Documents describe intent. Photographs record facts.
A construction project generates an enormous number of documents, and nearly all of it is a statement about the future. Drawings say what will be built. The programme says when. Method statements say how. Even progress reports, which sound like accounts of the past, are usually someone's summary of how the future is going.
None of that is a criticism. A project has to be planned, and planning produces documents. The problem arrives later, when something goes wrong and everyone reaches for the paperwork to establish what actually happened — and discovers they are holding a stack of intentions.
What Reality-Driven Intelligence is
Reality-Driven Intelligence (RDI) is a public framework published by Evercam at rdi.evercam.io. It is not a product. It is a description of how site reality should be captured, verified, interpreted, acted on and measured — and, more usefully, of the order those things have to happen in.
The framework sets out five layers, from the bottom up: Reality Capture → Ground Truth → Interpretation → Action → Command.
Reality Capture is observation of the physical site: sensors, cameras, wearables, equipment telemetry and site activity. The raw material — what was there, at what time, recorded by something with no opinion about it. Everything above inherits its limits: intermittent capture makes everything intermittent, and nothing without a timestamp can be placed in sequence.
Ground Truth is time-aligned, location-aware evidence of what is actually happening — observation turned into something verified and traceable back to what produced it. This is where most of the difficulty lives. A photograph is not evidence; a photograph with verified provenance is. The difference only shows when someone contests it, which is exactly the moment you cannot go back and add it.
Interpretation is meaning: detecting patterns, exceptions, risks and likely outcomes. Is that the third pour or the fourth? Is standing water on that slab normal for this stage? This is the layer the industry talks about most, and it is worth nothing without the two beneath it. A model reading a contested photograph produces a confident answer to an unanswerable question.
Action is alerts, workflows, escalations, tasks and follow-ups — what happens as a result. The test is whether the loop closes. Most construction data initiatives fail here, not because nobody understood the data, but because understanding it was not connected to anybody's decision.
Command is where the stack pays for itself: leaders directing attention, coordinating the response, and measuring whether any of it changed the outcome.
Why the order is the point
The stack is not a menu. You cannot buy a layer.
This is the single most useful thing about the framework, because it explains a decade of disappointment in construction technology. The industry has spent years being sold Interpretation — analytics, dashboards, increasingly AI — on top of a Reality Capture layer that is patchy and a Ground Truth layer that mostly does not exist. The output looks impressive and cannot be relied on, because the thing underneath it was never verified.
The same trap is being set one layer higher. There is great enthusiasm for automated decision-making in construction and in the insurance around it. Decisions are the Action layer. An Action layer resting on unverified Interpretation resting on intermittent capture will make confident, fast, well-presented mistakes.
Build from the bottom. It is slower, and it is the only thing that works.
Bought by the project, valuable far above it
Now notice something about the shape of the stack.
Reality capture is almost always bought as a project management tool. Time-lapse for progress meetings, walk-throughs for coordination, sensors for the wet trades. The purchase sits with the project team and the case for it is operational. That is how the kit gets on site, and it is the right way round — a tool has to earn its keep where it is used.
But the framework does not end with the project team. Its top layer is written for leadership: directing attention across the work, coordinating response, measuring outcomes. A framework that terminates in the boardroom is making a claim — that verified site reality is not a site tool that happens to produce pictures, but an asset whose value runs up and out of the organisation that bought it.
Follow it up. Finance signs payment applications and carries valuations that rest, in the end, on someone's account of what has been built; a verified account of the site turns that into something checkable. Commercial teams carry disputes that turn on sequence and condition — precisely what continuous capture holds and periodic records do not. The board carries a portfolio it sees mostly through reports written by the people it is trying to oversee. And outside the organisation entirely sit the parties carrying the project's risk, who have historically seen almost none of it.
One purchase, made for operational reasons, producing an asset most of whose potential readers never open it. That is the situation the framework puts a name to.
And in one place, that value can now be counted
The honest difficulty with everything above is measurement. Most of those benefits resist a number — you cannot run the project twice, once watched and once not, and compare. For the severe physical damage losses in construction insurance — the fires, the collapses, the major escapes of water — events are too infrequent and too unalike for a book to accumulate credible experience of what continuous evidence contributed.
But not everywhere. Where the figures are already agreed, worth can be calculated instead of observed. In construction insurance that place is delay in start-up cover, where the sum insured, indemnity period, waiting period and rate are all written into the policy before anything happens. Our Reality Capture Value calculator does that arithmetic: on a project's own figures, it shows how many days off the waiting period would repay the entire capture spend. Not a forecast — arithmetic, checkable by anyone with a handful of numbers and a reason to check.
One corner of the picture, counted. The rest of the stack's value arrives the way the framework says it does: from the bottom, layer by verified layer.
Why photographs, specifically
Structured data answers the questions you knew to ask when you installed it. Imagery answers questions asked in the future.
That is the property that makes reality capture foundational rather than merely useful. A moisture sensor tells you about moisture. A camera tells you about the moisture, the scaffold, the sequence of the pour, the material stacked where it should not be and the weather that day — including the parts that only become interesting eighteen months later, when someone asks a question nobody anticipated.
It is also the only layer of the stack with no stake in the answer. A photograph taken by a fixed camera on a Tuesday has no position on the dispute that arrives two years later. Every document on the project has an author with an interest. The camera does not.
Where this leads
The framework applies well beyond insurance — anyone trying to build something reliable on construction data is working somewhere in this stack, whether they name it or not. It is worth reading in full at rdi.evercam.io.
Pikt works in the same stack. It gives risk engineers continuous visibility of the projects they oversee, by connecting to the reality capture and sensing already running on site and turning it into one time-stamped view of each project. That is the bottom of the stack made continuous, and made available to the people carrying the risk.
Everything else in construction risk is built on what the site actually was. What has been missing is not the account. It is its continuity.
Pikt — continuous site visibility for construction risk engineering. Pikt is an independent technology company. It is not an insurer, MGA or broker, and gives no insurance advice.