Named For The Moment It Was Created

File names that made sense on the day they were written and are unsearchable a month later.

Pattern

Files accumulate with names like notes, final, v2, analysis_new, or a bare surname. Each was obvious when written.

Ten files are searchable, 50 are annoying, 300 are a problem. A month later the folder is unnavigable and the only way to find anything is to open files one at a time.

Why it looks right

The name is written at the moment of maximum context, when nothing about it is ambiguous. Spending thought on a naming convention feels like procrastination against doing the actual work.

It also degrades gradually. Ten files are fine, fifty are annoying, three hundred are a search problem — and by then renaming is a project.

Why it fails

A file name is the only metadata guaranteed to survive every move, copy, sync and export. It is the one thing a search sees without opening anything.

Names that encode the moment — final, new, v2 — carry no information a month later, because every file was final and new once. Names that encode only a subject with no date cannot be ordered. And inconsistent names defeat prefix matching, which is the cheapest retrieval an agent has.

For an agent this is worse than for a person. The model picks files by name, and writes new files patterned on the names it sees, so an inconsistent folder produces increasingly inconsistent output.

Instead

Decide the convention before the first file, and make it mechanical enough to apply without thought:

YYMMDD_subject_variant · date first for chronological sort · no diacritics in identifiers · version by changelog, not by filename

Two rules do most of the work. Never encode status in the name — final and v2 are what version history is for. And keep one convention per surface rather than one per person, because a convention that varies by author is a convention that cannot be searched.