480 restaurant chairs passed the rocking test that the buyer's own order specified, and 294 of them had 4 mm of play at the backrest joint, which only showed up when somebody pushed the backrest sideways.
Illustrative example This case is written from the patterns we see in furniture export orders, and the figures are representative rather than from a specific client file. It shows how a check works and what it finds. We will replace it with a real, anonymised case as soon as we can publish one with the buyer's permission.
The order, and the test that everybody ran
A European restaurant group was refitting fourteen sites over one winter and placed a single order for 480 contract restaurant chairs with a factory in Foshan. Solid rubberwood frame, beech back frame, upholstered seat pad, built for daily commercial use and for flat-pack shipping. The unit price was $58, the order was $27,840, and 30 per cent, $8,352, was paid on order. The remaining $19,488 was held back against a pre-shipment report.
The joint requirement was in the order, and it was written in words the buyer had learned the hard way on an earlier order with a different factory: no perceptible play at any joint. What followed it was the method, and this is where the order went wrong. Test: rock the chair on a hard floor in all four directions and confirm no movement.
Both sides ran that test. The factory ran it on every chair at final inspection and recorded it on the traveller. The buyer's own goods-in team in Europe ran the same test on the pilot shipment, in the same words, because the order said to. Nobody skipped it and nobody read it carelessly. It is a good test of what it measures. It simply measures something other than what the requirement was about.
Then the pilot site reported a complaint that no rocking test would ever produce. The operations manager at the first restaurant wrote that the chairs knock when they are pushed back against the table, and asked for one to be sent back so the group could see it. Twelve of the eighty chairs in that room were showing it, she said, and she had stopped counting.
This is an illustrative example drawn from the recurring pattern in contract seating orders, where a joint requirement is written correctly and the method written next to it tests something else. The figures are representative of the category rather than taken from a named customer file.
Where to go next
→ our dining and chair verification scope, including joint and hardware checks
What a rocking test actually loads
A chair rocked on a hard floor is loaded in two ways. Rocking it diagonally puts the legs into bending and the whole frame into torsion about the floor contact, so the joint that decides the result is the leg-to-seat joint, where the frame has to resist the entire chair twisting over one corner. Pressing down on a corner of the seat loads the same joints directly. Both of those are vertical and fore-and-aft loads, and they are precisely the loads a standard chair stability test is designed to apply.
Now take the backrest joint on the same chair. The backrest frame is bolted to the rear legs through a steel bracket, and the bracket is located by a hardened dowel pressed into a reamed hole in the leg. The dowel takes the side load. The bolt takes the pull-out load and squeezes the bracket against the leg so that the two faces bear on each other. Both of those are sound ways to build a rigid joint, and both of them are satisfied by the same hardware.
A rocking test does not ask that joint to move sideways. It presses down on it, or twists the frame over it, and in both cases the bracket is pulled hard against the leg and the dowel, if it is present, is carrying the side load in compression. The joint is being asked to do the one thing it is best at. A chair can therefore be immovable in every direction that a rocking test loads and still have four millimetres of play in the direction it does not load, because the only thing standing in the way in that direction is the clearance in a hole.
That is not a defect in the rocking test. It is a defect in using it as a joint test, which is what an order that specifies it as the method has done. The requirement and the method were both present, both reasonable on their face, and together they produced a chair that nobody had checked.
Where to go next
Where the four millimetres came from
The brackets are slotted. Each one has a 10 mm lateral slot so that the backrest can be shifted left or right during assembly and lined up with the seat, and a 4 mm hardened dowel is meant to fill that slot and turn the bracket into a locating feature rather than a clamping one. The dowel is pressed into a reamed hole in the rear leg before the bracket is bolted on.
The dowels were not supplied for part of the run. The line had consumed the last of them early in the week, the replenishment had not arrived, and assembly went on with the brackets bolted through empty slots. Nothing in the assembly stops this. The bolt goes in, the bracket sits flat against the leg, and the chair looks and feels correct to everyone standing at the bench.
A bolt is not a locating feature. An M6 shank is 6 mm and the slot is 10 mm, so the joint has 4 mm of float in the slot and nothing at all resisting it in that direction beyond the friction of a bolt that was never asked to do that work. What it does instead on those chairs is drive the bracket face into the leg face at the end of the travel, which is the knock the pilot site reported.
Here is the whole defect in one line of arithmetic: an M6 shank is 6 mm, the slot is 10 mm, so the joint moves 4 mm one way and 0 mm the other, and the joint has no resistance in that direction at all until the bolt finally takes up.
It is worth noticing what did not happen. The brackets were the right part. The bolts were the right length and were torqued correctly. The dowel hole was reamed in every leg, because the reaming operation ran on a separate line that never stopped. The bracket sat flat, the leg was square, and the chair passed every test the order asked for.
Where to go next
→ our dining and chair verification scope, including hardware and joint checks
How the play is measured
The test is not difficult, but it has to be written down, because the measurement is the only part of it that anybody argues about.
