A verification of 240 upholstered dining chairs passed on a sample of 32 and recommended release. The photo appendix ran to nine images, and all nine showed the same chair, from the same angle, in the same corner. The buyer counted them before approving the balance payment and found that not one photograph could be tied to a unit that had been inspected.

The order that passed

A hotel group ordered 240 upholstered dining chairs from a factory in Foshan. One model, 240 units at $88 each, $21,120 in total. Thirty per cent was paid on order. The remaining $14,784 was due against a pre-shipment verification report, and the report recommended release.

It was a Level 2 order and quality check, quoted at two man-days. It recorded a lot of 240 units and a sample of 32 units drawn under ISO 2859-1 general inspection level II, code letter G. No findings of any class. Model, quantity, colour, appearance against the approved sample and twelve dimension checks all conformed. The recommendation was to release the balance.

The photo appendix ran to nine images. All nine showed the same chair. All nine were shot from the same standing three-quarter angle, in the same corner, under the same light. Not one of the thirty-two sampled units was photographed. One was.

The buyer read the report on a Tuesday, before approving the payment. The approval was scheduled for the end of the week, which meant the report was going to be paid against and not looked at again.

The coverage arithmetic in the appendix

A report recommendation applies to the lot, not to the units that happened to be looked at. When the appendix cannot be tied to the sample, the reader is being asked to extend what was seen on one unit across all 240, and the report gives no way to check whether that extension is fair.

None of the lines below is a lie. The inspector inspected 32 units and found them conforming. The photographs simply carry none of that information. That is the specific failure of a photo appendix: it is decoration attached to a finding, and when a lot has no findings it becomes decoration attached to the page.

How the buyer found it in twenty minutes

There is no dispute to resolve and no site visit required. The audit is a counting exercise on the document itself, and it can be run on any report before the money moves.

• Open the photo section and count the images. Nine.
• Open the sampling section and find the unit count. Thirty-two.
• Read the caption under each image and write down the unit it shows. All nine captions read the same way: the chair, general view. No unit is identifiable.
• Ask whether any image can be matched to a unit the findings are recorded against. None can.
• Divide the distinct photographed units by the sampled units: one of thirty-two, which is 3.1 per cent. Then divide by the lot: one of two hundred and forty, which is 0.4 per cent.

What the audit did and did not prove

It is worth being precise about that, because it is where a buyer's instinct goes wrong. The audit did not prove the inspection was bad. It proved that the document the buyer was about to pay against could not be checked by the buyer, which is the only function that document has.

An inspection that found nothing wrong and a report that cannot be read are two different outcomes, and only one of them is a quality signal. Reading the second one as the first is how a good inspection gets paid for a bad report.

The inspector could not answer the question

The buyer asked one question in an email: which unit is in photograph three. The reply came back the same afternoon. The photographs were taken on the last chair off the assembly line at the end of the run, with the finished goods stacked behind it. The inspector could not say which unit it was, because the unit numbers were on the production sheet three metres away in the workshop and nobody had carried them across to the camera.

That is the part that matters. The information had already been collected. The inspector knew, at the moment the shutter fired, which chair was in front of them, and the unit number was recorded on a clipboard. The report's findings section was able to name unit 17, unit 26 and unit 31, because that data was captured inside the inspection workflow. The photographs were taken outside it.

So the gap was not a failure of diligence. It was a failure of attachment. The unit identifier existed, and it never made it into the picture.

What a photograph can and cannot establish

The second row is the one that catches people out most often. A photograph of a chair says nothing about whether it is 460 millimetres tall. Without a rule or the measuring tool in the shot, it is an image of a chair, and its evidential weight is exactly the same as a photograph of a different chair in a different room.

The fourth row is the one that matters most for a report with no findings. An empty findings section and a findings section that was never worked on look identical from the outside. The photographs are the only place where that difference could have shown, and in this report they did not.

Repetition where coverage was needed

Counting the nine images as nine pieces of evidence is the natural mistake, and it is what makes the appendix look adequate at a glance. It is one unit photographed repeatedly. A unit needs two photographs to be documented properly: one overall, showing the unit and its identifier, and one close, showing the feature being claimed. The report had the first kind, and it was of a unit nobody could identify.

