One hundred and forty solid oak dining tables stood on three legs, and the fourth was 3 mm in the air. The floor was flat, the frame was square, and the factory had measured every leg at the wrong end of it.

The order, and the one line that had a number in it

A European home-furnishing retailer was building a spring dining collection and ordered one hundred and forty solid oak dining tables from a factory in Guangdong. Each table was 1,800 by 900 mm with a top 750 mm above the floor, a hardwood apron and frame, four square tapered legs, and a plastic glide under each foot. Nothing about the order was unusual and nothing about it was extravagant. It was a dining table that had to stand still in a restaurant.

The order carried one line about levelness. All four legs shall rest on the floor at the same time, and the table shall not rock. One sentence, no datum, no instrument, no condition. It is the kind of line everybody agrees with, because who argues with a table standing still.

Production ran. The tables were finished, packed and stacked, and the buyer's goods-in manager had already booked the container. At the pre-shipment visit the inspector started with the wobble check, because it takes thirty seconds, and because on a table it is the check that finds more than it costs. The first table rocked.

Ninety seconds of measuring, and two possible answers

A rocking table has two possible causes and they need completely different answers. The floor may not be flat. Or the table may not be. The inspector did not guess which one it was, because getting that wrong is how a factory ends up being blamed for a building defect.

The first test was the rotation test. The table was lifted a few centimetres, turned a quarter turn on the spot, and set down again. If the rock travels with the table when it is rotated, the fault is in the table. If it stays in the same place on the floor, the fault is the floor. The table was rotated three times in three directions and the same leg came up clear every time.

The second test was the press-down test. The table was left standing and pressure was applied in turn to each of the four corners. Each time the table rocked onto the opposite diagonal. The corner that rose was always the one opposite the corner being pressed, which is exactly what a table standing on three points does. The two legs along the diagonal being pressed carry the load, and the other two lift.

With the table resting on its three legs, the gap under the fourth was measured with a feeler gauge. Three millimetres. The same reading was then taken on thirty-two tables spread across the lot rather than taken from the top of the stack, and nine of them came up with a leg clear of the floor between 1.5 and 3.5 mm.

What a 3 mm gap does to a table

The number is small, and that is precisely why it survives inspection. Nothing is broken. The frame is sound, the joints are tight, the top is flat, and the table photographs perfectly. What it does is move, and a table that moves is a table people cannot use comfortably.

The mechanics are worth setting out, because they explain why the fault shows up in service and not on the delivery note. Three legs on the floor define a plane. The fourth leg sits 3 mm above it, so the table stands level on three points and waits for somebody to lean on it. Press the corner over the clear leg and the frame rotates about the diagonal joining the two legs furthest from that corner. The clear leg comes down 3 mm and the opposite corner rises 3 mm. The rock amplitude is the full 3 mm, corner to corner, every time.

In a restaurant that is a table that moves. A diner leans on an elbow and the plate slides. A glass walks towards the edge. Over a year of that, the frame is being twisted 3 mm each time, and the joint that absorbs it is the leg-to-apron joint on the corner that never touches the ground while the table stands empty.

Under load the asymmetry becomes visible. With 75 kg at the centre of the top, as the static load test in EN 1730 applies it, the table lifted on the corner over the short leg instead of distributing the load across four legs. The short leg does no work at all. The other three carry the table and its load through an eccentric path, and the frame is the part that pays for that.

So the failure arrives as a warranty claim rather than as a rejected container. The bolts at the corner blocks back out, the joint opens at the apron line, the table develops a crack where the leg meets the frame, and by then the cause is a number that was taken at the wrong end of a piece of wood two months earlier.

The floor was flat, and that is worth stating

The packing area was power-floated concrete. It was measured, because a finding that blames the factory should not rest on an assumption about the floor, and because the buyer's specification was about to be rewritten to include a floor reference, and that clause needs a number to work from.

A one metre straightedge was laid on the floor in three places and the maximum gap under it was read with a feeler gauge: 1.0 mm. For reference, the German construction tolerance DIN 18202 puts a finished floor at 4 mm over a one metre gauge, and at 1 mm over a 100 mm gauge for a floor with high requirements. A 1.0 mm reading over a metre is a flat floor by any measure.

But a straightedge one metre long is a shorter instrument than the table. The leg pattern spans 1,620 by 820 mm, so the question that matters is whether the floor is flat across nearly two metres, not across one. The inspector measured across the footprint as well, at 2 m, and got 1.5 mm. Then he repeated the test with a reference table from an earlier batch that had passed its own inspection, standing on the same floor. The reference table sat on all four legs. The floor was exonerated.

That is the step that separates a furniture finding from an excuse, and it also produces the more useful lesson. The factory floor was flat, and the floor in the buyer's restaurant will not be. A table that is exactly right on a power-floated slab can still rock on an uneven screed or a sprung timber floor, and no amount of inspection in Guangdong can know what is under the customer's table. The specification has to deal with that. The factory cannot.

The frame was square. The leg was short.

With the floor ruled out, two candidates remained. A leg that was the wrong length, or a frame that was out of square. They have different fixes, so the distinction had to be made on paper rather than in a conversation.

