Modelling a product from scratch is the easy half of this trade. Modelling something that has to land inside a space built by somebody else, years ago, to nobody's drawing, is the half where money is lost. I do not measure those spaces myself: the survey always arrives from the client, with the enquiry, in whatever form they happen to keep it. So my work starts where somebody else's tape measure stopped, and most of the expensive surprises on a shopfront, a fit-out or a steel structure trace back to a survey that looked complete on screen.
A survey I did not take is evidence, not truth
What arrives varies. A dimensioned sketch on squared paper, a photo of a sketch on the back of a job sheet, a landlord's PDF with dimensions nobody has checked against the building. All of it is useful. None of it is automatically a dimension I am prepared to cut aluminium or steel against.
The distinction I draw on every job is between reference dimensions and cut dimensions. A reference dimension tells me where things roughly are: the column is about there, the shutter box runs that way, there is around that much headroom above the door head. It is enough to design around, to check clearances, to decide a layout. A cut dimension is a number a part gets made to, and it has to have been measured directly, at the right place, by somebody who knew a part would be cut to it.
If a mesh turns up in the pack, that is exactly how I treat it: a photograph with depth. It shows what is there. It does not tell you how big it is, and it looks equally convincing whether it is right or wrong.
The dimensions I ask for by name
When a survey is thin, I do not guess and I do not scale a PDF. I send back a short list and ask for those numbers specifically. On a shopfront it fits on one sheet:
- the structural opening width, taken at three heights: head, middle and cill, never once in the middle;
- floor to underside of the soffit, taken at both ends and at the centre;
- both diagonals of the opening;
- the projection available before the lease line or the walkway edge, which is a legal limit as much as a physical one;
- what the thing is fixing into and how deep it is: lintel, steel post, blockwork, or plasterboard with nothing behind it.
That last one is not a dimension at all, and it is the one most often missing. A beautifully measured opening is no help if the fixing detail assumes a lintel and the wall turns out to be a stud partition.
Asking costs one message. Finding out on installation day costs a return visit, a re-cut panel and a client who now watches every drawing.
Out of square is the normal case
New-build drawings show rectangles. Buildings do not supply them. An opening 30 mm wider at the cill than at the head, a soffit that falls 15 mm across its length, a reveal that bows out of plumb in the middle: this is ordinary rather than defective, and a survey has to record it as found instead of tidying it into a rectangle on the way into CAD.
Both diagonals are the cheapest check there is. If they differ, the opening is a parallelogram and you now know by how much. Model it that way. A model that quietly averages three width readings into one number has thrown away the most useful thing the survey collected, and the difference does not disappear, it just moves to installation day, where it costs more.
The same applies to floor level. On a run of joinery or a plinth-mounted totem, a floor that falls 20 mm over four metres decides how tall the scribe has to be and whether the top line still reads level to somebody standing in the doorway.
Where the site error was allowed to land on the LEWLEX stairs
The external steel stairs I documented for LEWLEX are a good example, because steel gives you no second chances. Two flights, two landings, C160 stringers, WEMA grating on the steps, the whole structure hot-dip galvanised and then bolted together on site. After galvanising, nothing gets trimmed, filed or persuaded. Whatever the drawings said is what arrives on a lorry.
That structure lands on concrete pad footings poured by somebody else, on ground nobody surveyed for me. So the adjustment does not sit in the steel, it sits underneath it: every leg stands on a base plate with chemical anchors, and the plate is set to level on a bed of self-levelling mortar or on steel shims. The ground does not have to be right. The plates do, and they are the last thing set before the frame goes up. The contact zone between steel and concrete then gets sealed after installation, because that is where external stairs start to rot.
The lesson generalises past stairs: decide early which element is allowed to absorb the site, make it cheap and make it late. The nine-sheet package is in the portfolio as steel external stairs, and that kind of frame-and-baseplate thinking is most of what structural design actually is.
Design so the error has somewhere to go
Once the real geometry is in the model, the job is not to design to it exactly. It is to design so the remaining uncertainty is absorbed where nobody sees it. The moves are old and they still work:
- Scribes. A separate, oversized piece that meets the building and gets cut on site. Never make the visible panel do this job.
- Shadow gaps. A deliberate 10 to 20 mm recess hides a few millimetres of taper that a tight butt joint would advertise from across the street.
- Slotted fixings. Round holes commit you to the survey. Slots let the installer take up the difference without drilling on site through a panel that is already painted.
- Oversize backplates and trims. Let the thing behind be generous and the thing in front be exact.
- Two-part details. A carrier fixed to the building, a face fixed to the carrier, adjustment between them.
And spread the adjustment rather than banking it in one place. Ten millimetres shared across four joints is invisible. The same ten millimetres in a single joint at eye level is the thing the client photographs.
What has to travel with the model
A survey that exists only as a mesh and a memory is not a deliverable. What I want in the pack, every time:
- A dimensioned survey sheet with each critical dimension named, so the model and the site talk about the same number.
- Photographs keyed to positions on the plan, not a folder of forty unlabelled shots.
- A stated datum. One point everything is measured from, marked on site and on the drawing.
- An assumptions list, written plainly: what was not accessible, what was measured through a finish, what is believed rather than known.
- A note of who took the measurements and when, because a survey done before the shell was stripped out is a different document from one done after.
Getting the survey, the model and the production pack to agree is most of what turns a set of somebody else's measurements into files that can safely go to the floor, which is the same discipline as production support.
What to do with this
- Split every number into reference and cut. Only cut dimensions come off a tape, at the right place, on purpose.
- Ask for the missing dimensions by name before modelling. One message beats one return visit.
- Record the building as found, out of square included. Do not let CAD idealise it for you.
- Decide which element absorbs the site, and make it the cheap, late, invisible one.
- Send the datum, the assumptions and the survey sheet along with the model. If the survey cannot be argued with, the installation usually cannot either.
