A 45 mm gap opened up between the mattress and the guard rail on 600 children's beds. The inspection report that cleared them measured 35 mm, and that number was correct.

The order

An online children's furniture retailer selling across the UK and the EU placed an order for 600 toddler beds with a fixed side guard rail, each supplied with its own mattress, from a factory in Foshan. The bed is a beech frame with a slatted base 300 mm above the floor and a guard rail screwed to the head end and the long side. It is the kind of bed a two-year-old climbs out of in the dark and a four-year-old sits on while they eat. The order was worth $132,000: $168 a bed and $52 a mattress, $220 a set, 600 sets. Thirty per cent was paid as a deposit, $39,600, and the balance of $92,400 fell due when the factory asked for the loading date. The goods went out in seven 40 ft high-cube containers, about 86 sets to a container.

The buyer's specification ran to eleven pages. It called for beech, for the slats to be at least 13 mm, for the rail to be 250 mm tall at the head end and 200 mm at the foot end, for the mattress to be 120 mm, and for the mattress to be delivered with the bed rather than shipped alongside it. Three lines in that specification described the relationship between the mattress and the rail, and every one of them was written in a way that could not be measured.

The first of those three lines read: the clear opening between the top surface of the mattress and the underside of the guard rail shall not exceed 40 mm, measured with the mattress in the condition supplied. Thirty-five millimetres on a factory floor, with a brand new mattress lying on the deck and nobody sitting on it. The second said the mattress shall be 120 mm. The third said the rail shall stand not less than 280 mm above the mattress surface. The buyer had specified a clearance, a mattress thickness and a rail height, and had joined them with a condition that describes the showroom rather than the bedroom.

The factory built to it. On every one of the 600 beds the rail underside was at 455 mm above the floor and the mattress measured 118 to 122 mm, which put the opening at 33 to 37 mm on every unit. The inspection measured the opening on the 80 beds in the sample, found 33 to 37 mm against a limit of 40, and passed the lot.

The clearance the drawing defined

The opening is a difference between two heights, and it is worth writing it out, because that is where the specification went wrong. The rail underside sits at 455 mm above the floor. The mattress support deck sits at 300 mm. The mattress is 120 mm, so the mattress surface is at 420 mm and the opening is 35 mm. Thirty-five millimetres is what the factory built, on every bed, and it is what the inspection recorded.

The specification asked for no more than 40 mm, so the design carried 5 mm of margin. Five millimetres is not much for a dimension that involves a mattress, and the reason it is not much is that the drawing treated the mattress as a fixed component, in the same way it treated the deck and the rail. It is a beech frame, a slatted base and a foam block, and two of those three do not move.

The third clause explains why that 5 mm of margin could not be spent on the mattress. The rail top sits at 705 mm above the floor, so it stands 285 mm above the mattress surface, and the specification required not less than 280 mm. The rail could not be dropped 10 mm to claw back the opening, because that would have put it under its own height floor. The frame was drawn to sit in a 5 mm corridor, at the top of one limit and the bottom of another, and the mattress was going to be the thing that moved.

This is the point at which the case looks like a small tolerance argument, and it is not. A tolerance argument would be: the mattress was specified at 120 mm, it came at 118 mm, and 2 mm of thickness became 2 mm of opening. That happened, and it is inside the tolerance. What actually happened is that the mattress was never the dimension that mattered, because 120 mm is the thickness of a mattress lying flat and empty on a factory floor, and the mattress in the child's bedroom is a different shape.

What a 15 kg child does to a 120 mm mattress

The mattress was not substituted and it was not faulty. It was 120 mm of high resilience foam at the density the buyer's own specification set, it matched the approved sample, and it passed the material checks. The one thing the specification never asked for was how much of it would be left after a child had been sitting on it.

That question was answered on a bench at the factory, with a load. A 250 N mass was placed over a 300 by 300 mm area on the mattress and left for ten minutes. Across the eight mattresses pulled from the inspection sample, the compressed thickness came out between 97 and 110 mm. All eight were below the 115 mm the frames needed in order to hold a 40 mm opening, and five of the eight were below 105 mm.