Put the chair on a level hard floor with nothing behind it. Hook a 100 N spring scale to the top rail of the backrest, 700 mm above the seat, and apply the load sideways, in the width of the chair rather than fore and aft. Read the movement at the joint line with a feeler gauge on the bracket-to-leg face, and read the movement at the top rail with a dial indicator. Do both, in each direction.
The reading at the joint line is the one that goes in the report, and the reading at the top rail is the one that is easiest to misread. The backrest is a cantilever. A 100 N load at the top rail puts a moment on the joint that bends the backrest post, and the post's own bending shows up in the top-rail reading without being clearance at all. On the affected chairs the joint moved 4.0 mm and the top of the backrest moved 6.8 mm on average. The 2.8 mm difference is the post bending under the same load. A report that quotes only the top-rail figure overstates the joint by 70 per cent.
Direction matters for a second reason. Push the backrest one way and a rigid joint reads zero both times. Push it the other way and a clearance joint reads zero one way and 4 mm the other, because the bolt bears against the far side of the slot and takes up the slack. A joint that reads 4 mm in both directions is a different fault entirely and needs a different remedy.
Read it again ten seconds after the load is released. Clearance returns, and a residual above about 0.5 mm means something has taken a set or a fastener is slipping, which is not the same problem and must not be reported as the same problem.
And if the measurement shows play, the next question is where it is, because that decides whether the remedy is a part, a hole or a torque. Hold the bracket and push the backrest. If the bracket face slides across the leg face, the joint has clearance and the locator is missing. If the bracket itself moves and the bolt head turns, the fastener is not clamped. If the leg moves under the bracket, the fault is in the frame below the joint and the bracket is innocent.
| Measurement | Compliant chair | Affected chair |
|---|---|---|
| Rocking test, four directions | No movement | No movement, all 50 sampled chairs passed |
| Gap at the bracket-to-leg face, backrest pushed one way | 0.0 mm | 0.0 mm |
| Gap at the bracket-to-leg face, backrest pushed the other way | 0.0 mm | 1.5 to 4.0 mm, mean 3.4 mm |
| Lateral travel at the top rail, 100 N at 700 mm | 0.4 to 0.9 mm | 6.0 to 8.4 mm |
| Residual travel 10 seconds after release | Under 0.2 mm | Under 0.2 mm |
| Bracket-to-leg contact | Flat, full bearing contact | 4 mm gap on the loaded side, no contact |
Where to go next
→ what a report records when a lot passes and a lot fails
What the report recorded
The report recorded both tests, and it recorded them in that order, because the rocking test result is the one the factory and the buyer's own team had both relied on. Fifty chairs, all four directions, no movement on any of them. That result is not in dispute and it is not wrong.
Then the lateral push test: 31 of the 50 sampled chairs showed measurable play at the backrest joint, between 1.5 and 4.0 mm, with 3.4 mm the mean. The lot was rejected on that characteristic. Under ISO 2859-1 at general inspection level II a lot of 480 is code letter H with a sample of 50, and at AQL 2.5 the lot accepts three findings and rejects four. Thirty-one is not a marginal result. The sampling arithmetic was never in doubt here, because a lot with 61 per cent of its chairs affected is not a lot that a sample of fifty was ever going to pass.
A restaurant chair inspection china buyer who reads only the rocking test result learns nothing about the 294 chairs, and a restaurant chair inspection china file that carries both readings and the boundary between them is the one worth holding. The second test cost the same shift as the first. The difference is that one of them answers the question the order actually asked.
The report also recorded what it was not asked to certify. It did not say the chairs were unsafe, because a chair with four millimetres of play at the backrest is a chair that knocks and moves, not a chair that fails a load requirement. It did not say the frame was wrong, because the frame was right: the legs were square, the dowel holes were reamed, the brackets were the correct part and correctly torqued. It did not claim that all 480 chairs were affected, because 31 findings in a sample of 50 is a rate and not a count, and the only honest way to count them is to test every chair.
And it recorded the pattern, which was the most useful thing in it. The affected chairs were not scattered through the lot. They fell in a band of the run. Opening cartons from the beginning, the middle and the end of production and testing chairs from each showed the band, and the band started on the day the dowels ran out. The report did not state the cause, because the cause was the factory's to establish. It stated the location, the extent and the boundary.
What the factory found, and what 100 per cent cost
The factory's own investigation was quick, because the missing dowels were a parts-supply failure and the line kept a delivery record. The dowels had run out on the fourth day of a nine-day run, a replenishment had been signed for and had not arrived, and the line continued for two days. The chairs assembled in that window, and only those chairs, went out without a locator.
The factory then did the thing that should have happened before the report was written, which is to screen every chair in the order rather than to argue about the sample. A push test against a go/no-go gauge takes a few seconds a chair, and 480 chairs took one shift. Two hundred and ninety-four of them failed it, 61 per cent of the order, and the number matched the sample almost exactly.
This is where the value of the inspection stopped being the sample and became the boundary. A report that said only 31 of 50 failed would have left the buyer with a rejected lot and no idea whether the problem was a workmanship trend, a supplier failure or a single bad afternoon. The band turned it into a parts-supply event with a start and an end, and it told the factory exactly which chairs to rework and exactly which 186 chairs were fine.
