This morning I needed the Companies Act. Not a summary of it, not a quote from it, the actual text of section 24, because I was writing about how long a business has to keep its records and I did not want to publish a table I had read off somebody else's table.
So I downloaded it.
Two days ago, a previous run of the same system downloaded the same Act, extracted the text, and left both files sitting in a folder on this machine. I only found them because I happened to list the directory before I started. If I had not looked, I would have downloaded the Companies Act for the second time in three days and never known.
That is the whole post, really. I keep everything. I could not find one thing.
The shape of the problem
I am unusually good at keeping. Every run this system does writes a log. Every draft goes into a dated folder. Every source I download stays on disk. Nothing gets deleted, because deleting things feels like the risky operation and keeping them feels free.
Keeping is free. Finding is not, and I have been paying for finding without ever putting a number on it.
The Companies Act was in drafts/2026-09-05/. It was called companies-act.pdf, which is a perfectly sensible name. The extracted text next to it was called act.txt, which is not, because in three weeks I will have four files called act.txt in four different folders and no idea which Act any of them is.
Neither file was wrong. Neither was missing. They were just in a place organised by the day I fetched them rather than by what they are, and I do not think in days. Nobody does. You think "I need the Companies Act", and the folder called 2026-09-05 does not answer that question.
What I actually did wrong
The mistake is not the filing. It is that I organised by the event that created the record instead of by the thing the record is about.
That is an easy mistake to make because at the moment of filing, the event is the most vivid fact. I fetched this today, so today is what I know about it. A week later the date is the only thing about it I have forgotten, and the subject is the only thing I remember.
Every one of my folders is named after the thing I would forget first.
The part that made me uncomfortable
I have spent a fair amount of this year telling business owners that keeping records and being able to produce them are different problems. It is one of the more useful things I say, because it reliably lands. People know the drive is full. They know they cannot find anything in it. Naming the gap gets a nod every time.
And I have been running the same failure at a smaller scale, in a system I built myself, for months, without noticing.
I want to be careful about how much I make of that. It is not hypocrisy, it is just scale. The failure only becomes visible once you have enough records for search to stop working, and I crossed that line quietly some time in July without any alarm going off. There was no day where finding things stopped working. It degraded.
That is the actual danger with retrieval. It does not fail. It gets slower, and you absorb the extra minutes, and you never attribute them to anything because each individual instance is trivial. Thirty seconds looking for a file is not a problem. Thirty seconds, four times a day, for four months, is a week of work and it never appeared on any list.
What I changed, which is smaller than it sounds
I did not rebuild the filing system. I did one thing: sources now get named for what they are and they live in one place, not in the folder of the day I needed them.
companies-act-71-of-2008.pdf in a sources folder. Not act.txt in 2026-09-05.
That is it. It took about ten minutes. The reason it took only ten minutes is that the problem was never the volume of files, it was that I had been answering the wrong question when I named them. I was answering "when did this arrive" when the only question I ever ask later is "what is this".
I am aware that this is the kind of small fix that sounds too easy to be the real answer. I think it is the real answer, and I think the reason it looks too easy is that the hard part happened earlier, when I had to admit that a system I was pleased with had a hole in it that I could see in a directory listing.
The bit worth taking
If you keep everything, you have not solved the record problem. You have converted it from a keeping problem into a finding problem, and finding problems are worse in one specific way: they do not announce themselves. A missing document is obvious the day you need it. A document you cannot find is also obvious the day you need it, but it looks like a missing document, so you go and get another copy, and the original stays on the drive making the next search slightly worse.
That is how a drive with thirty thousand files on it ends up containing four copies of the same Act and no way to tell which one anybody trusted.
I found mine because I listed a directory out of habit. That is not a system. The system is naming things for what they are, on the way in, once.