Six hundred glass-top coffee tables went out to retail stores across the UK. Twenty-three came back with a six millimetre dip in the middle of the top. The glass was flat. The frame was not.
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.
Six hundred coffee tables, and a dip in the middle
A homeware retailer in the UK ordered six hundred glass-top coffee tables from a factory in Foshan. The top was 1,200 by 600 mm of 10 mm toughened safety glass on a black powder-coated steel frame, 420 mm high, with a single centre rail. The order was worth forty-four thousand seven hundred dollars. A deposit of thirty per cent, thirteen thousand four hundred and ten dollars, was paid, and the balance of thirty-one thousand two hundred and ninety was due against the loading date.
The first eight months of retail sales went normally. Then returns started arriving, and they all described the same thing. The middle of the top had sunk. The table wobbled. A table lamp that should sit flat sat in a shallow bowl, and on some units the lamp base rocked. Twenty-three returns came back out of two hundred and forty tables sold.
The retailer's first instinct was the glass. It looked like a glass problem, and glass problems are familiar. The top is the part everyone distrusts, and a replacement panel is a straightforward remedy. The retailer wrote to the factory asking for twenty-three replacement tops.
The factory replied that all twenty-three tops had passed the load test at the end of the line, that the panels were 10 mm as ordered, and that the retailer's own stores had installed the tables on a variety of floors. It offered to supply new glass. It did not offer to explain why a table that passed a load test two years ago now holds a bowl.
This is an illustrative example built from the pattern behind most long-term furniture complaints: the failure is time-dependent rather than visible at the point of manufacture, so the goods that were checked and the goods that fail are, on every measurable dimension, the same goods. The figures are representative of the category rather than of any particular order.
The question the retailer was not asking
The useful question was not which top had failed. It was whether the top had failed at all.
A replacement panel is only a remedy if the panel is the defective part. If the panel is intact and is sitting in a frame that has moved, then twenty-three new panels arrive, twenty-three tables still dish, and the retailer has paid twice: once for the glass and once for the freight. The original fault is still in the frame, and the remaining five hundred and seventy-seven tables are still carrying it.
So the check was booked as an Order plus Quality Check, two man-days at one hundred and ninety-nine dollars a day. One day at the factory to measure new units out of the same batch, one day at the retailer's warehouse to strip down the returned units and work out what had actually changed.
How the measurement was set up
Two numbers decide this case, and both of them are boring. How much the top drops under a stated load, and how much of that drop is left behind afterwards.
The first is elastic deflection. The glass bends, the load comes off, the glass comes back. The second is residual deflection. This is what is left behind, and it is the number that separates a table that is stiff enough from a table that is slowly changing shape.
Every unit was measured with a dial indicator on the centre line at mid-span, with a 1,200 mm straightedge and feeler gauges as the cross-check on the returned units. Each measurement was taken in the same four states. The glass alone on a flat bench. The glass on its frame, unloaded. The glass on its frame with a two hundred newton load applied at the centre of the top over a three hundred by three hundred millimetre pad and held for sixty seconds. And then the same point again, fifteen minutes after the load came off.
The frame was measured separately, in the same four states, with the glass removed and the unit standing on a flat plate on all four feet.
What the two sets of readings showed
Six new units were pulled out of the batch at the factory. Four returned units were taken apart at the retailer's warehouse. Nothing in the two columns below is an opinion, and every number is a reading.
| Measurement | New units at the factory | Returned units at the retailer |
|---|---|---|
| Glass top alone, flat bench, over 1,200 mm | 0.1 to 0.3 mm | 0.2 to 0.4 mm |
| Top on its frame, unloaded, at mid-span | 0.3 to 0.6 mm | 5.4 to 6.2 mm |
| Top on its frame, under 200 N at the centre | 0.9 to 1.4 mm | 6.8 to 7.1 mm |
| Same point, 15 minutes after the load came off | 0.3 to 0.6 mm | 5.4 to 6.2 mm |
| Frame alone, end frames out of parallel | 0.1 to 0.3 mm | 5.4 to 6.0 mm |
| Frame alone, feet in contact on a flat plate | All four in contact | Two in contact, two clear |
| Centre rail bracket at the end frame | Bolted, four M6, holes round | Bolted, holes elongated, one bolt loose |
The glass never sagged
The top line of those readings decides the case. On the returned units the glass measured flat: a maximum deviation of four tenths of a millimetre over 1,200 mm, with no fracture, no chip at the corners, and no dished profile of its own. Laid on a flat bench on six points, a toughened panel that had spent two years carrying a lamp in the middle of it came off the bench as flat as a new one.
