Source
LIT-810, LIT-338. Filed by the promotion audit’s narrowing-loss sweep (batch 066, ADR-047 v4) on 2026-10-20, to carry the wider claim that the narrowing of THEORY-163 dropped. THEORY-163 is Active and is this THEORY’s evidence; this one stays Proposed.
What was actually shown
Treating a colour lexicon as an encoder from Gaussian perceptual meanings to words, the authors fit each World Color Survey language and English to the information-bottleneck curve: all lie near it, at β just above 1, nearer than hue-rotated variants of themselves (93% beat all 39), and soft, inconsistent naming is what IB optima look like at finite β. THEORY-163 states this, with the caveat that the need distribution that makes the fit best was estimated from the naming data cross-validated over languages. The authors write of colour naming across languages in general, and of its evolution as a path along the bound.
What would settle it
The promote_when is inherited from THEORY-163’s before it was narrowed, and QUESTION-214 asks the same question.
What this does not say
It does not carry the authors’ further reading that languages are driven to the bound by a pressure for efficient compression: the paper gives no process that would take a language there, and its “evolution” is a path through optima, not a dynamics (NOTE-631). That reading would be a separate thesis. Rotations are one family of alternatives only.