The arithmetic is not subtle. A 15 kg child sleeping against the rail puts their weight through roughly 500 cm² of contact, which is about 2.9 kPa. The bench test applied 2.8 kPa. The test was not harsher than the child. It was the child, written down in a way that could be repeated, and it is the reason the answer is defensible: a load, an area, a dwell time and a measurement point.

And the compression is not even. A child sleeping against the rail depresses the mattress at the rail line by around 10 mm, which opens the gap to 45 mm. A child kneeling at the rail and pulling themselves up concentrates their weight in one place, and the local indentation at that point reached 23 mm, which opens it to 58 mm. The gap was widest exactly where a child uses the rail to get out of the bed, and 13 mm worse than the reading in the middle of the bed.

The measurement, and the condition it was taken in

The inspection report was not wrong. It recorded the opening correctly, and it recorded it under the condition the specification had asked for, which was with the mattress in the condition supplied. The report and the specification were in agreement, and both of them were describing a bed that did not exist.

A clearance that depends on a compressible component needs the condition stated with the number, because the number alone means nothing. Five things have to be fixed before a reading like this can survive being repeated in a warehouse in another country by a different inspector: the load and the area it acts over, the dwell time before reading, the position on the bed where the reading is taken, the instrument, and the state of the mattress underneath. A steel rule laid across a soft surface and read at eye level does not resolve half a millimetre at the rail line, and a reading taken in the middle of the bed is not the reading that matters.

What follows is what the same bed looks like under five different conditions. Only the first and the fourth would have been recorded on a report that said the opening was 35 mm.

Eighty beds, and the number that decided it

The lot of 600 fell under ISO 2859-1 code letter J, which gives a sample of 80 at general inspection level II. At AQL 2.5 that sample accepts 5 defects and rejects 6. The opening was measured on all 80 and came out between 33 and 37 mm, so the characteristic passed with 3 to 7 mm in hand on every single bed. On the dimension, there was nothing to argue about.

The compressed-thickness check was the one that moved the case, and it was not an AQL characteristic. A material property is not a defect count. It is a pass or a fail on the mattress itself, and one failure is enough to hold the lot. All eight mattresses pulled for the compression test came back below the 115 mm the frames needed, and the buyer stopped the loading with three containers sealed and the balance of $92,400 unpaid.

This is the structural point of the case, and it is easy to miss because the report reads like a clearance argument. If the inspection had measured the opening under load instead of unloaded, the lot would have failed on its AQL characteristic, at 80 out of 80, with the same sample and the same two days on site. The measurement was not the only thing that could have gone differently. The number on the report could have been the right number in the wrong place, and the inspection bought the buyer a specification problem instead of a manufacturing one.

The gap is not a property of the bed

Most furniture dimensions are properties of the product. A dining table is 740 mm high because that is the height it was made at, and it is 740 mm high whether or not anybody is sitting on it. A seat is 450 mm because that is the height it was made at. Those numbers can be read off a drawing and checked with a rule, and they mean the same thing in the factory and in the customer's house.

A clearance measured against a compressible component is a property of the product in use, and it is a different thing. The factory contributed two fixed numbers, 455 mm and 300 mm, and it controlled both of them completely. The mattress contributed a number that changed with load, with temperature and with time, and the specification asked for it the way you ask for a table height. The clearance was a difference, and a difference inherits the worst behaviour of the things being subtracted.

The same logic applies to everything else on the bed. A spring core mattress would barely have moved. This foam mattress lost 23 mm in ten minutes. The buyer had specified the mattress as 120 mm and had not specified what it does, because for a mattress lying on a deck, in the condition supplied, thickness and behaviour are the same property. They stop being the same property the moment somebody climbs into the bed, and the purchase order did not mark the moment.

It is worth saying plainly what this is not. The factory did not cut corners, did not substitute a cheaper foam, and did not misunderstand the specification. It built 600 beds that met every stated requirement, and every stated requirement was about a mattress that nobody was sitting on. A kids furniture inspection that only records a free-hand opening on an unloaded mattress has recorded a fact about the factory, not about the bed.

Rewriting the specification

The buyer rewrote the three lines and did not change the bed. The frames that were already built were correct, and the rail was not moved, because the rail could not be moved. What changed was the word condition, and what the specification now asks for instead of asking for.

