Site observation report for architects, construction administration

The same observation keeps coming back, and it isn’t the contractor

19 September 2026

Read three years of a practice’s site reports end to end and the same handful of observations turn up again and again. Different jobs, different contractors, different cities. The parapet upstand flashing. The threshold at the terrace door. The junction where the soffit meets the column.

The comfortable reading is that trades are careless. The useful reading is that the practice keeps drawing something that is hard to build.

Five Contractors Cannot All Be Wrong in the Same Place

One contractor getting a detail wrong is a contractor problem. The fifth one getting the same detail wrong on the fifth job is a drawing problem, and the only reason it does not feel that way is that each instance arrives a year apart, on a different site, in a different report.

Nobody experiences it as a pattern. They experience it as a Tuesday.

This is the part of construction administration that gets talked about least. The site report is understood as a defensive document, and it is one. It is the record that says the architect was there, what was seen, and on what date. It is also the only structured feedback the studio ever gets about the thing it drew, written by the person best placed to judge it, and most practices throw that away by filing it.

The Report Is the Only Feedback Loop a Studio Has

Consider what else is available. The contractor’s RFIs show where the drawings were unclear, which is related but not the same question. The client says whether they are happy. Neither of those reveals that the standard terrace threshold has been flagged on four of the last six projects.

The observation log does, but only if it can be read across jobs rather than within one. Plenty has been written about what belongs in a site observation report for architects, and the honest answer is that the format matters far less than whether it stays the same from visit to visit and from project to project. A log that can be searched is worth more than a beautifully laid out one that cannot.

That asks for a few boring commitments, none of which are about software.

Observations get numbered continuously across the job rather than per visit. Observation 47 stays observation 47 on every report and in every email about it. Per-visit numbering means that item 3 refers to eleven different things by practical completion, and the correspondence stops being searchable at exactly the moment somebody needs to search it.

Each observation carries a location precise enough to find again: grid line, room number, floor, elevation. North elevation is not a location on a building with four bays of it. And each one carries two photographs rather than one, because a tight shot proves a crack exists while a wide shot from three metres back proves it is the crack at the north stair landing and not one of four similar cracks elsewhere in the building. The second photo is the one people skip, and it is the one that is missing when the file gets reopened.

Carried-over items stay visible. An item first raised on visit 4 and still open on visit 9 is a different fact from one raised today, and only a carry-over section shows the difference.

The Ten Seconds After the Photo Decide Everything

None of this survives the drive home. Where the architect was standing, what they were looking at, whether the sub had already been told about it last visit: all of that is available on site and gone by the evening. Accuracy drops with every hour between the walk and the write-up, and an observation dated three days after a visit is easier for another party to argue with than one dated the afternoon it happened.

So the write-up happens on site or it happens badly. In practice that means capturing in walk order on a phone rather than in a notebook to be typed up later. Several iPhone and iPad apps do this now, Sitewise among them, stamping each photo, voice note or sketch with the time it was taken and a GPS pin when location is switched on, then exporting a formatted PDF or DOCX before anyone gets back in the car. The specific choice matters less than whether the record gets made while the thing is still in front of you.

What This Does Not Tell You

Recurrence is a signal, not a verdict. A detail flagged on four jobs might be badly drawn. It might equally be a detail that works on paper and gets value-engineered on the way to site every time, or one where the same specialist subcontractor has been novated onto four projects because they are the only firm in the region doing that work.

The observation log will not separate those. What it will do is show that the question is worth asking, which is more than most practices have now.

It also says nothing at all about the jobs where the detail worked, because nobody writes an observation to record that something was built correctly. That gap is real and worth naming. The log is a record of what went wrong, so reading it as a portrait of a practice’s work will make that work look worse than it is.

Read the Log Once a Year

Pull every observation from the last twelve months across every job and sort by what it was about rather than by project. An hour of somebody’s time. Most practices have never done it, which is why the exercise usually turns something up inside the first twenty minutes.

If the same junction appears more than twice, take it back to the detail. Either it needs redrawing, or it needs a note on the drawing saying what the last four contractors got wrong. Either way the studio is designing with information it already owned and had never looked at.

The site report was always going to be written. It may as well be worth reading.

Comments on this guide to Site observation report for architects, construction administration article are welcome.