The frame came off the same bench as a parallelogram. On the returned units the two end frames stood five point four to six millimetres out of parallel, and the centre rail sat correspondingly low. On the new units the same frame stood within three tenths of a millimetre of parallel, and all four feet touched the plate.
That is the whole finding in two numbers. The glass was not deformed and the frame was. The top on the returned table was not a deformed top. It was a flat top sitting in a warped frame, and the six millimetre dip was the frame's shape being read through the glass.
It also explains why the retailer could not reproduce the complaint in the store and could not photograph it convincingly. A unit standing on a floor that is itself out of level does not show six millimetres of dish. It shows whatever the floor adds to the frame, which on a level stockroom floor is nothing at all.
Why a stiff plate follows a frame that moves
Glass in a table is not a structural member in the way people assume. Its stiffness makes it an effective load distributor. It takes the lamp and the magazines and the stack of books and passes them to the frame as a spread load, instead of letting one point do the work. The frame carries the load and the glass carries the frame's shape. If the frame racks, the glass cannot help it.
The joint is where this table fails, and it fails because of how it is built rather than how it is loaded. The centre rail is joined to each end frame through a steel bracket and four M6 bolts. That is a pinned connection. Bolted through a bracket, the joint carries the vertical load and the horizontal load, and it carries almost no moment. Under a load in the middle of the top, the centre rail wants to sag, and the joint permits it by rotating.
Bolted joints also have clearance. Under repeated loading, seasonal temperature cycling, and the different expansion rates of a glass top and a steel frame, the bolts settle into their holes and the joint takes up its slack. Once it has taken up, the frame is no longer free to recentre itself. It is held in whatever shape it had reached, and each further load deepens it. That is a mechanism with a time constant, which is why the failure appeared over two years rather than in the first week of use.
Two things in the retailer's own records finish the mechanism. The returns cluster around tables where a lamp had been standing in the middle, which is exactly where a pinned joint has to work hardest. And the elongation was measurable at one end of the centre rail and absent at the other, which is what a joint that rotated under load leaves behind.
The frame on the delivery floor was not the frame on the drawing
The approved drawing for this table shows the centre rail welded between the two end frames. A welded joint across the full section is a moment connection. It resists rotation, and the frame holds its shape the way a welded frame is supposed to. The frames that shipped carry a bracket and four M6 bolts instead, and on all six new units checked at the factory that difference was visible at one joint in under a minute.
That gives the report a documentary finding to stand on, and it changes the nature of the claim. A table whose frame does not match the approved drawing is not a table with an opinion attached to it. It is a table that was not built to the drawing it was approved against.
It also explains why the goods passed every check that was performed on them. The frame is not visibly out of square when it leaves the line. It is straight, it is level, and it is correctly put together. What it is not is the frame on the drawing, and no dimension in the routine inspection report asked that question.
What the report could and could not establish
The report follows the eight sections we use for every order, and the two that matter here are the second, which lists what the inspection was measured against, and the eighth, which records what could not be verified. The second is where the drawing discrepancy belongs. The approved drawing is the standard, the delivered bracket is the departure, and the report says so in one line.
The eighth section has to carry the honest limits. The inspector could not establish how much of the joint rotation happened before the units left the factory and how much happened in two years of retail use, because returned units arrive with their service history and new units arrive without any. The inspector could not establish whether the frames left the factory already open by a fraction of a millimetre, which is why the finding is written against the joint design rather than against a single unit.
The inspector could not exclude an out-of-spec load either. If somebody stood on a table in a shop, the joint would rotate too, and no bench measurement can tell that case apart from the lamp. That possibility is recorded rather than dismissed, and it is the reason the recommendation is a design fix applied to the whole run rather than a repair applied to the twenty-three returns.
What the inspection established without qualification is simpler and more useful than any of that. The glass is sound, the frame is not, and the delivered frame is not the approved frame.
The report also put a number on the alternative. Repairing the twenty-three returns would have cost more than rebuilding the centre rail joint on the full run, and would have left five hundred and seventy-seven tables with the same pinned joint in the same condition. The recommendation was a joint detail change and a re-run of the remaining production, with the glass left alone.
