The Paper Trail

Test your filing by trying to break it

A filing system’s quality is unobservable while you’re filing. Everything looks fine: things go in, nothing is visibly wrong, the folder count grows. The first real measurement happens when somebody asks for a specific document, and by then the measurement has consequences.

So take the measurement early, on purpose, when nothing depends on the result. It takes about half an hour and it finds problems that years of careful filing will otherwise hide.

The test

Pick a month at random from a year you consider done. Then, without browsing, try to answer five questions:

  1. Find one specific spend over a set amount. Any one. Can you produce the document?
  2. Say what that spend was for. Not what was bought — why.
  3. Reconcile the month. Does every payment in that month have a record, and does every record correspond to a payment?
  4. Find something you know is in there by a word rather than a date. The plumber. The hotel. The replacement part.
  5. Produce everything relating to one vendor across the whole year.

Time each one. Note where you had to browse instead of search, and note anything you couldn’t do at all.

That’s the whole exercise. Its value is that it exercises retrieval, which is the only thing an archive is for and the only thing that never gets practised.

What each failure tells you

Question 1 fails — you can’t find a specific document. Either the file isn’t there or the naming doesn’t support finding it. Establish which, because they are completely different problems: the first is a capture failure and the second is a naming failure. The naming question is the subject of finding it again.

Question 2 fails — you have the document but not the reason. The commonest single failure, and it is a capture-time problem that cannot be fixed retrospectively for records already filed. It tells you the habit needs changing today, not that the archive needs reorganising — writing it down at the time.

Question 3 fails on the payment side — there are payments with no record. Some proportion is normal and expected. What matters is the pattern: if the gaps cluster in one channel, that channel isn’t being captured. Recurring charges are the usual culprit, for the reasons in recurring charges leave the thinnest trail.

Question 3 fails on the record side — there are records with no payment. Usually duplicates, or an order confirmation filed as though it were a receipt, or a claim that was never settled.

Question 4 fails — search by word doesn’t work. The naming scheme is date-only, or inconsistent, or the file is an unsearchable image with no descriptive filename. Very common and very fixable going forward.

Question 5 fails — you can find things one at a time but not as a set. Structural. Usually means the archive is organised by a hierarchy that cuts across the query, so the answer requires visiting many folders. Flat plus good names outperforms hierarchies precisely here.

Read the times, not just the failures

The failures are the obvious output. The timings are the more useful one, because they tell you whether the system will actually get used under pressure.

Under a minute per question means the system works. Anything you can answer in five minutes is fine too — you’d tolerate that once a quarter. The number to watch for is the one where you stopped searching and started browsing, because browsing is unbounded: it takes as long as it takes, it feels like progress the whole time, and it is what turns “find me the invoice” into a lost afternoon.

Note also which questions you didn’t want to attempt. That reluctance is data. It’s usually pointing at the part of the archive you know is bad, which is the part a real request will land in.

Keep or bin

KEEP OR BIN — what the test finds

  · A written note of what failed and how
    long each answer took
                    → KEEP. The only baseline you'll
                      have next year.

  · Files you found by searching a word you
    wrote yourself
                    → KEEP the convention that made
                      that work. It's the thing doing
                      the work.

  · A hierarchy you had to browse to answer
    question 5
                    → BIN, or flatten it. If retrieval
                      needs browsing, the structure is
                      decoration.

  · Records with no matching payment
                    → INVESTIGATE. Usually duplicates
                      or confirmations filed as
                      receipts.

  · An order confirmation you'd been
    treating as the receipt
                    → NOT A RECEIPT. Proves the order,
                      not the payment or final amount.

  · The urge to restructure everything
    afterwards
                    → BIN it. Half-migrated is the
                      worst state an archive can be in.

  · Whether what you produced would satisfy
    anyone who asked
                    → ASK LOCALLY. Sufficiency is set
                      by your tax authority or
                      adviser, not by how tidy it
                      looks.

Fix forward, not backward

The strong temptation after a bad result is to go back and repair the archive. This is usually the wrong call, and it’s worth saying plainly.

Capture-habit failures cannot be repaired backwards at all. A missing reason is missing. The only available action is to change what happens at capture from today — why batch entry fails covers what that change looks like.

Naming failures can be repaired backwards, but only newest-first and only in complete passes. The value of a fix decays with the age of the record, and a partially renamed archive is genuinely worse than a consistently badly-named one, because you can no longer predict which convention any file follows.

Structural failures should be fixed once and never revisited. If the hierarchy is the problem, flatten it in a single sitting. If you can’t finish it, don’t start.

The realistic outcome of a first test is: change one thing about capture, adopt one naming convention going forward, and leave the old material alone except for a newest-first pass if you have the appetite.

Do it once a year

Cheap, and it catches drift. Conventions slip, a new payment channel appears uncaptured, a backup silently stopped — the last of which the test finds if you deliberately answer question 1 from the backup rather than the live copy, as in a synced folder is not a backup.

Pair it with the annual disposal pass in running a disposal review and the whole thing is one afternoon a year covering retrieval, retention and backup verification.

What this doesn’t settle

Whether what you managed to produce is adequate for any purpose, whether the format is acceptable, whether the gaps you found matter, or how long any of it needs to survive. Those are rules that vary by jurisdiction and record type, and they come from your tax authority or adviser.

The test answers a question no rule addresses and no rule can: given that something is in your archive, can you actually get it out.

organisingrecords