Main idea

The transcript of a user interview, the hypothesis you draw from it, the decision you then make and the brief you generate happily end up in the same file, in the order they appeared. Nothing is lost: it is all there, in a single stream.

What is lost is the status of each sentence. Three weeks later, you no longer know whether a claim is a user's verbatim or an interpretation you made of it, whether a decision still holds or whether another has replaced it, whether a figure comes from a source or from an estimate dropped in passing. The content is intact and its provenance has vanished.

Keeping the sources, the memory of what you retain from them and the outputs in distinct places is therefore not a filing requirement. It is what lets you, over several months, keep saying where what you claim comes from — and it is the only protection against reasoning that feeds on its own outputs without knowing it.

Why it matters

This gives a reason to structure that does not depend on reading comfort. A single stream reads very well; it simply becomes impossible to defend the moment someone asks what a conclusion rests on.

It also changes what you expect from an assistant writing in that space: producing in the right place, not producing in the right format. An interpretation dropped into the sources folder becomes a source for the next session.

Nuances and limits

The boundary is not always sharp. A synthesis note contains rephrasings that are already interpretations, and the requirement to separate does not settle where to put a sentence that does both.

And separation costs at the moment of writing, while it pays off much later — a bad ratio for a subject that will not last.

Open questions

• What do you do with a deliverable that becomes, over time, the only remaining trace of a source you did not keep?