The joinery arrives on a Tuesday. Four reception units, built to the dimensions on the approved drawing. The site team offers them up to the wall and the last one is 15 mm too wide.

Somebody will suggest the shop measured wrong. Usually nobody measured wrong. The drawing said 2 400 and the wall says 2 385, and both of those numbers are honest.

Getting comfortable with that gap is one of the things that separates a second-year engineer from a first-year one. Buildings do not come out the size they were drawn. The job is keeping the difference small enough that nothing has to be rebuilt.

Where the difference comes from?

No single villain here. It accumulates.

Setting out carries a small error from the start, because instruments have tolerances too. Concrete shrinks as it cures, and it does not shrink evenly. Blockwork picks up a millimetre or two per course, and by the time you have twenty courses those millimetres have found each other. Steel moves with temperature, which matters more than people expect on long spans in hot weather. Then finishes go on, and plaster has a thickness that varies with the plasterer’s day.

Add a wall built slightly out of plumb, which is entirely normal within tolerance, and the opening at the top is not the opening at the bottom.

The drawing does not model any of this. It shows the geometry as designed, and that is exactly what it should show. cad drawings are prepared to define intent and dimensions clearly, not to predict how a particular crew will pour a particular slab in a particular week.

So the question on site is never whether there is a difference. It is how big the difference is and whether anyone has measured it yet.

What a tolerance actually is ?

A tolerance is permission to be slightly wrong, agreed in advance.

Two kinds matter and people mix them up. Construction tolerance covers what the structure and finishes are allowed to deviate by. Fabrication tolerance covers what a manufactured element is allowed to deviate by. They are different numbers, set by different parties, and the joint between them has to absorb both.

That is why joints exist where they do. A 10 mm shadow gap around a panel is not only a design decision. It is the space where the difference between the wall and the panel goes to live. Remove the gap for aesthetic reasons and you have removed the tolerance, and now somebody is scribing panels on site with a jigsaw.

Practical version: if a detail has no gap, no shim space and no scribe, ask who is absorbing the deviation. If nobody can answer, the answer is the installer, at their own cost, badly.

Why fabricators insist on measuring the opening ?

Anything made off site to fit a specific space gets measured on site first. Joinery, cabinetry, doors, glazed screens, cladding panels, stone. The fabricator sends someone out with a laser and a clipboard, and the numbers they bring back override the drawing.

This is what shop drawings are for. They take the designed element and document it at the size it is actually going to be built, with the site conditions folded in.

The economics explain the insistence. A reception desk built to a drawing dimension and delivered 15 mm oversize is not adjustable on site. It goes back to the shop, gets remade, and arrives three weeks later. A survey visit costs a morning.

Worth knowing as a site engineer: when the fabricator asks whether the walls are finished before they measure, they are asking a real question. Measuring to bare blockwork and then applying 15 mm of plaster produces exactly the problem this whole visit was meant to prevent.

How dimensional control is done on site ?

The principle is that every measurement traces back to something fixed.

A project has control points established by survey, usually referenced to a site datum. From those come the grid lines, and from the grid come the offsets used to set out individual elements. Level control works the same way, running from a benchmark rather than from the floor you are standing on, because that floor has its own deviation.

Instruments in normal use: total station for setting out, laser level for transferring levels, plumb laser for verticality on multi-storey work, plus a tape and a spirit level for the hundred small checks a day nobody records.

The habit that matters is checking against the control, not against the previous element. Measuring each partition from the one before it is how a 3 mm error becomes a 40 mm error by the end of the corridor.

Where error accumulates fastest ?

• Multi-storey vertical alignment, where each floor inherits the last one
• Long runs of repeated elements: partitions, columns, façade bays
• Interfaces between two contractors, where each set out from a different reference
• Sequences with several finish layers, since each has its own thickness variation
• Anything set out from a wall rather than from a grid line
• Retrofit work, where the existing building was never square to begin with

Retrofit deserves a note. In an old building, nothing is plumb, nothing is level and no room is rectangular. Everything gets measured. Design dimensions there are a starting hypothesis rather than an instruction.

A checklist before you order material

Before anything is fabricated or delivered:

• Confirm the opening or space has been measured on site, not scaled off a drawing
• Confirm whether the measurement was taken to finished surfaces or to structure
• Record who measured, when, and with what
• Check the finishes schedule for layers not yet applied
• Identify which joint absorbs the deviation, and how much it can absorb
• Confirm the fabrication tolerance the supplier works to
• Note any element that cannot be adjusted on site at all
• Get any dimensional change back into the drawings before the next trade uses them

That last one gets skipped constantly. A dimension that changed on site and never reached the drawing will surface again three months later, when the next subcontractor works from the old number.

The gap is the job

A drawing is a statement of intent, produced before the building exists. The building is a physical object made by people working in weather. Those two things will never match exactly, and a project that expects them to is a project full of surprises.

Measure early, write it down, tell the people whose work depends on the number. Most dimensional disasters on site are not measurement failures. They are communication failures wearing a measurement costume.