about when the factory stops working, made before the factory has spent the last two weeks of the container assuming the money is coming.

There is a version of this order where the buyer pays the balance on the Wednesday, the armrests go out to nine kitchens, and somebody has to explain to a customer why the chair moves. Three armrests out of eighty in the sample, a thin file, and goods that were otherwise right. Every one of those facts would have been argued about on a Friday afternoon with a container already booked.

The point of an inspection is not to find defects. On this order it found three, and what protected the money was the section of the report that said what had not been checked. None of it requires the buyer to distrust the factory. It requires the buyer to know what has been checked, and to have written down what happens if it has not. That is a smaller ask than it sounds, and it is the difference between a balance payment that closes a transaction and one that gambles with it.