The tops were bowed 5 mm across 1,800 mm, and every one of them passed the factory's flatness check
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 situation
A European contract furniture supplier ordered 150 solid-wood dining tables from a factory in Foshan for a hotel restaurant fit-out. Each table was 1,800 × 900 × 750 mm with a 25 mm solid ash top on a four-leg frame. The order was worth $48,000, and 70% of it — $33,600 — was still unpaid when the tables were finished and ready to load.
The purchase order carried one line about the tops: *tabletop surface to be flat, deviation not more than 1.5 mm.* There was no other dimensional requirement anywhere in the document for the top. It was a normal line, written the way thousands of furniture purchase orders are written, and it was accepted without comment by both parties.
The factory ran its standard final check on every top: a 900 mm aluminium straightedge laid across the width of the panel at three points, with a feeler gauge run under it. Readings came out between 0.4 mm and 0.9 mm. Every top was recorded as conforming. All 150 were packed, photographed and loaded into three containers.
The buyer booked an inspection before releasing the balance. The inspector found the tops flat on the face and 5 mm out of shape along the length.
This is an illustrative example drawn from a pattern we see regularly on solid-wood table tops. The figures are representative of the category rather than taken from a named customer file.
"Flat" covers three different things, and only one of them was checked
"Flat" is used in furniture specifications to mean at least three separate properties, and a panel can be good at one of them and fail the others badly. The factory's check and the buyer's requirement were not talking about the same property, and the 1.5 mm figure was never attached to any of them.
| Property | The question it answers | How it is measured | Where it goes wrong |
|---|---|---|---|
| Surface flatness | Is the face of the panel a plane? | 900 mm straightedge laid across the face, feeler gauge run under it | Treated as if it settles the whole top |
| Bow, or camber | Is the length of the panel straight? | A straightedge longer than the span, the panel supported at two stated points, the deviation read at midspan | Measured on the bench, where a panel lying along its whole length cannot bow at all |
| Planarity | Are the four corners of the panel in one plane? | The two diagonals of the assembled top compared against each other | Never checked at all on a top that is one solid panel |
The three failures behave differently in service
The distinction is not academic on a solid-wood top, because the three failures behave differently in service. A rough face is a finishing problem and a fairing block fixes it. A bow changes the geometry of the table, and no amount of sanding takes it out. A twist — the panel's corners out of plane — is what makes a table stand on three legs and rock on the fourth, and it is invisible to any check that only looks at a surface.
**A bow is also a shape measurement, and a shape only exists relative to the points you hold the panel at.** Hold a 1,800 mm panel at its two end edges and it has 1,800 mm of length to be straight over. Hold it at the 1,200 mm between the table legs and it has 1,200 mm to be straight over, and that is a different question. The factory answered the third question, the buyer asked the first, and the table in the restaurant was living with the second.
What the straightedge actually found
The inspector did not begin with a gauge. The first thing checked was the panel's own construction, because a solid-wood top glued up from 90 mm boards carries a history: the straightness of each board along its length, the board's moisture content at the moment of glue-up, and the springback of the press. All three are visible in a panel before it is ever framed.
The boards were 90 mm wide, kiln-dried, and cut from a single lot. The offcuts from that lot showed 3 to 4 mm of lengthwise bow over 1,800 mm on two boards out of six — enough, when the panels are laid up edge to edge, to put a few millimetres of curve into the finished top without any single board being wrong.
The factory's final check could not see any of this. A 900 mm straightedge laid across the width of the panel measures the flatness of one 900 mm slice of the face. It says the surface is smooth. It says nothing about where the panel's ends are, and it says nothing at all about the diagonal.
| Check | Straightedge | Support condition | Reading | What it establishes |
|---|---|---|---|---|
| Factory final check, all 150 tops | 900 mm, across the width | Panel lying flat on the bench | 0.4 – 0.9 mm | The face is flat. Nothing about the shape of the panel |
| Inspector, loose panel | 2,000 mm, along the length | Crown down, supported at the two end edges | 3.4 – 5.0 mm at midspan | The panel's own bow |
| Inspector, assembled table | 2,000 mm, along the length | Legs on a level floor, 1,200 mm leg span | 3.9 – 5.0 mm | The bow in the condition the table is used in |
| Frame squareness | Tape, both diagonals | Assembled | Diagonals differ by 6 mm | The frame was pulled out of square by the panel it was fixed to |
| Table stance | Feeler gauge under each foot | Assembled, on a level floor | 3.2 mm clear under one foot | The table rocks and will not sit still |
What the number was worth, and what it cost to get it
The 1.5 mm in the purchase order was applied by the factory to a 900 mm face check because that was the check the factory had a gauge for. Nobody chose that. **It is what happens when a specification states a number without a method: whoever owns the equipment defines the method.**
Sampling was ISO 2859-1, general level II, which gives a sample of 8 from a lot of 150. All 8 exceeded the 1.5 mm figure in the purchase order, 6 of them by more than 4 mm. Because the deviation was in the panel rather than in the assembly, the inspector recommended screening all 150 rather than widening the sample, and the buyer held the balance for the eleven days that took.
The support points decide the number
A table flatness check is a three-point measurement. The straightedge rests on two points and the deviation is read at a third, and which two points you hold the panel at is not a detail of the method — it is the measurement.
For a table top there are three support conditions available, and all three were used on this order.
- **The whole length, 1,800 mm.** The panel supported at its two end edges. This is the panel's own geometry, and it gave 3.4 to 5.0 mm at midspan. Measured crown down, so the straightedge lay across the hollow and the gauge could be run under it.
