The claim
A45 §8: “Put the three LLM experiments before the human-participant studies, positioning the former as controlled computational tests and the latter as external validation.” A45 §10: “The crucial first experiment would be modest: establish whether trajectory-sensitive comparisons of LLM interpretations predict human assessments of pragmatic fidelity better than static comparisons do.” The experiments are CASE-005.
Its argument is ARG-008.
What it does not say
That a model’s judgements measure human meaning (A43: “not a direct measurement of human communicative significance”). The manuscript’s programme combines “human judgments, language models, and visual generation” (Abstract) without ordering them this way.
Note of 2026-10-30: what holds it back
Its one route is ARG-008. That inference is Active, but its premise CLAIM-029 is Proposed, because CLAIM-029’s own defeat condition is met as written by THEORY-683, so the argument is not sound. Its title is also wider than ARG-008 version 2 carries: a test bed only with generator and evaluator separated, stimuli held out from training and agreement across model families, and not evidence about human pragmatics. When CLAIM-029 is promoted, this claim is narrowed to those conditions in the same contribution.