It also told the buyer something worth knowing beyond this order. A factory that can lose a locating part for two days without a line stopping, without a check catching it and without the traveller recording it has a change-control gap that will cost it on the next order too, whatever that order happens to be.
Where to go next
What it cost, and what it was worth
The rework was straightforward: bracket off, dowel pressed into the reamed hole, bracket back on, bolts torqued, reading re-checked. The factory quoted $5.60 a chair on the 294 affected chairs, which is $1,646, and the 186 chairs outside the band were not touched. A re-check of 50 chairs drawn from the first and last of the rework found no play above 0.5 mm, and the inspector recorded it as a separate one man-day at $199.
Two inspections, $398 and $199, $597 in total against an order of $27,840. That is 2.1 per cent of the order value and 3.1 per cent of the balance being held. The rework it produced cost $1,646, and it happened while the chairs were still in the factory, where a press and a bench were a few metres away.
The alternative is worth setting against that, because it is the number the buyer was actually exposed to. The pilot site had 80 chairs installed and 41 of them knocked. Every additional site would have taken more chairs with the same fault and the same complaint, and the buyer's own goods-in check in its warehouses uses the rocking test, because the order told it to. Those chairs would have passed the buyer's own incoming inspection and reached the restaurant floor.
The fit-out programme for fourteen sites did not have room for that. The winter refit carried a fixed opening date per site, and 61 per cent of the chairs arriving with a joint nobody had tested was not a problem the buyer could absorb at the end of it.
Where to go next
→ what a man-day covers and what it costs
The lines that would have caught it
The order already had a joint requirement, so this is not about adding words to a document that had none. It is about the four lines next to it. Each of them costs nothing to write and each of them removes a different way for a chair to pass a check that was never aimed at it.
The rocking test is not a bad check and it should stay in the order, because it finds the leg-to-seat looseness that a chair really does fail on. It simply should not be named as the method for a joint requirement, because the moment it is, a requirement and a method that disagree stop looking like a disagreement and start looking like compliance.
| Line in the order | As it was written | As it should read |
|---|---|---|
| Joint rigidity | No perceptible play at any joint | No perceptible play at any joint. Backrest-to-leg joint: maximum 0.5 mm lateral movement at the bracket-to-leg interface |
| Test method | Rock the chair on a hard floor in all four directions and confirm no movement | Add: with a 100 N load applied sideways to the top rail at 700 mm above the seat, the backrest-to-leg joint shall show no perceptible play, measured at the joint line with a feeler gauge and at the top rail with a dial indicator. The rocking test is a frame stability check and does not satisfy this clause |
| Acceptance and evidence | not stated | Record the reading in millimetres, not a tick. Name the sample size and the positions in the run the chairs were drawn from |
| Parts and change control | not stated | A change in the parts that locate or clamp a joint, including a substituted locating dowel, requires written approval before assembly continues |
Where to go next
Six lines to carry into the next order
None of this needs a laboratory, a loading rig or a specialist. It needs a force gauge, a dial indicator and a decision to write down the number that somebody will later read your goods against.
- Say no perceptible play as a number in millimetres, at a named point. Zero is not a measurement anybody can repeat.
- Name the load, the direction, the height and the measuring point. A load with no direction is a load with two answers.
- Keep the rocking test, and rename it what it is. A frame stability check is not a joint test, and a chair that passes one has not passed the other.
- Ask for the reading rather than the tick. Chair 1 to 50 joint readings recorded, maximum 0.4 mm, is evidence. All joints checked OK is not.
- Ask for the first chairs of every shift, not only the first chairs of the order. A parts-supply failure does not begin at the beginning of an order.
- Name the parts that locate a joint and put a substitution clause on them. A locating dowel is not a consumable, and its absence is not visible from the outside.
Where to go next
→ our dining and chair verification scope
→ what to put into a chair specification before you order
What the report could not settle
Three things stayed outside what the inspection could establish. The first is what happened at the pilot site before the chairs came out. The forty-one chairs that knocked were already installed and in daily service, and the report can tell the buyer what was wrong with them but not what to do with a restaurant that has been running on them.
The second is the cost of the chairs already installed. The rework happened at the factory, not at the restaurant, so somebody was going to carry that cost and the report was not the place to decide who.
The third is whether the same line made other things for the same buyer with the same dowel. The report is an order verification statement against the buyer's own documents for this order, not an audit of the factory's part supply, and it says so on its face. A second order at the same factory, or a different chair in the same catalogue, sits outside what fifty chairs and two man-days can settle.
What the report did settle is narrower and more useful. It recorded that 480 restaurant chairs on a $27,840 order carried a joint requirement that the rocking test could not reach, and that the requirement had been checked and passed on every chair, and that both the check and the requirement came from the buyer's own order. It recorded 294 chairs with four millimetres of play, a boundary around them in the production run, a cause the factory could act on, and a balance that could be released against evidence rather than against a nodding inspector.
Have an order you need verified? Send us your PI and factory address. Three levels, published prices, $149 per man-day from.