The rewritten clauses, in the order they appear in the purchase order:

• State the mattress's compressed thickness, not its nominal thickness. A minimum of 115 mm under 250 N applied over a 300 by 300 mm area for 10 minutes is a requirement that can be inspected on a bench. 120 mm is a description of a size, and it is satisfied by a mattress that loses 23 mm of it.
• Attach the load to the clearance. The opening shall not exceed 40 mm with the mattress at its minimum compressed thickness. Written this way, the clause survives any mattress the buyer substitutes later, and it survives the factory changing foam supplier.
• Write the measurement position down. At the head end of the rail, 100 mm from the end post, where a child kneels to pull themselves up. At the rail is not a place, and the worst reading on this bed was 13 mm worse than the reading in the middle of the bed.
• Measure the opening from a fixed reference. The rail underside is 455 mm above the floor and the deck is 300 mm. Both are the factory's to hold. A clearance measured from a compressible component is an opinion on the day it is written.
• Set both limits on the opening, and keep the rail height floor. The opening shall also be at least 15 mm, so a thicker mattress cannot push the rail up or bear on it, and the rail shall stand not less than 280 mm above the mattress surface, measured at the compressed mattress thickness and not at the nominal one. Those two clauses together are what stop the answer being found by moving the rail instead of by fixing the mattress.
• Sample the mattress and hold the lot on the material check. A compressed-thickness failure is a fail on the mattress, not a defect count, and it does not wait for an acceptance number.

What the specification cost

The difference between a specification that works and a specification that does not is small in the factory and enormous after the container is sealed. These are the figures as the buyer modelled them, using the published man-day rate and the order as placed.

The remedy available before loading was the cheaper one: change the mattress. The buyer replaced 600 foam units at $52 with 600 spring core units at $68. The new mattress retained 117 mm under the written load, so the opening ran 35 mm unloaded and 38 mm loaded, back inside the 40 mm limit measured on the rail rather than on the deck. The net cost was $9,600 on a $132,000 order.

The alternative the factory offered was to re-hang the rail 10 mm lower on every bed, $22 a bed of rework and $13,200 in total. The buyer rejected it. It would have closed the opening to 25 mm, but it would have put the rail at 275 mm above the mattress surface, under the 280 mm floor the buyer's own specification set, and it would have fixed the symptom while leaving the purchase order still wrong for the next order.

The cost of the same change after loading is a different order of magnitude. A batch of 600 children's beds with a documented entrapment dimension does not stay in a warehouse. If the problem had surfaced in a retailer's own sample or in a complaint, the cost was going to be the repair of every unit, the re-packing, the freight, and a conversation with a trading standards officer, and the repair is not cheaper per bed at that scale than it was in the factory. Ninety-five dollars a set to put right and re-ship, $57,000 against a lot that had already been paid for.

The inspection itself was three man-days at the full verification level, $897. Against a $92,400 balance still unpaid, that is the cheapest thing in the table.

Two findings, and the decision that followed

The report carried two findings and kept them apart, because they were two different findings. The first was that the goods as manufactured conformed to the purchase order as written, on every dimension the order actually specified. The second was that the order did not specify the condition under which its own clearance was to exist, and that under the condition of use the clearance was 45 mm typical and 58 mm at the worst point on the bed.

Those two sentences are not contradictory, and a report that merges them is much less useful than one that separates them. A merged report invites the buyer to argue with the inspector. A separated report says: the goods are right, the document is not, and here is the number that proves which one is going to matter in a customer's house.

The buyer did three things with it. It stopped the loading with three containers sealed and the balance of $92,400 unpaid. It had the factory replace the mattress and produce 600 units retaining at least 115 mm under the written load, which put the opening at 35 mm unloaded and 38 mm loaded, back inside a 40 mm limit measured on the rail rather than on the deck. And it rewrote the three lines of the purchase order, so that the next 600 beds arrive with a specification that describes a bedroom and not a factory floor.

The frames were never the problem and the factory was never the problem. A 45 mm opening between a mattress and a guard rail is not a defect in a beech frame that was built to a drawing and measured on every unit. It is a number that was never written down, and it was always going to be exactly the wrong number the moment somebody sat on the bed.