The glass thickness was never the problem
Ten millimetre glass on a 1,200 by 600 millimetre top is zero point seven two square metres. The glass best-practice guide the UK and European retail trade works to sets out nominal thicknesses by area for glass that is not supported over its whole area: four millimetres up to zero point two five square metres, five up to zero point five, six up to zero point seven five, eight up to one point five, and ten above that. At zero point seven two square metres the minimum is six millimetres. The order asked for ten, and the order was given ten.
That matters for how the buyer argues the claim. The retailer cannot stand on the glass being too thin for the table, because it was not. What it can stand on is the support condition, because that same guide treats thickness as a function of how the glass is carried. A panel carried on a continuous edge strip and a panel carried on isolated pads are different cases with different limits. The delivered frames carry the top on a perimeter strip and on pads under the centre rail, with the full 1,200 mm between the end frames unsupported at the moment that matters.
The furniture standard behind the order says the same thing in a different register. A domestic table has to take a vertical static load of one thousand newtons and, where it carries glass, a vertical impact from one hundred and eighty millimetres, with the glass itself qualifying as safety glass under EN 12150-1 or as a Type B or Type C break under EN 12600. Those are the loads the table is designed for. A table that passes them and then dishes under a lamp in a living room has not failed a standard. It has failed the room it was sold into, and only the joint made that possible.
The specification, rewritten
None of the seven lines below needs a laboratory to check, and every one of them would have closed this case before a single panel was cut. The defect was not exotic and it was not hidden. It was a bracket instead of a weld, in a joint nobody had written down, on a frame that passed every dimensional check because it was straight and square when it left the line.
- State the support condition, not just the glass. Instead of 10 mm toughened glass to EN 12150-1 on the frame, write the glass down as carried on a continuous EPDM strip along each long edge with a centre rail at the mid-length, and give the effective unsupported span in millimetres.
- Give the deflection limit with its load, its method and its hold time. Maximum deflection of 2 mm is not a limit, because it does not say under what. Write: under a 200 N load applied at the centre of the top over a 300 by 300 mm pad and held for sixty seconds, vertical deflection at mid-span measured with a dial indicator shall not exceed 2.0 mm.
- Add the unloaded limit and the residual limit, and they are not the same number. The same point with no load applied shall not exceed 1.0 mm, and shall not exceed 1.0 mm again fifteen minutes after a loaded measurement. A frame that is taking a set passes the loaded check and fails the second one. The second one is what catches a joint that is rotating.
- Write the joint. Steel frame is not a specification. Name the connection: the centre rail welded to the end frames across the full section, or a bolted joint with a stated bolt size, a stated number of bolts and a stated bracket depth. Say which of the two is acceptable.
- Add a parallelism requirement for the bare frame, measured with the top removed. All four feet in contact with a flat plate, and the end frames within 1.0 mm of parallel. This takes two minutes on the line and it is the check that would have caught the fault before the first panel was cut.
- Tie thickness to span so that a later change to the frame cannot quietly change the glass. Toughened glass with a minimum thickness stated for the unsupported span, or thickness as recommended for the span and support condition, with the recommendation written down.
- Ask for one assembled first article, glass on, loaded and measured, before the rest of the batch is built. It costs a man-day and it would have closed this case in a day instead of two years.
Where to go next
The whole case cost two man-days
Two days of inspection, three hundred and ninety-eight dollars, and a finding that replaced a shipping lane of glass replacements with a joint detail. The retailer had asked for twenty-three new tops. The report showed that twenty-three new tops would have produced twenty-three tables that still dished, and would have left five hundred and seventy-seven tables with the same joint in the same condition.
The defect was not hidden and it was not exotic. It was a bracket instead of a weld, in a joint nobody had written down, on a frame that passed every dimensional check because it was straight and square when it left the line. What it took was a dial indicator, a flat bench, and the decision to measure the glass and the frame separately instead of taking the complaint at face value.
That is what a coffee table inspection is for, and it is why a load test on its own is not the same as a check. A load test asks whether the table survives a load. This case was about what the table did with the loads it survived, over two years, in a room with a lamp on it.
Where to go next
→ what a man-day covers and what it does not
→ how the sample size for a lot this size is worked out
→ send us the order details and we will tell you what a check can cover
Have an order you need verified? Send us your PI and factory address. Three levels, published prices, $149 per man-day from.