The case
Four thesis-level defeat conditions in the pragmatic-transport line state one test.
- CLAIM-115: “predicted as well by propositional or embedding similarity as by measured preservation of pragmatic observables (stance, footing, alignment, expected uptake)”.
- CLAIM-139: “predicted as well by topic-factor, embedding or surface similarity as by independently measured relational profiles”.
- CLAIM-005: “Embedding or propositional similarity between source and rendering predicts held-out human communicative-fidelity judgements as well as the context-indexed distribution distance does.”
- CLAIM-078: “Structural and decision-theoretic measures add no prediction of communicative-identity changes beyond text similarity”.
So the test has three parts.
- The criterion: human judgements of communicative fidelity, or of shared kind, for a set of renderings.
- The predictor: the theory’s measure, mostly human judgements of stance, footing and the like for the same renderings (CASE-040’s interpretive judgements are of “tone, stance, solidarity, and authority”).
- The rival: a machine score of propositional, embedding, topic or surface similarity.
The theses survive if the predictor beats the rival.
What it can show
That judged pragmatic observables predict fidelity judgements better than text similarity does. Failing that would defeat the theses.
What it cannot show
It has not been run. As stated, its rival is one that no account the record names holds (CLAIM-233), and its predictor and criterion share a method that the rival does not (CLAIM-232). So a pass does not separate the theses from functional, relevance-theoretic or speech-act accounts. That is ARG-026, Rejected. CASE-055 is the repair: a behavioural criterion, predictors rated by other raters, and rivals that hold static stance or proposition with force.
Filed 2026-10-28 by the ADR-061 migration, as the premise CLAIM-232’s and CLAIM-233’s argument needed.
Note of 2026-10-29: the conditions as revised
The quotations above are the conditions as stated when this case was filed. At the seventh status review CLAIM-115, CLAIM-139 and CLAIM-005 were revised for CLAIM-238: each test now fixes its observables, their weights, an equivalence margin and a power before the data, and CLAIM-005’s distance is computed with one outcome correspondence fixed across a class of items and scored on held-out ones (CLAIM-207). That repairs the null with no margin. It does not repair what this case says it cannot show: the rival is still one no account the record names holds (CLAIM-233), and predictor and criterion still share a method (CLAIM-232). CASE-055 remains the repair for those.