What a conservator taught us about condition surveys

The first version had a text field. You described the damage - "water stains, some foxing, faded ink at the bottom" - and the description became the restoration instruction. It demoed well, and it was completely wrong, which we found out by showing it to someone who does this for a living.

Two operators, two restorations

The objection was simple and, once said, obvious. Two people looking at the same water-damaged parish register will write two different sentences about it. One says "heavy staining across the lower third." The other says "tide line and mould in the bottom corner, some cockling." Different sentences produce different restorations. And neither restoration can be justified afterwards, because the justification is a sentence somebody typed once and then forgot.

For a landscape photograph nobody cares. For a legal or historical record, an operation that cannot be reproduced or explained is not a treatment - it is an event that happened once. If a court or a conservation committee later asks "why does this page look like this," the honest answer would be "because of a paragraph an operator typed on a Tuesday," and that answer is not acceptable.

Conservators do not describe damage in prose

They record it against a controlled vocabulary, grouped the way damage actually arrives - by cause, because damage from the same cause tends to need the same treatment and to co-occur. Our conservator walked us through the groups, and they became the structure of the survey.

• Storage and age - water tide lines, foxing, mould, insect and rodent damage, brittleness, yellowing.
• Handling - roll and fold creases, tears, tape residue, staple and punch holes, faded and abraded ink.
• Capture problems - glass glare, frame shadow, keystoning, uneven lighting, camera shake.

Recording against that vocabulary means two operators surveying the same page tick the same boxes, because the boxes are named for observable conditions rather than for impressions. Same survey, same instruction, same treatment. The subjectivity that lived in the prose has nowhere to hide.

So we deleted the prompt box

The condition survey became the interface. There is no free-text field where damage is concerned; the restoration instruction is derived from the structured survey, and the survey travels with the job as its record. Ask why a page was treated the way it was, and the answer is a checklist somebody can read, disagree with, and re-run - not a paragraph nobody can reconstruct.

If you cannot reproduce the treatment, you have not restored anything. You have produced one image, once, for reasons nobody wrote down.

Paper tone fell out of the same conversation

Somewhere in the middle of all this, the conservator pointed at a restored certificate and asked why we had made the paper white. We had not decided to - the restoration had just done it, on the way to cleaning the page. But whether a sheet goes to archival white or keeps its aged cream is a curatorial decision, not a restoration one. Conflating the two meant the tool was quietly making a judgement it had no business making.

So paper tone came out as its own control, separate from text recovery. You can recover every stroke of faded ink and still keep the sheet exactly as aged as it arrived, because those are two different questions and a good tool does not answer the second one on your behalf.

What we took from all of this

The broad lesson outlasted the specific features. A restoration tool for serious records is not judged by how good a single output looks. It is judged by whether the same input reliably produces the same output, whether the reasons are legible, and whether the tool refrains from making curatorial decisions that belong to the institution. A prompt box fails all three. A structured survey passes them, and it was a conservator, not a product meeting, that made us see it.