Teardownspublic

A file management system

Five conclusions matched a design I'd built separately, and not one of the reasons overlapped. It also named three gaps in mine exactly.

5 agree · 3 I missedSource video — youtu.be/MM-MPS57qKA ↗

The value of this piece is arriving at the same answers for entirely different reasons. A nine-minute video about organising files agreed with my own folder design on five conclusions. And the reasoning does not overlap at all — he arrives via human factors, “make it complicated and people run away”; I arrived via systems, “a path that depends on a taxonomy breaks.” The video also named three gaps in my design precisely. One of them was a number I had picked with nothing behind it.

the skeleton — five levels deep at most, low numbers for what you use often [01] personal · [02] work · [03] reference · [04] temp sharing · [05] backup · [99] archive [99] is always reserved for the archive — the home of things you'll never reuse but can't delete four levels of date naming — with one boundary running through them year · year+quarter findable by search you needn't know the parent folder year+month · full date findable only via the parent folder the finer it gets, the less search helps ⇒ and every file carries one standard keyword — "project + keyword" hits 90% of the time one of five — document, minutes, worksheet, briefing, template — is always in the name
The sharpest part of this system is that how precisely you write the date is itself the boundary of findability.

What the video says

It starts from a balance — between “how tidy is it” and “how findable is it.” The ceilings are five levels deep and ninety-nine folders wide. The reasoning is simple: “past five, we get lost.”

It opens by openly mocking one numbering scheme. It puts an official “two digits, decimal point, two digits” rule on screen and says “memorise this not-at-all-complicated numbering system and… wait, where are you going?” Complicated systems don’t get used is this video’s consistent axis.

One folder is his own invention. “Temp sharing” — a place to isolate the parts you’ve carved out of confidential material to share. A legitimate temporary home for things that belong nowhere.

The first of five tips matters most.

Organise by where you’ll use it, not by where you found it.

The other four: learn your file manager’s search syntax, plant keywords in the description field for shared files you can’t rename, at most five stars (“if everything is starred, nothing is”), and when something is shared with you, choose one of copy, shortcut, or ignore — a shortcut lets you file it into your own system without moving the original.

The conclusion is modest — “there is no perfect file system. Over-optimising actually costs you productivity. Pick one and stick to it.”

What broke — same answers, different reasons

held against my design — it split into three groups five that agree reject numbering · don't move originals cap the index · pick one and hold "organise by where you'll use it" three that diverge depth 5 vs 3 — my reasoning was thin number prefixes — depends on storage tags — my claim was too broad four it never touches directories machines use "is this safe to delete" time-series output · automated checks ⇒ five conclusions agree and not one reason overlaps — his is human factors, mine is systems reaching the same point down different roads raises the odds that the point is right ⚠ and his own screen shows the system failing — the archive holds folders with duplicate numbers
Separate agreement, conflict and gaps, and the question stops being "who is right" and becomes "right under which conditions".

The important conflict is the number prefix. He uses them. I banned them. Both are right — the conditions differ.

He is on ID-based storage. Rename a folder and the links survive. I am on local absolute paths. The moment a number enters a path, the path becomes dependent on the taxonomy. The same technique is harmless on one side and fatal on the other. Folded into a rule: numbers are fine on ID-based storage and banned on path-based storage.

And he is paying for it too. His own screen shows the archive containing folders with duplicate numbers. The numbers have lost their meaning. His system’s failure is captured on his own screen.

I got two things wrong. Depth 3 was a number I picked on instinct, and his 5 has real usage behind it. But the levels were different — his 5 is document depth, my 3 is root-rule depth. Correction: root rules at 3, free below that, 5 as the overall recommended ceiling. And calling tags an anti-pattern was too broad a statement — for things you cannot rename, tags are the only tool available.

Held against my own setup — the half this video can’t cover

Everything this video addresses is a document a human reads. No code repositories, no caches, no lock files, no pipeline output. A large share of my working folders are directories machines use, so this video doesn’t cover half of it.

There is no “is this safe to delete” judgement. There is a backup folder but no concept of something being regenerable. Which is why the archive becomes a graveyard for “things I can’t bring myself to delete” — he says so himself. My design asks “can I rebuild this” first, which blocks that graveyard at the source.

There is no answer for time-series output either. Dates go only in filenames. When results accumulate automatically by the hundred or thousand, a naming convention alone collapses — you have to partition into folders.

And the only enforcement is human willpower. “Pick one and stick to it” is all there is. The result is printed on his screen as disorder. Without a script that checks the rules, rules decay with time.

Verdict

What Verdict
A temporary home for things that belong nowhere adopt. My temp folder is defined to take only regenerable things, so anything unclassifiable had nowhere to go
Cap the index adopt. “If everything is starred, nothing is” — a discipline device my design lacked
Correct the depth ceiling adopt. Root rules at 3, 5 overall. My 3 had nothing behind it
Tags as anti-pattern corrected. They must not replace primary structure, but for things you can’t rename they’re the only tool
Number prefixes still rejected — conditionally. Fine on ID-based storage, banned on path-based
Enforcement by willpower rejected. This goes to automated checks

The lesson I paid for here: that the reasons differed mattered more than that the conclusions agreed. Five matched, and not one reason overlapped. Reach the same point down different roads and the odds that the point is right go up.

And ⚠ a technique that “works for them but not for me” usually has a hidden condition. The number prefix was exactly that. Stopping at “it works for him, not for me” would have produced no rule; asking why it splits produced one — the kind of storage. Find the condition and you get a rule; miss it and you’re left with a preference.