A building keeps its own record. Not in the archive, though that helps, but in its surfaces: the worn hollow in a stone step, the paint layers at a door frame, the change in mortar where an opening was filled in 1937, the scorch above a stove nobody remembers.

This information is not protected by heritage designation, because designation attaches to significance and none of these things are significant in that sense. They are just traces. And they disappear during refurbishment as a matter of course, not through vandalism but through the ordinary operation of a specification that says clean, make good, redecorate.

I want to argue that removing them is a decision, and that it should be made deliberately rather than by default.

What surfaces record

Wear records use. A stair tread hollowed on the left tells you which side people walked for a century, and it is almost always the side nearest the wall. A threshold worn at one point tells you where the door actually swung, which is sometimes not where the current door swings.

Repair records failure. Where a building has been patched, something went wrong there: water, movement, fire, a change of use that required an opening. The repair is a diagnosis written in a different material, and it is frequently the only evidence of a problem that will recur.

Filled openings record function. A blocked door in a party wall means two buildings were once one, or once connected. A ghosted window under render means the room behind it was lit differently, which usually means it was used differently.

At a boathouse in Copenhagen, the original slipway ran into the water and was still there, unused since the 1970s. The obvious move was to floor over it and gain 40 square metres of exhibition space that the client badly wanted. We kept it open. It floods twice a month at spring tides, which the client accepted, and it is now the thing visitors photograph. It is also the only part of the building that explains what the building was for.

Cleaning is not neutral

The specification language is what makes this invisible. "Clean down and make good" reads as maintenance. It is in fact a substantial intervention, and it is generally the largest single act of removal in a refurbishment project.

Abrasive cleaning of brickwork removes the fired surface, which is the durable part, and exposes the softer body beneath — a well-documented problem that continues to happen because the result looks better on completion. Chemical cleaning of stone removes the patina that has taken a century to form, and the building then looks new, which people frequently want and rarely think through.

Even paint removal is a decision about which moment in the building's life is the correct one. Stripping back to bare timber does not return the building to an original state; it returns it to a state it was never in, since the timber was painted from the beginning. The choice of layer is arbitrary, and it is usually made on the basis of what looks best rather than what is true.

None of this means cleaning is wrong. Sometimes soiling is genuinely damaging, and sometimes a building's condition is simply grim and the occupants deserve better. It means the question should be asked out loud, with somebody in the room able to say what the cleaning will remove.

The problem with legibility as a rule

I have argued before that new work should be identifiable as new. There is a version of that principle which becomes its own kind of erasure, and I should be honest about it.

If every intervention must be visibly contemporary, the building's history acquires a hard edge: everything before a certain date is old, everything after is new, and the two-hundred-year continuum of small adaptations in between gets flattened into a single category. Buildings do not experience their lives that way. They accumulate.

The more useful discipline is that each intervention should be honest about its own date without insisting on rupture. A repair that uses matching brick but a slightly different mortar is legible to anyone who looks, and invisible to anyone who does not need to. That is often the right register — particularly for repairs, as opposed to insertions, where the reasoning is different.

Where an insertion is genuinely new accommodation rather than a repair, contrast makes sense. At the boathouse the steel and glass box inside the timber shell is unambiguously of this decade and touches nothing. That is a different act from patching a rotten sill, and it deserves different treatment.

Practically, what this means

Survey the surfaces, not just the structure. Photograph before anything is touched, at close range, with a scale. Most projects photograph the building; very few photograph the wear.

Write the specification to require decisions rather than assume them. Instead of "clean down all brickwork", specify sample panels and a hold point, with someone nominated to approve. This costs almost nothing and converts an automatic removal into a considered one.

Record what is removed when removal is right. A short written note and a photograph is enough, and it is the difference between a loss and an undocumented loss.

Keep at least one place where the record survives. A single wall left unrestored, a stair tread not replaced, the slipway left open — one honest surface does more to explain a building than a display panel ever will.

The limit

There is a sentimentality in this position that I want to name.

Not every trace is worth keeping, and a building preserved as a record of itself is a museum rather than a working thing. People need buildings that are warm, dry, accessible and pleasant, and those requirements will sometimes require removing evidence. A client who needs 40 square metres and cannot afford to lose it is not being philistine.

The argument is narrower than it might sound. It is only that these removals should be visible as choices at the time they are made — because at present most of them are not choices at all, and nobody involved ever registers that something has gone.