Inspection & Clash in the Field

Human vision is extraordinary at one particular thing: detecting mismatch. Line two things up that should be identical and any difference between them - a shift, a gap, a missing piece - jumps out almost instantly, far faster than reading numbers off a drawing and comparing them mentally. Field inspection with AR is built entirely on this. Register the as-designed model over the as-built structure and the two should coincide; wherever they do not, your eye is drawn straight to the discrepancy. A wall that has crept off its line shows as the real block wall separating from its ghosted overlay. A duct rerouted on site shows as the real duct diverging from the model's path. A slab edge poured short shows as a strip of overlay hanging past the concrete.

This makes AR a fast, *wide* inspector. Instead of checking one dimension at a time, a site engineer walks a floor and the overlay presents the whole design-versus-reality comparison continuously, letting obvious departures surface across a large area quickly. That breadth is the value: it is a way to *find* the places worth attention among the vast majority that are fine, so limited inspection effort goes where it is needed. On a busy site where no one has time to measure everything, a fast visual triage that reliably surfaces the gross departures is genuinely useful.

But the same sentence that describes the strength names the limit. The eye flags a *candidate* - something that *looks* off. It cannot tell you whether the wall is 8 millimetres or 80 off, whether that is within tolerance or not, or whether the apparent gap is a real deviation or just overlay drift. Registration error, as earlier lessons stressed, can itself produce an apparent mismatch where the building is actually fine, and can hide a real one where the overlay happens to drift *into* alignment. So the overlay is a superb generator of questions and a hopeless source of answers. It says "look here"; it never says "this is wrong by this much". Everything this lesson goes on to describe - deviation-spotting, field clash, quality and progress checking - lives inside that division: AR finds the candidates fast and wide; verified measurement decides each one.

Office clash detection compares model against model - one discipline's ducts against another's beams - and resolves conflicts before anyone builds. It is one of BIM's most valuable routines. But it assumes the building is constructed exactly as coordinated, and real construction constantly breaks that assumption. A service is rerouted on site to dodge an unforeseen obstruction, and now fouls a structural member it cleared in the model. An element is installed slightly out of position and clashes with the next trade's work. A temporary support, a bracket, a fixing that was never in the model occupies a space something else needs. These are real-world clashes - conflicts that exist in the built structure even though the office model was clean - and they are exactly what a coordinated overlay can surface in the field.

Walking the structure with the model overlaid, a site engineer sees where the real installed work and the designed work no longer agree, and where a real element now conflicts with where the next stage must go. Caught early - before the next trade arrives, before the boxing-out, before the pour - such a clash is cheap to resolve; caught after, it can mean rework, delay and dispute. This is the field extension of clash detection: not re-running the office check, but noticing where reality has departed from it. For congested MEP zones especially, where a single reroute can cascade, this early visual catch is real value, and it is one of the more genuinely useful on-site uses of the overlay.

The discipline is identical to deviation-spotting. The overlay flags a *suspected* clash; it does not adjudicate it. Whether the real duct truly fouls the real beam, or merely appears to because the overlay has drifted, is decided by looking properly and measuring - a tape, a level, a total station where needed - against the verified drawings. Many an "AR clash" dissolves on measurement into registration error; some are real and must be raised through the proper channel and resolved by the design and site teams. The overlay's job is to make sure the real ones are *found* early, while there is still cheap time to fix them. It earns its place as an early-warning scanner for the messy gap between the coordinated model and the built reality - and stops precisely at the point where confirmation begins.

Beyond deviations and clashes, the overlay supports two everyday site jobs: checking quality (is what was built right) and checking progress (how much of it is built). For quality, walking the as-built with the model overlaid surfaces candidates for workmanship and conformance issues at a glance - an element that looks mislocated, a route that departs from design, a component that seems the wrong size or missing. For progress, comparing the model's planned state against what physically exists gives a fast, visual read of how far the work has got - which bays are poured, which services are in, which stage a floor has reached - useful for a site meeting, a client update or a progress claim, and far quicker than tallying it element by element from drawings.

Both are genuine efficiencies, and both are firmly bounded in the same way. A quality check by overlay produces a *list of things to inspect properly*, not a conformance decision: whether the mislocated-looking element is actually out of tolerance is a matter for measurement against the verified drawings, and whether workmanship is acceptable is a matter for the responsible inspector and the governing standards, not for how well two ghosts line up in a headset. A progress read by overlay is an *estimate for understanding*, not a certified quantity: it is excellent for grasping and communicating status, and it must not become the measured basis of a payment or a formal completion certificate, which require proper measurement and sign-off. In each case the overlay accelerates the *seeing* and leaves the *deciding* to the accountable process.

Kept in these lanes, the value compounds. A site engineer using the overlay covers more ground, notices more early, and directs the slow, exact tools - measurement, formal inspection, survey - to where they are actually needed, which on a stretched Indian project with lean supervision is a real gain in reach. Progress and quality understanding become more shared and more visual across the team, client and PM. But the moment an overlay read is promoted from "worth checking" to "checked", or from "looks about eighty per cent" to a certified quantity, the tool has been pushed past what it is, and the accountability that inspection and measurement exist to provide has quietly gone missing. The skill, again, is to let the overlay do the fast, wide seeing and to keep every binding quality and progress decision with measurement, the responsible inspector, the verified drawings and the standards.

Put the pieces together into an honest field-inspection workflow, and the boundary becomes a simple, repeatable habit. Step one - overlay: register the coordinated model, lightened to the zone and trades in question, over the as-built structure, anchored to survey control so the alignment is as good as it can be, and read as approximate. Step two - flag: walk the area and let the eye surface candidates - deviations, suspected real-world clashes, quality concerns, progress gaps - noting each as a question, with a photo or marker, not a verdict. This is where AR does its fast, wide, genuine work. Step three - confirm: for each flagged candidate, verify with the appropriate instrument and against the verified drawings - a tape or level for a length or plumb, a total station where position or tolerance is at stake, the responsible inspector for workmanship - discarding the ones that turn out to be overlay drift and confirming the ones that are real. Step four - record and act: raise the confirmed problems through the proper channel (a non-conformance report, an RFI, a coordination note), with the binding measurement, not the overlay, as the evidence, and resolve them with the design and site teams.

The shape of this workflow is the whole lesson. AR sits only in steps one and two - the seeing and the flagging - where its speed and breadth are a real asset. Steps three and four, where anything becomes binding, belong entirely to measurement, the accountable inspector, the verified drawings and the standards. The overlay never confirms and never records a binding result; it earns its keep by making sure the real problems reach steps three and four *early*, while they are still cheap to fix, and by sparing the slow tools from having to scan everything blind.

This is the same principle the module has held throughout, now in operational form: spatial computing is a tool for seeing and communicating, never a source of truth. In field inspection that means AR is a superb early-warning scanner and a poor adjudicator, so you build the workflow to use the first quality and never rely on the second. Do that, and on-site inspection genuinely improves - wider coverage, earlier catches, more shared understanding - without ever letting an unaccountable picture make a binding call about quality, position or progress. The overlay flags; measurement and the inspector confirm; the verified record and the standards decide. Keep those roles clean and the tool is a real help; blur them and it becomes a liability dressed as efficiency.