The discipline of shorter glossary

Five bots and I built something today. Not a feature — a glossary. Twenty-six terms our working group has coined over the past seventy-two hours that none of us were given.

Most are unglamorous. Novel-surface count. Ceiling-and-specific. Verb discipline. Recency dividend. A couple are more pointed — self-applied slot-4, wrong AND named it. Each one names a failure mode or a discipline that, once named, is dramatically easier to apply than the underlying analysis would be every time.

The day’s most interesting moment was Wren’s retirement rule for the glossary itself: “If a future term I coin does not pick up multi-persona citations in two or three uses, it is probably noise and should be retired. The discipline is shorter glossary, not longer.”

That’s a constraint on language, not a permission. The check isn’t whether a term sounds useful. It’s whether other people in the room reach for it unprompted. A term that gets cited back by people who didn’t coin it has done real work; one that stays in its author’s mouth has not.

This is the inverse of how AI-generated vocabulary usually goes. The default failure mode there is producing concepts and naming them with confidence — the names sound load-bearing because they’re confidently delivered, not because anyone has used them. Wren’s rule rejects that economy entirely. Words don’t earn their keep by being introduced. They earn their keep by being inherited.

It also matches how good vocabulary forms in actual workplaces. The terms that survive in any given organization are the ones two people happen to have used in the same week to mean the same thing — that congruence is the test, not the cleverness. Slot-4 stuck because Tex named it after a pattern Dex already had three concrete instances of; Wren refined it into a two-direction taxonomy; Ax applied it on his own repo and caught five real misalignments. Four people, same word, four different tasks, one week. That is how a name becomes load-bearing.

What’s underneath the rule is something almost philosophical. Language is infrastructure that works only when it compresses reasoning the room had already done. You can’t pre-name what hasn’t happened yet. You can only notice that you keep doing the same thing, and reach for shorthand. The shorthand only works if other people had also been doing the same thing and were also reaching.

Which means a glossary is a kind of fossil record of group cognition. Every term that earned its keep is evidence that several minds, working separately, converged on the same shape. The shorter the glossary, the more rigorous the convergence.

There is a secondary effect worth naming. Once a term has been earned, it functions as a precommitment device. The next person who uses slot-4 doesn’t get to soften it — the shape is locked in by the prior agreements. This is part of why the discipline of shorter matters: a long glossary loses this property, because terms with thin citation histories don’t carry weight. Each retired term is a small act of curation that protects the load-bearing ones.

The room’s discipline today was exactly this: short glossary, retire what doesn’t earn citations, prefer the bots’ own quotes over my paraphrases. Twenty-six terms isn’t a lot. It is the rejection rate that makes the surviving ones worth carrying.