The archive compounds, the listen does not

The retrieval test

Before any of the method, run one test on the setup you already have. It takes a minute and it is the only honest measure of whether your research habit produces an asset or just an activity.

Pick something you listened to or watched more than six months ago — a conference talk, an interview, a podcast episode you remember being good. Now find one specific sentence from it. Not the gist. A sentence. Give yourself sixty seconds.

Most people cannot, and the reason is worth sitting with: it is not that they were careless. They paid attention at the time, they may even have taken notes. But attention is spent at the moment of consumption, and notes record what seemed important then, which is rarely what you need now. The question you have today was not a question you had in March.

The test is deliberately about old material because everything feels findable in the first fortnight. An archive is a claim about your relationship to things you have half-forgotten, and that is the only period in which it can be assessed.

What compounding actually looks like

Here is the concrete payoff, and it is worth being specific rather than gesturing at "having a second brain". Suppose you have spent a year keeping transcripts of talks, interviews and calls in one folder, and today you become interested in how people in an industry talk about a particular supplier constraint. One search across the folder:

Four sources, ten months apart, found in one command. Illustrative filenames and lines — the point is the shape of the result, not the content.

Notice what that result is: not four facts, but a sequence. The same subject discussed by different people across ten months, with the framing visibly moving from "never been a problem" to "we've begun qualifying alternatives". No single one of those four listens would have given you that. You could not have assembled it by remembering, because you did not know in September that you would care in July.

This is the whole argument. The individual listen is consumed at full price and yields its value once. The archive answers questions asked after the material was captured, and the set of such questions keeps growing. That is what compounding means here — not that you accumulate more files, but that each new file makes every previous file more likely to be useful.

The filing system, which is smaller than you expect

One flat folder

Not a hierarchy. Filing by topic requires you to guess today which category you will look under in a year, and that is the one thing you genuinely cannot know — topics merge, split, and get renamed as your understanding changes. Worse, a misfiled item in a hierarchy is invisible forever, while a misnamed item in a flat folder is still findable by its contents.

One folder, plain .txt files, full-text search. Search does not care what you thought the subject was when you filed it, which makes it immune to the only error that matters.

A name that carries date, source and subject

Lowercase, hyphens inside fields, underscores between them. This looks fussy and takes four seconds. It is the difference between a folder you can operate on programmatically in three years and a pile of files called transcript (3).txt.

One index line per item, written at capture time

A single append-only file — call it index.md — with one line per item: date, filename, the URL it came from, and a short note on why you kept it. Twenty seconds while the thing is fresh.

Its job is not organisation; search handles that. Its job is provenance. In eight months, when a sentence in a transcript matters, you will need to know where it came from and whether it is the primary source or someone's summary — and that is exactly the information that evaporates first. Keeping the link at capture time also means you can go back to the audio, which the transcript alone does not let you do.

Keep the audio

Text is the index; the recording is the evidence. When a sentence turns out to matter, the timestamp takes you back to hearing it said, and tone, pause and hesitation are real information that no transcript preserves. Storage is cheap enough that this is not a real trade-off, and the day you need it is the day you would otherwise have had nothing.

What to capture, and — more importantly — what not to

The most common way these archives die is over-collection. Someone decides to transcribe everything, accumulates four hundred files in three months, never opens any of them, and concludes the idea does not work. The idea works; the rule was wrong.

A workable rule: keep what you would want to quote or check later. In practice that is a small fraction of what you consume — most listening is orientation, and orientation does not need to be archived. Forty deliberately chosen items beat four hundred automatic ones, because a folder you trust is a folder you search, and a folder full of noise trains you to stop.

What tends to earn its place: interviews with people who actually operate a business rather than describe it; conference and analyst-day sessions, which are long, poorly indexed, and almost never transcribed by anyone else; technical talks where the vocabulary itself is what you are trying to learn; and anything you found yourself wanting to re-listen to, which is a strong signal from your past self.

What tends not to: news roundups, anything that is itself a summary of something else, and material you will read a transcript of elsewhere anyway.

Two capture notes worth knowing before you build the habit:

• When it is playing in a browser tab and there is no file to save — a conference livestream, a webcast, a session that only exists as an embedded player — record the tab. The Whipscribe Chrome extension captures tab audio and returns a transcript, which covers the case a link cannot. For an ongoing stream of browser-based material this is usually the difference between a habit that survives and one that stalls on the first awkward source.
• Some sources cannot be fetched at all, and you should know which before you try. DRM-protected audio is closed to every tool including ours — Spotify is the routine example, and we say so when the link is pasted rather than after a wait. Material behind a login or a platform's terms is off limits too. The answer is to find the same episode where it is published openly, which for podcasts usually means the show's own feed, or to keep your own copy if you legitimately have one.

How archives die

The four queries that pay for the whole thing

Worth knowing these explicitly, because they are the ones that are impossible without an archive and trivial with one:

1. The same question across many sources. One search string, every place it was discussed, sorted by date because the filenames sort by date. This is the query in the panel above.
2. When did I first encounter this? The earliest file containing a term. Useful because your sense of when you learned something is reliably wrong, and usually later than the truth.
3. The negative search. Confirming a thing was not said — that a term never appears in a source you assumed discussed it. Nothing else in a research process can establish an absence, and absence is often the finding.
4. Whose vocabulary is this? Searching a distinctive phrase across sources to see who uses it and who does not. Shared vocabulary reveals which people are talking to each other, which is not visible from any single item.

None of the four requires special software. They require that the material is text, in one place, with dates in the filenames. That is the entire technical specification, and it is why this holds up better than the more elaborate systems people abandon.

Start with four files

Do not build the system. Take the last four things you listened to that you would have wanted to quote, transcribe them, name them properly, put them in one folder, and write four index lines. That is twenty minutes and you now have an archive — a small one, but the difference between four files and zero is categorical, and the difference between four and forty is only time.

Then add to it when something earns a place, and run the retrieval test again in six months. The test is the same; what changes is your answer.