So the appendix was short by thirty-one units and long by seven shots. That pattern, repetition where coverage was needed, is what happens when photographs are requested as a deliverable without a specification attached to them. The repeatable shot gets repeated because it fills the slot, and the hero shot is the easiest one to repeat.

In practice the identifier does not need a production line to carry it. A numbered card set behind the unit, or the unit's own carton label while the unit is still in its carton, is enough to make every frame self-identifying. The cost is a card and a second hand.

The clause to put in the order

These are written to be pasted into a purchase order or an inspection instruction. They are about coverage and attachment, not about image quality or lighting.

• Every unit carrying a recorded finding is photographed twice: one overall frame with the unit identifier in shot, one close frame filling the image, with the measuring tool or a scale rule visible where the finding is dimensional.
• Every unit in the sample is photographed at least once, overall, with its unit identifier legible. Where the lot exceeds 500 units, cover a stated random subset of at least the sample size and name the subset in the report.
• Every dimension quoted in the report is photographed with the measuring tool in frame and the reading visible.
• Every file is named with its unit identifier and its check, and the report carries an index table listing unit, check and filename for each photograph.
• Photographs must show units drawn from the lot under inspection. Display models, showroom pieces and retained samples are not acceptable.
• The report states its own coverage in one line: units sampled, units photographed, units carrying a finding, and units with no photograph at all.

Why the clause costs nothing

The last line of that list is the one that changes how the report is read. A report that states its coverage can be audited in seconds instead of reconstructed. A report that does not has to be taken on trust, and the buyer has no way to tell the difference between an inspector who photographed every unit and an inspector who photographed one chair.

None of these clauses asks for more inspection time. The same man-day that produced nine photographs produces thirty-two, because the cost of the second chair is walking to the next unit. The clause adds a card, a camera roll and a line in the report template. It does not add a man-day, and it does not add to the invoice.

What the buyer did, and what it cost

The buyer asked for the appendix to be redone: every sampled unit photographed with its identifier, and the twelve dimension checks repeated so that the numbers could be matched to frames. The inspector proposed a re-visit instead, and the buyer held the balance payment until it was delivered.

The re-visit took one man-day, $199, and produced thirty-two photographs indexed by unit identifier, with the twelve dimension checks repeated and no findings. The delay was eleven days, during which $14,784 of balance payment stayed where it was. The inspection had been right the first time. The document had not been usable, and the cost of that was eleven days and $199 against a fit-out programme with a fixed opening date.

The original inspection had cost $398 at two man-days. The re-visit cost $199, half of it again, to produce the evidence the first visit should have produced and did not.

The buyer put the clause into the next order. The first inspection under it came back with thirty-four photographs covering all thirty-two sampled units and two findings, each finding tied to a named unit. Nothing needed re-checking, because the appendix was the verification record.

What photographs do not replace

A clause like this is often over-read. Thirty-two photographs of thirty-two units is still thirty-two units. The sample's weight against the lot comes from the sampling table and the accept and reject numbers, not from the number of pictures. Photographs make the sample legible. AQL makes it binding. A report needs both, and neither substitutes for the other.

There is a second trap on the other side. Specifying a photo count invites the same behaviour the count was meant to prevent, because thirty-four photographs of four chairs will satisfy a written requirement for thirty-four photographs. That is why the requirement is a coverage rule tied to sampled units and to findings, and why the report states the coverage itself rather than a count.

One line in the report

The shape of the document was familiar. A QC report example with the sections in the right order, the tables filled in, a photo appendix at the back and a recommendation at the end.

The whole case reduces to a single number that should have been in it and was not: how many of the units you inspected did you photograph. Nine photographs of one chair against a sample of thirty-two is not a quality problem. It is an auditability problem, and it is the cheapest kind there is to fix.

A report that says what it did not cover is worth more than a report that appears to cover everything. Our own format puts the untested items in their own section for exactly that reason, and it would be dishonest to publish photographs that read like coverage while the coverage is one unit in thirty-two.