Four folders, one structure
Four folders is enough
A common mistake in setting up a knowledge store is to design an elaborate taxonomy on day one. The taxonomy is wrong by month two, the worker stops trusting it by month three, and the system is abandoned by month six. The minimum viable structure is much simpler. Four folders, each answering a different question. Ideas hold the things that might one day matter. Projects hold the things actively being shipped. Tasks hold the discrete pieces of work with finish lines. People hold the relationships and what is owed in each direction. Almost every fragment captured fits one of these four; the few that do not are usually reference material that belongs in a fifth, archival folder.
Each folder answers a different question
The folders are not arbitrary categories but functional ones. Ideas answer the question what might be worth doing. Projects answer the question what is being actively pushed forward. Tasks answer the question what specifically needs to happen next, with a deadline. People answer the question what is the state of each working relationship. A folder that cannot be summarised in a question that specific is usually a sign of category drift, and the easiest cure is to merge or remove it. The taxonomy works because each folder has a particular job, and the items inside each folder share that job.
Linking is what makes the structure useful
Folders alone are organisation; the structure becomes useful when items link to one another. A project links to the people involved and the tasks beneath it. A task links to its parent project and to any reference material. A person links to the projects they are part of and the conversations that produced commitments. The links are not for aesthetic completeness. They are what allows an assistant asked to brief the worker on a project to gather, in one pass, the relevant tasks, the relevant people, and the recent conversations — without the worker having to remember where any of it lives.
The machine-readable layer
A small additional discipline turns the structure from useful to genuinely scriptable. Each file carries a short header — typically two or three lines at the top — that names the type of item, the relevant date, and any tags. The header is plain text, in a format anyone can read, and exactly the format an assistant can parse without ambiguity. With this in place, a single command can produce a daily briefing by reading every task with a due date in the next forty-eight hours, every project marked active, and every person who has been waiting on a reply for more than three days. None of this is exotic. It is what happens when a structure is designed to be read both by a human and by a machine.
Open whatever you currently use as a notes store. Of the items inside, how many could you assign to one of: idea, project, task, or person — and which would refuse to fit?