Construction site inspections fail after the walk, not during it

Somebody has to stand on the finished work, measure it, photograph it, and put their name to a record saying it matches the drawing. That is the job. It runs under several names: field inspection and QA/QC inspection are what a contractor's own quality system calls it, the list of defects drawn up near handover is a punch list in the United States and a snag list in Britain, and on public work the formal version has hold points written into an inspection and test plan.

The reason it matters is that the record outlives the walk. Somebody who wasn't there has to be able to rely on it later.

The software category built on top of that has settled on a single promise. Photograph the work, measure it, compare the result against the drawing, and the defect surfaces before anything covers it up. Detection, in other words: find the thing the eye missed.

People who do this every week describe a different failure. The walk finds plenty. What goes wrong is everything the walk hands off to.

Chris Hatch, a commercial developer, published the sequence he runs a project through, eighty-one numbered steps from site selection to sale. Two of them are about watching the job. Step 57 is "Setup an oxblue camera or some other construction monitoring device so you can keep tabs on your project 24 / 7 / 365". Step 58 sets a site walk "every 1-2 weeks depending on the scope and scale of your project".

Then step 69 arrives, and the camera isn't the answer to it. "Final couple of site walks are critical to be done in person, with your architect and with the PM or CM for your tenant."

The operator with a camera running every hour of the year still books the last inspections in person, and says so in the same document.

He isn't unusual in the cost he accepts for that. Sean Sweeney, another developer, tells owners to "Show up unexpectedly at your construction site", to know the floor plans well enough to catch a wrong layout, and to check the light switch locations. His own habit is quieter: "I also walk my projects alone on the weekends. I like it quiet so I can think. I usually always notice something."

An owner-developer put the bill in plainer terms, asking who actually wants to do it: "How many want to walk the building at 5:30a in 20 degree weather? Once a week, for 2 years?"

Ask what a practised inspection actually looks at and the answer is rarely dimensional. The same owner-developer set out the five things she watches when she walks a job site: cleanliness, staging, logistics, security and safety, and leadership. Is there a designated rubbish area. Are materials stored where they'll be damaged or easily stolen. Are some subcontractors stopped, waiting for others to finish. Does every trade have a foreman on site who can be reached.

Her reason for reading those things is diagnostic. They tell her, in her words, about "a GC's ability and the project's likelihood for success".

Not one of the five is a measurement. She is grading the organisation that will be building for her for the next two years, on the theory that a site that can't manage its own bins is telling her something about what ends up behind the plasterboard. A tool that checks whether the slab came out the right size is answering this buyer's second question.

Marilyn Moedinger, an architect who has also worked as a general contractor, was asked what to record before work gets covered up. Her answer was about filing rather than photography: "Photos, but make sure you can tell where they are taken and what they are taken of. Save them with helpful file names ('south kitchen wall') otherwise you have a bunch of zoomed in stuff that makes no sense later".

Nobody in the trade makes a stronger case for automating this, and it's worth noticing what she isn't asking for. Not a system that looks at a wall and judges it. She wants the photograph to arrive already knowing where it was taken and what it shows. The defect was visible at the time. The record of it was what turned out to be useless.

That distinction decides when the photo is worth anything. A site photograph carries weight only once somebody disputes the work, which is generally months later, by which point whoever took it has moved on and the file is named by a camera rather than by a person.

Comparing a photograph to a drawing assumes the drawing can be trusted. Three accounts suggest the reference is shakier than the premise allows.

An engineer described spending months helping refit a new school block after it failed inspection, and named the cause: "The original construction was terrible... Not a single contractor involved knew the required specification." Nothing there was mismeasured. Everyone was building accurately to the wrong number.

Sometimes the drawing is the defect. Caleb, a steel fabricator, was on site getting ready to install a large I beam when something on the print stopped adding up. He called the engineer. "Five minutes later he called back in a panic. He called out the wrong beam in the stamped prints. We have to haul this sucker out and bring in a new one." Stamped, approved, wrong.

Even the instruments disagree. A mason who works alongside other trades warns them to check every tape before a job involving several people, and keeps test boards on site so each arriving trade can check theirs against a common reference. Any tape that disagrees gets "culled and relegated to drystone work". Four trades working off four rulers is a tolerance problem that no photograph settles.

There is a document meant to close exactly this gap, and a heritage contractor sets out what makes one real. A method statement is "the sequence, the protection and the hold points", and if it doesn't name the fabric, the sample and the stop rule, "it is not a method". That reference is what an inspection is supposed to test against, and on most jobs it isn't written down anywhere.

Go back to Hatch's step 69, because its second half is the argument. After the walk that has to happen in person: "Create one clear punch list and then diligently follow up to ensure every single item is addressed with photographic evidence of what was done." Step 70 is another walk, once the list is complete, before anything is handed to the tenant.

One list. One photograph per item. Chased until every item is shut. That is the work, and almost all of it is clerical.

The same shape appears when the evidence is demanded by someone official. A site worker described spending over an hour with a safety regulator, producing method statements, rescue plans, rigs, ropes, toolbox minutes and inspection tickets. The inspecting had happened weeks earlier. The hour went on proving it.

A caveat about the numbers, because the obvious ones are missing. The figures that would size this properly, hours a week lost to inspection paperwork, days to close a punch list, the cost of a failed inspection, aren't ones practitioners state. They appear almost only in the marketing of companies selling inspection software, which is a reason to treat them carefully rather than to repeat them here.

Which leaves a better thing to count than photographs captured or defects detected: items closed with their evidence attached, and how long each one sat open first.