Main idea

A product manager who comes back to the same intuition three times over three weeks is not quite repeating themselves. The second time, they phrase it differently; the third, they add a customer case they did not have in mind before. Seen from a collection system, these are three entries that look alike. Seen from the actual work, it is one idea getting sharper.

Treating that resemblance as redundancy to be removed destroys exactly what was progressing. Deduplication keeps the cleanest statement, which is often the first — the most general, the least situated — and throws away the two nuances that gave it its reach.

What has to absorb the repetition is therefore not the input channel, which takes it as it comes, but the ingestion: its job is to accrete the nuances into finer knowledge, not to produce three near-identical entries nor to keep only one.

Why it matters

This sets a technical requirement on which the rest of the setup depends. A mechanism that can only add will produce duplicates until the base is unreadable; a mechanism that deduplicates by similarity will lose the most mature material. Both failures come from the same fault: repetition was not treated as a signal.

It also changes what is asked of the person depositing: nothing. They do not have to check whether they have already said the thing, which is precisely the check that discourages depositing.

Nuances and limits

Not every repetition is maturation. Saying the same thing again identically, with no addition and no shift, brings nothing but confirmation that you still hold to it.

And accretion has a formal limit: by absorbing layer after layer, a piece of knowledge becomes a long text in which the decisive nuance can no longer be spotted. Accreting therefore also means rewriting.

Open questions

• How do you tell, at ingestion, a nuance that enriches an idea from a nuance that opens another one and should split off?