Site visits are the least-taught part of architectural practice and one of the most consequential. Most of us learn them by accompanying someone who was never taught either, which is why the standard version has settled into a ritual: arrive, walk the works, attend the meeting, agree the minutes, leave.

That version is not useless, but it is mostly reassurance. The visits that matter are the ones where something is caught while it is still cheap, and that requires a different approach to the same two hours.

Arrive early and walk it alone

The most valuable part of a site visit is the fifteen minutes before the meeting, walking without a client, a contractor, or a purpose.

Accompanied walks are guided walks. You are shown what somebody has decided you should see, in an order that makes sense to them, with commentary that frames each item before you have formed a view. This is not deception; it is what anyone does when showing a visitor around. But it means you see the works as narrated rather than as they are.

Alone, you notice different things. The junction that has been detailed differently from the drawing without anyone thinking to raise it. The stack of materials that is not what was specified. The area nobody is working in, which is usually where a problem is being avoided. The head height in a corridor that feels wrong before you measure it.

Then attend the meeting, having already formed a view.

What to actually look at

The junctions, not the surfaces. Surfaces are what people show you and are also the last thing to go on. Junctions are where the design either survived or did not, and they are visible for a short window before they are covered.

Anything at the point of concealment. Services about to be boxed in, insulation about to be plastered, a damp-proof membrane about to be screeded over. There is a small window in which these can be seen at all, and it is the highest-value moment of any visit. Ask what is going to be covered before the next visit, and look at that first.

Dimensions you assumed. Ceiling voids grow. Screeds get thicker. Head heights in corridors drop below what the section assumed, forty millimetres at a time, through a sequence of individually reasonable decisions. Take a laser measure and check the ones that mattered in the drawing.

Substituted products. Look at the labels on what is on site, not at what the submittal register says. The two diverge more often than anyone expects, usually without ill intent — a supplier was out of stock, a like-for-like was taken by someone with no reason to know why the original was chosen.

Where the water goes. Roofs, thresholds, upstands, and any horizontal surface meeting a vertical one. Water ingress accounts for a large share of post-completion problems, and almost all of it is visible during construction to someone who thinks to look.

Say it on the day

The most common failure is not seeing a problem. It is seeing one, deciding to raise it in the next report, and letting a week pass while the work continues.

A deviation caught on the day is usually a conversation. The same deviation caught a week later has fourteen more metres built on top of it, and it has become an instruction, a cost, and a relationship problem. The asymmetry is severe enough that the correct default is to say it immediately, in person, even at the cost of being wrong occasionally in front of people.

Being wrong occasionally is fine, incidentally. A site team quickly learns whether an architect raises things because they matter or because they are performing vigilance, and the ones who raise real things and accept correction gracefully get listened to.

Write it down the same day

Site notes written the following morning are already a reconstruction. Write them in the car, on the train, or standing in the car park, while the specifics are still specific.

The useful record is not "discussed progress at level two". It is what was seen, where, what was said about it, by whom, and what happens next. Photographs with location and date, and a written note against each — a photograph without a note is almost worthless six months later, when nobody can remember what it was showing.

This matters commercially as well as practically. Contract administration disputes are decided on contemporaneous records, and the practice with the better notes is generally the practice that is right.

What the visits give back

The traffic is not one-directional, and this is the part that gets undersold.

Administering construction makes a practice better at drawing. You find out which details were unbuildable, which sequences you assumed wrongly, and which parts of the package the site team ignored because they were unclear. Practices that hand over after planning never receive this feedback, and they reproduce the same difficult junction for a decade because nobody has told them it takes a bricklayer forty minutes.

On a school extension in Ghent, the timber cassettes were craned in over eleven weekends, which meant eleven consecutive Saturday visits watching an unfamiliar system go up. Almost everything we now know about tolerances in prefabricated timber came from those mornings, and it went straight into the next project's package. No course would have taught it.

Roughly half our people are on site most weeks. It is the main reason our technical drawings have improved, and it is the least glamorous form of professional development in the profession.

The one thing to remember

You are not there to inspect the work. Inspection is the contractor's obligation and the building control officer's role, and an architect who behaves as an inspector both oversteps and, oddly, sees less.

You are there to find out whether the building being made is the building that was designed, and to say so quickly when it is not. That is a narrower job and a more useful one, and it is done mostly by looking at things nobody has drawn your attention to.