The Thing You Can't Reconstruct
A plain text box that may be worth more than everything else in the product. Every archive has this hole: the artefact survives and the reason evaporates.
Date
Category
Writer
Our studio has a folder called REF. It has about four thousand images in it, accumulated over years, and it is the single most valuable asset we own after the client work itself.
It is also almost useless, and the reason is precise.
We can tell you, for any image in it, the filename, the dimensions, the date it was added, the dominant colours, and with a bit of effort where it came from. What none of us can tell you, for almost any of them, is why it was saved.
Someone saw something. Something in them said keep this. That judgement was made against everything they were thinking about that week: a project, a problem, a conversation, a shape they had been circling for a month. None of that is in the file. None of it is in the metadata. And it was gone within about a day.
Fig. 1. Photograph by Tookapic, CC0, via Wikimedia Commons.
The field
There is one field in JEFIELD that I think is worth more than everything else in it, and it is a plain text box.
When you open something you kept, you can write why you kept it.
That is the whole feature. No template, no prompt, no required format, no character limit, no AI suggestion, no autocomplete. A field, and whatever you put in it.
Why this one is different from everything else
Consider what the system can reconstruct on its own if that field is empty.
Property | Recoverable? | How |
|---|---|---|
Filename, size, format | Yes | Read from the file |
When it arrived | Yes | The event log |
Dimensions, duration, page count | Yes | Parse the bytes |
Text inside the document | Yes | Extraction |
What it depicts | Yes, increasingly | A model, given enough compute |
Similar things | Yes | Comparison across the corpus |
Why you stopped | No | Nothing. It was never in the artefact |
Table 1. Six of seven rows will get cheaper every year. The seventh will not, ever.
That last row is the interesting one and it is worth being precise about why it is permanent rather than merely difficult. It is not that the reason is hard to infer. It is that the information was never in the artefact. It was in you, at that moment, in the relationship between the thing and everything else you were holding in mind.
No model, present or future, however large, can recover information that was never recorded anywhere. This is the one place in the whole product where the usual trajectory does not apply.
Everything else in the record is recoverable and will only get more so. This is the only thing that is not.
How it is held in the model
This required a split I had avoided for a long time, because it doubles the conceptual surface of an item. A trace now carries two different kinds of thing.
That separation is what lets the original stay untouched while the thing accumulates a personal layer. Rename it and the real filename is still there, in the detail view, permanently. Write a reason and the file itself has not changed by a byte.
There is a merge benefit too, which I did not design for and gratefully accepted. Because these are per-field values resolved in canonical order, two devices that both rename and both pin the same item converge correctly with no merge code written for that case.
The bug that told me what it was worth
For a while, the personal note was not searchable.
Search read the log. Search read the document's own text, the page's title and description, the contents of archives, the host of a link. And the one field explicitly designed to hold a person's own words about their own thing was the one field search could not see.
Nobody decided this. It happened because the note was added later, in a different part of the code, and the index was built from the places that already existed when the index was written.
It is a perfect small illustration of how systems encode priorities nobody chose. The machine-extracted text was indexed automatically because it arrived through the pipeline that had an index attached. The human-written text arrived through a different door and got nothing.
I have started looking for this pattern deliberately. Which inputs to your system arrive through the well-served path, and which arrive through a side door? The answer is rarely a decision anybody made and is usually a decent description of what the system actually values.
What it will tell me
This field is the strongest experiment in the project, in both directions, and I have committed to the reading in advance so I cannot reinterpret it later.
If people write in it, and those notes make re-finding work: that is the strongest evidence this product has ever produced. It would mean the thing that makes the field valuable is something only a person can put there, and nothing can be built to compete with it. Not by a better model, not by a better-funded team.
If nobody ever writes one: that is nearly as strong in the other direction. It would mean that at the moment of keeping, people are not willing to spend even five seconds on the one thing that cannot be reconstructed later. Which would tell me something important about what keeping actually is. Not an investment in a future self, but a way of closing a loop in the present.
I do not know which it will be. Both readings are written down, dated, before the result.
The version outside software
Every archive has this hole.
A studio folder full of references with no record of what any of them was for. A camera roll of eleven thousand images, perfectly dated. A saved article nobody can remember saving. A bookmark bar that is an archaeology of past intentions.
The artefact survives beautifully. The reason evaporates almost immediately, and it was the reason that made the artefact worth keeping.
Provenance is usually discussed as a chain of custody: who had it, when, where it came from. Museums and archives do this well. But I have come to think the more valuable provenance is intention: what was the person thinking when they decided this mattered.
Nothing captures that automatically. It has to be volunteered, in the five seconds while it is still available, by the only entity in the world that has it.
We have started writing one line in the filename when something goes into REF. It is ugly, it breaks our naming conventions, and it has already been more useful than four years of the folder without it.
Sources
Photograph by Tookapic, CC0, via Wikimedia Commons. commons.wikimedia.org/wiki/File:Pen-writing-notes-studying.jpg
Related reading
Latest Articles.
© Moc Mallen Studio
Thoughts, ideas, and perspectives on design, simplicity, and creative process.