- **The service span, 1,200 mm.** The legs are inset 300 mm from each end, so in service the panel is carried at 300 mm and 1,500 mm. The bow a customer feels is the bow across that span, and it gave 3.9 to 5.0 mm. It is worse than the full-length figure because the same curve is being asked to be straight over a shorter distance.
- **Lying flat on the bench.** The panel supported along its whole length. It gave 0.4 to 0.9 mm, and it is the only one of the three that is meaningless as a shape measurement, because a panel supported everywhere cannot bow anywhere. It is also, not coincidentally, the check the factory ran on all 150 tops.
Two things follow, and both of them belong in a purchase order. The first is that **a bow figure without its support condition is not a checkable number.** 3.9 mm and 5.0 mm are the same panels measured over two different spans, and a report that quotes a bow without saying how it was supported has said nothing that anybody can repeat next week. The second is that the reading changes with the assembly, so the loose panel and the assembled table were measured separately and the assembled reading was the one reported, because it is the condition the goods ship in.
Where the honest limit sits
**What the inspection established:** that 150 tops had a face flatness inside 0.8 mm, a bow of 3.4 to 5.0 mm over the panel's own length, and a 6 mm difference between the frame diagonals; that the assembled tables stood with up to 3.2 mm of clearance under one foot; that the factory's 1.5 mm check could not and did not detect any of it; and that the 1.5 mm figure in the purchase order did not state the method it applied to.
**What the inspection did not establish: the cause.** The board lay-up explains how a bow of that size can exist, and the inspector could see lengthwise bow in the offcuts of the same lot, but attributing a specific panel's shape requires the factory's own records: which lot, what moisture content at glue-up, how long the panel was left before framing, and which way round the boards went. The factory's log showed the panels were measured and framed on the same day, which narrows it further without closing it. The report recorded the measurement and left the attribution to the process data, because a report that says *the factory glued the boards up wrong* without that data is making a claim the inspector cannot support.
**The frame was checked separately, and it cleared.** On two units the inspector loosened the top fixings and re-measured. The panel's own bow was unchanged at 4.8 mm and 5.0 mm, and the frame's diagonals still differed by 6 mm. So the frame was not the cause of the bow — and the frame was also not innocent in the rocking, because a panel fixed to a square frame and screwed down pulls that frame out of square. One finding, two consequences.
**What the inspection could not establish at all: how the tops will behave in the customer's building.** Everything was measured in Guangdong, in the factory, in the humidity of that week. A 25 mm solid ash panel moves with its moisture content, and the equilibrium moisture content of a heated European interior in January is around 9%, while a factory in a humid Guangdong month sits well above that. **A bow figure taken in Guangdong is not a bow figure in Rotterdam.** The panel was not at its equilibrium moisture content when it was measured, and the report could not say what shape it will be on the customer's floor. That is not a reason to dismiss the finding — a table that measures 5 mm out of shape in the factory does not become straight on the way to Europe — but it is a reason to write the moisture content and a delivery re-check into the order.
For the record, the remedy cost less than the argument would have. The factory re-cut and re-glued the 150 panels from the same lot with the board orientation recorded, left them nine days in the assembly hall before framing, and the buyer held the balance for the eleven days that took. The 1.5 mm line in the purchase order was never changed.
What the specification should have said
Six lines. None of them costs anything to write, and each one turns an argument into a measurement.
- **State the method with the number.** *Bow shall be measured as the maximum deviation between the upper face of the panel and a 2,000 mm straightedge laid across the panel with the panel supported at its two end edges, the deviation read at midspan with a feeler gauge reading to 0.01 mm.* A number with no method attached is not a specification, and whoever owns the gauge supplies the method by default.
- **Give a number for each span that matters.** *Bow over the full panel length of 1,800 mm shall not exceed 2.0 mm, and bow over the 1,200 mm service span between the leg frames shall not exceed 1.0 mm, both measured as described above.* One figure cannot cover a panel checked across its width and a panel checked between its legs.
- **Specify face flatness and planarity separately.** *The upper face shall be flat within 0.8 mm under a 900 mm straightedge applied in both directions, and the diagonals of the assembled top shall agree within 3 mm.* A twisted panel is a different failure from a rough one and it needs its own line.
- **Set the moisture content, and say where it is measured.** *Panels shall be conditioned to between 8% and 12% moisture content, and the moisture content of each panel recorded at the point of assembly using a resistance-type pin meter at the panel centre. The buyer reserves the right to require a bow re-check at delivery against the 1,200 mm span limit.* This is the clause that makes the finding hold in the destination country rather than only in the factory.
- **Require the panel construction that prevents the defect, and the records that prove it.** *Boards shall be checked for straightness before assembly, the board lay-up shall be recorded, and the panel maker shall confirm in writing that each finished panel was measured for bow over the 1,800 mm length before the frame was fitted.* Checking a panel before it is framed costs one gauge and one minute; it is simply not in the order, so it does not happen.
- **Make the panel check and the stance check a gate on production.** *Before frame assembly begins, three finished panels shall be presented for measurement on the assembly bench and assembly shall not start until the result is recorded and accepted in writing. Assembled tables shall stand on all four feet on a level floor with no foot more than 1.0 mm clear.* The first-article check is the cheapest control on the list, and it is the one that would have stopped 150 panels.
The point of all six is not to catch the factory out. It is to make the state of the goods checkable by somebody who is not there, on a day that is not the day the panel was pressed, by a method both sides agreed to before the panel was cut.
Where to go next
→ dining table and chair verification in our dining scope
→ table and desk verification for commercial and office tables
→ what a report records when a lot passes and a lot fails
→ how sampling works under ISO 2859-1
Have an order you need verified? Send us your PI and factory address. Three levels, published prices, $149 per man-day from.