The diagonals of the leg pattern were measured on a table that rocked: 1,816.4 mm one way and 1,817.9 mm the other, a difference of 1.5 mm. The frame was square inside its own tolerance. Then the legs were measured from the shoulder, where the leg meets the apron, down to the bottom of the finished foot. Three read 448.0 mm. One read 445.0 mm. The table was level, square and correct, and one of its legs was 3 mm short.

That last detail is the whole case, and it explains why the fault survived to the packing stage. The factory checked leg length at the shoulder, because that is where a tape goes while the leg is still on the bench and the top is not yet on. Every one of those four readings was 448 mm, because the legs were all cut to the same length and all fitted at the same height. The 3 mm was at the other end of the leg, in the section that gets planed off after cutting, and nobody put a tape on a finished foot.

The inspector asked for the cutting and sanding records. The cutting list showed a batch of legs cut to 448 mm nominal with a 1 mm tolerance. The sanding station was set from the first leg of each bundle of ten, and one blank in a bundle came through with 3 mm more material still on the bottom face because the rough-sawn stock thickness varied and the setting was not re-cut for it. One short leg in ten, four legs to a table, and about a quarter of the lot ended up with a table carrying one.

This is a process setting, not a careless assembly. Nobody cut a leg short on purpose, and nobody ignored a measurement. The measurement simply was not taken at the end of the part where the error was.

What the report had to say, and what it had to leave out

The report recorded what was measured, under what condition, with what instrument, and how it differed from the specification. It did not carry the phrase within tolerance, because that phrase is a conclusion and a conclusion needs the numbers behind it.

The finding was graded Major. The tables are safe, the frames are sound and nothing about them will injure anybody. They do not do what the order said, which is stand still, and a table that moves in a restaurant is not fit for the service it was bought for.

One note in the report mattered more than the finding itself. These tables would have passed the 5 mm wobble limit that general on-site furniture checklists work to. They were written up because the buyer's purchase order specified no rocking at all, and the buyer's number was stricter than the inspector's default. That is a sentence worth reading twice, because it says the finding came from the specification and not from the inspector's opinion, and it says exactly where the next specification has to be tightened.

• The reading with its instrument and condition: 3.0 mm gap under the fourth leg, feeler gauge, table standing unloaded on the packing area floor
• The separation of floor from furniture, with its numbers: rotation test repeated in three directions, straightedge readings at 1 m and 2 m, and a reference table standing on the same floor
• The extent of the defect, stated as 9 of 32 sampled tables, with the individual readings listed rather than summarised in a sentence
• The root cause, stated as a leg length and a sanding setup at the factory, rather than as a workmanship failure on the assembly line
• The sampling decision with its figures: lot of 140 tables, ISO 2859-1 at general inspection level II, code letter G, sample size 32, major defects at AQL 2.5 with an accept number of 2 and a reject number of 3, and 9 found, which passes the reject number
• That no type approval testing under EN 12521, EN 15372 or EN 1730 had been carried out on these tables, and that the report was a specification check and not a compliance certificate
• That the floor the tables will stand on in the buyer's own premises was outside the scope of the inspection, and that the specification had been rewritten to deal with that rather than to pretend it away

How the specification should have been rewritten

Nothing in this order was exotic. The fix is entirely in the purchase specification, and it belongs in a purchase order the first time and in a standing specification thereafter. Each clause below was written so that an inspector can turn it into a pass or a fail without interpretation.

What to carry to the next order

Nothing on this order was mislabelled, nothing was substituted, and the factory did not cut a corner. It built a good table to a drawing that was internally consistent, and it checked every dimension it had been given. The failure was in one number that was taken at the wrong end of a part, and in one sentence that did not say what it meant.

Five habits would have caught it, and all five are cheap.

• Measure the end of the part that touches the customer. The leg was 448 mm at the shoulder and 445 mm at the foot, and the factory check was at the shoulder. A check is only as good as the end of the part it is taken at.
• Separate the floor from the furniture before writing a word. Three rotations and a straightedge take ninety seconds and decide whether the report is a non-conformance against the factory or advice for the customer.
• Record the floor you measured on, in the report. The packing area was flat and the restaurant will not be, and the answer to that sits in the specification rather than in the factory.
• Put a number in the order and schedule the table leg stability test before the container is booked rather than after the goods-in team finds it. These tables would have passed a generic 5 mm checklist line. They were caught because the buyer's purchase order was stricter than the default.
• Expect variability to need levelling, not remaking. Nine tables in thirty-two is a rate rather than a build defect, so the right instruction is to level and re-check the whole lot before loading, not to reject and sort it into good and bad tables.

What the inspection was worth

This was one day of an inspector's time at the factory: press, measure, rotate, measure the floor, separate the two causes, and write it up. It sits at the Full Verification level of the service, $299 per man-day, because the work included diagnosing the cause rather than only counting units and looking at them.

Against that sits a container that could have been loaded with nine rocking tables in every hundred, a retailer whose goods-in team would have found it, and a specification that would have kept doing the same thing on the next order. A finding that is on the report is worth arguing about. A finding that is not on the report is a table that rocks in a restaurant for the life of the contract.