Documented, Observed, Inferred

Documented evidence is supported by a record or supplied material: a drawing, a maintenance log, a compliance certificate, a survey, a permit with a clear scope, an owner’s confirmation. Observed evidence is visible in what is available to the assessment: a photograph, aerial or street imagery, a site inspection, a listing image. Inferred evidence is reasoned from era, type and contextual signals: a building of this age and type in this location probably has this kind of plant or these materials.

Each grade has a place. Inference is often the only evidence available for concealed elements, and it can be very good inference. It is still inference.

The danger is not inference. It is inference that has forgotten it is one.

The grades are easiest to apply when each comes with examples of its evidence and of what would move it up.

The clearest recent example of an inference hardening into fact is the 30-year lifespan widely attributed to reinforced autoclaved aerated concrete during the 2023 school closures in England. The figure was repeated across news reports, professional commentary and guidance. The Institution of Structural Engineers has since said the claim is misleading, traced to a 1995 exchange of correspondence, with no specific data behind it.

Every mechanism this paper describes is visible in that case: a figure repeated without its source, summarised without its caveats, and finally treated as an engineering fact. A register that grades evidence would have carried it as inferred and unsupported, and priced the investigation rather than the date. What Age Conceals sets out the case in full.

Inference hardens into assumed fact in predictable ways. It is repeated across documents without its source. It is summarised into a headline. It fills a field that has no way of saying how it was derived. An inferred construction year appears in a report, is copied into the next one, and becomes the building’s age.

Dates are a common example. A planning application or permit tells you that work was proposed or approved at a certain time. It does not, on its own, establish when a building was completed. A capital view that treats the permit date as the construction date may misjudge the age of every system in the building.

Evidence grade feeds directly into confidence. A line whose condition and cost basis rest on documented evidence can carry high confidence. One resting on inference should carry lower confidence and, where the stakes justify it, an action to confirm. This is what lets an owner see where the capital position is firm and where it is soft.

It also shapes how uncertainty is priced, which is the subject of Pricing What You Don’t Know.

Documented: “The roof was re-sheeted in 2014; invoice and scope supplied by the owner.” Observed: “Aerial imagery from 2015 onward shows a uniform new profile across the western bays.” Inferred: “Buildings of this type and era in this location typically had asbestos cement roofing originally; the eastern bays may retain it.” Each statement is useful. Each supports a different level of confidence. And each tells the reader what would strengthen it: the inferred statement points directly to a hazardous materials survey.

Keeping the grade visible stops the inferred statement from being read as the documented one six months later in someone else’s report.

Assessments increasingly draw on public data, imagery and AI-assisted reading of documents. These tools are powerful at producing plausible statements. Plausibility is not evidence. A disciplined method has to keep asking of every claim: where did this come from? What Public Evidence Can Tell You discusses the limits of desktop reading, and the Moyne Ross approach to AI sets out how models are kept inside this discipline.