Main idea

Adding an industrial maintenance standard or textbook to an already established base does not end in a batch of brand-new notes. It touches ten to fifteen notes already written: a glossary definition made more precise, a nuance added to an existing entry, a link created between two concepts that were not speaking to each other, a contradiction flagged with what an earlier source asserted.

The volume produced is low, the volume modified is high — the exact opposite of what an indexer does, stacking documents without ever touching the ones before. That difference, and not the file format, is what separates a consolidated base from a tidy repository.

One consequence lies in the cost structure: ingesting a source does not cost a fixed price, it costs in proportion to what the base already contains.

Why it matters

This gives an observable indicator of a base's condition. An ingestion that produces only creations, on a corpus that ought to overlap with itself, signals that the concepts are not meeting: you are stacking readings instead of confronting them.

It also says what you have to be technically able to do before you start. A setup that only knows how to add will never consolidate, whatever the quality of its extraction.

Nuances and limits

The ratio is reversed at the start: on the first sources everything is creation, and it would be absurd to read that as a failure.

And a high modification rate is not good in itself. Rules that are too broad "touch" many notes without improving any of them, and produce the agitation of consolidation without the consolidation.

Open questions

• Beyond what size of base does the cost of an ingestion stop growing with it — or does it never stop?