The claim
A73: track “two different quantities”, D_local(i) = W_d(q_(i-1), q_i) and D_global(i) = W_d(q_0, q_i), over the distribution of inferred communicative configurations at each generation. Of E4’s two chains (CASE-032): “the first chain preserves a particular higher-order relational pattern while the second undergoes a pragmatic phase change, so to speak, in which the speaker’s relationship to the audience and the subject is reorganized. The exact point of transition could be studied by identifying the first generation at which a different communicative configuration becomes dominant.”
Where it went
It was stated over latents, and went when U28 demoted them (CLAIM-008). It does not need them: the same event can be defined on judged observables. The manuscript’s §8 bounds cumulative error by per-step error (CLAIM-056) and says the trajectories of invariants “need not coincide”, but it has neither the pair of distances nor the transition. CLAIM-031 is the same point made about constraints.
Note of 2026-10-29: what holds it back
No chain the record holds has per-generation judgements of communicative configuration. CASE-032, the one chain with a shift of act, has three generations written in one context, and its per-generation numbers are not in the transcript. So whether the change comes at a marked generation, which is what makes it a phase change and what the defeat condition tests, has no observation either way. CLAIM-056 gives only that surface error can stay small while the act changes.