Main idea

A glossary doesn't sit on the same floor as the other pieces of a knowledge system. It doesn't replace atomic notes — it supplies the words they use to assert. It doesn't replace thematic notes — it makes their syntheses readable to someone who didn't write the notes they assemble. It doesn't replace sources — it helps translate their concepts into the language of the project, when an article or a book uses a word in a sense that isn't the team's.

So it is a light infrastructure: it produces no new knowledge, it makes other people's usable. Its value isn't measured by what it adds to the base, but by what stops being fragile elsewhere in the base once it exists.

That position explains why it is judged badly. A layer that produces nothing always looks removable, and its absence never shows up directly: it shows up in links that carry nothing and in syntheses nobody rereads.

Why it matters

This stops the glossary being treated as a documentation project competing with the notes, to be arbitrated against them in hours spent. It only takes time in proportion to the words that actually carry a decision.

It also says how to defend it: not by the number of entries written, but by what it lets the other layers do.

Nuances and limits

The infrastructure has its own upkeep cost: a wrong or outdated entry is worse than a missing one, because the notes that use it inherit its error without seeing it.

And the role flips if the glossary starts carrying theses: it becomes a layer of content again, with all the update obligations that come with it.

Open questions

• How do you measure what an infrastructure layer prevents, when its effect can only be read in what fails to happen?