A phrase overheard on the tram. The way a man folds his newspaper before delivering bad news. A name that sounds like a whole biography. Fragments are the raw material of writing, and they share one property: at the moment of capture, you cannot know what they are for. Any system that demands filing at capture time is asking a question the fragment cannot answer yet.
Fragments resist filing by nature
A folder called ‘dialogue ideas’ assumes the fragment is dialogue. But the overheard line might become a character's tic, a chapter title, or the seed of an essay three years out. Filing forces a premature decision about future use—and the friction of that decision is exactly why most fragments die unrecorded, in the seconds between noticing and finding a pen.
Capture verbatim, at the speed of noticing
The value of a fragment is its exact wording, its specific gesture—summarize it and it dies. Voice keeps the rhythm of a sentence as you heard it; the camera keeps the peeling poster; text keeps the precise line. One fragment per note, unpolished.
- Never clean up a fragment while capturing it—the roughness is the material.
- Speak the sensory detail you will otherwise lose: the accent, the weather, the smell of the stairwell.
- New fragment, new note—stacked fragments bury each other.
One breadcrumb of context, not a taxonomy
Add a clause, not a category: where you were, who spoke, what it might touch if you already know. ‘Woman at the pharmacy, to no one in particular.’ That is not organizing—it is leaving a scent trail for a future search, and it costs three seconds.
Retrieval by resonance
Months later you will not remember the fragment; you will remember a mood. Search by meaning works at that level: ask for ‘notes about old hands’ or ‘things people say when they are lying politely’ and fragments surface across voice, photo, and text—captured in different years, never filed anywhere. Two unrelated fragments landing in the same result is how scenes get born.
Searching for a mood rather than a word
The queries that work for fiction look nothing like the ones that work for admin, because you are searching for a quality rather than a fact. Ask by sensation—‘notes about waiting rooms’, ‘people being polite while furious’, ‘things that smell like childhood’—and let the overlap do the work.
Two habits make this more productive. Search when you are stuck rather than when you are collecting, because a query written while you need something specific is far sharper than one written out of curiosity. And read past the first result: with fragments, the value is usually in the third and seventh items, which arrived from unrelated years and have no business being next to each other.
The uncomfortable part: most of it is not good
Capturing without judgment means capturing a great deal that is mediocre, and the archive will not sort it for you. A search for a mood returns forty fragments of which perhaps three are alive, and there is no metadata that distinguishes them—liveliness is not a property a system can index.
This is a real cost and not a solvable one. The alternative, filtering at capture time, is worse: the judgment you would apply in the moment of noticing is unreliable, and the fragments that turn out to matter are frequently the ones that seemed pointless when you wrote them down. Collecting badly is the price of collecting at all.
There is also a failure mode specific to writers, which is collecting instead of writing. An archive of ten thousand fragments can feel like productive work while functioning as an elaborate way to not start. The fragments are raw material; nothing about capturing them well makes the actual sentence any easier to write, and no retrieval system has ever finished a chapter.
File nothing, keep everything, verbatim and dated. The fragment's job is to survive until it finds its story—and search by meaning, not your filing discipline, is what makes that reunion possible.