What a model can never prove: as-built evidence

The second answer is the interesting one

You can now ask a model whether the escape routes comply, and it will answer. Distances, widths, counts, a tidy compliance sheet. It is a genuine advance, and anyone who works with models has been waiting years for it.

The interesting part, though, is the second answer: the list of things the system itself declares it cannot check. That list always existed, but it lived in the head of whoever was supervising. Automation is writing it down for the first time — and in doing so it draws a precise boundary around where field work begins.

What automated code checking does genuinely well

Geometry, distances, areas, counts, the presence or absence of declared properties. All of it deterministic: the same question about the same model returns the same answer today and in three months, with no fatigue and no drifting judgement. No manual review competes with that on speed or consistency.

Worth giving full credit without the obligatory “yes, but”. Done well, automated checking removes an entire category of avoidable errors before they reach site, and frees review hours for the problems that actually need judgement.

The second list

Every honest implementation ends with a caveats section. Conditions that depend on what is physically installed. Elements nobody models — ironmongery, signage, sealing, small equipment. The actual use a space will be put to, which is not always the use in the design statement. And decisions that simply have not been taken yet at the moment the model is checked.

These are not tool failures. They are category boundaries, and saying them out loud is a sign of maturity, not weakness.

Intent versus fact

Here is the distinction that organises everything else. A model is a statement about what should be. A building is a statement about what is. Two different objects that look very alike, and that get conflated in meetings constantly.

Final compliance — the kind you defend to a third party — concerns the second one. You can have an immaculate model and a building that departs from it at a dozen minor points and two significant ones. Nobody lied. Construction happened.

Where model and building diverge, and why that is normal

A material substituted because the lead time went to fourteen weeks. An opening moved 200 mm to dodge a beam that ran elsewhere in the model. A service rerouted because the duct simply did not fit.

None of that is fraud and none of it is sloppiness. It is building. The problem is never that divergence exists; the problem is undocumented divergence. The first is site management. The second is a debt somebody pays later, usually with interest.

The evidence problem: handover documentation and as-built records

The moment a wall closes, the only remaining proof of what sits behind it is whatever someone recorded before it closed. There is no later appeal. Either the evidence exists or it does not.

And a great deal rests on that. The monthly valuation, which argues about executed quantities. Practical completion and its snagging list. The handover pack and the O&M manual, which will follow the building for its whole service life. The warranty claim, when damp appears three years in. And the refurbishment somebody designs in fifteen years, whose first problem will be working out what is actually behind that wall.

What good field evidence looks like

Four conditions, and they are demanding: place (where exactly, not “on the second floor”), a reliable date, an identifiable author, and enough context that a stranger understands the photo a decade later without phoning anyone.

A loose photo in a camera roll fails all four. It has metadata, yes, but it has no place in the sense that matters — the point on the drawing — no context, and no relationship to the rest of the record. Worth being blunt: this is a documentation problem, not a photography problem. Better cameras do not fix it.

Why this gets more important, not less

You might expect the opposite: if design-stage control tightens, surely there is less to check afterwards. It works the other way round. The stricter and more explicit the input control, the more visible the distance between an approved model and an executed building becomes — and the more people ask about it.

A verified model raises the bar for everything downstream. The question stops being “was it designed correctly?” — that is answered and signed — and becomes “and was it built that way?”. Only as-built evidence gathered in the field answers the second one.

What it means for the person on site

Almost nothing, and that is how it should be. Nothing has to change about how you work. What has to be true is that capture is fast enough to survive a real day, in the rain, with gloves on, with someone waiting for you.

In PinMy that is a few seconds: open the drawing on the phone, tap the point where the service runs before it gets closed up, dictate the Description by voice, attach the photo. It keeps its place, date and author, and it is found by looking at the area. At month end you can produce a PDF report for one file listing every pin and comment on it; that report is in beta and publishes its own limits. If you also work with models, we have written about when IFC beats a PDF on site and about how to structure a site visit.

What PinMy is NOT

PinMy does not check building codes, does not verify compliance with anything, and does not issue signed justifications. That is the supervising engineer’s work and professional judgement. PinMy is the field-capture layer, and its job is that the evidence exists and can be found.

On 3D it is worth being exact, because this is where the sector overclaims: a pin on an IFC model in PinMy is anchored to a point in three-dimensional space, not to a model element. We do not read element properties, we do not interpret model data and we do not version models. It is a dated snapshot for field documentation — not BIM coordination, and no replacement for Revit or for your common data environment.

FAQ

Will automated model checking replace the supervising engineer? Not the interesting question. What it does is cleanly separate two jobs that used to be tangled together: checking what was designed, and checking what was built. The second still requires somebody looking.

How much evidence should we keep? The working rule: everything that will end up concealed, and everything that departs from the design. What stays visible can be looked at again. The rest cannot.

Does a phone photo count as as-built documentation? It does if it keeps its place, date, author and context, and if someone can find it without phoning you. A loose photo in a camera roll meets none of those four.

The other side of the line

Automation is drawing a sharp line around what a model can answer for. That is good news even for people who will never open a model, because until now the line was blurry and everyone put it wherever suited them.

The other side of that line was always somebody’s job. What is new is that it now has a name, and can be asked for. If defending what was built ever lands on you, send this to the colleague still keeping the proof in a camera roll.

• Start free: pinmy.co
• Book 15 minutes: tidycal.com/pinmy