The case
A design, stipulated by the record on 2026-10-10 in reply to the audit’s objections to the central tests. It is in the line of CASE-005, CASE-021 and CASE-040, which are also proposed designs, and of CASE-005’s Experiment 2 in particular, whose three-way selection it restores. It is the study the revised theses CLAIM-220, CLAIM-202 and CLAIM-204 name in their defeat conditions. Every margin, power and rate it needs is to be fixed in its preregistration, by the owner; none is fixed here.
Stage 0: measurement (CLAIM-227). Before any fidelity test:
- definitions of stance and footing given to the judges in writing;
- a fixed response format;
- reliability between judges and between two passes by the same judge, against a target fixed in advance;
- known-groups validation: renderings that differ only in the attributed speaker (CASE-018’s reattributions used as stimuli) must be rated as differing in footing, and renderings that differ only in wording must not.
Its results bear on QUESTION-076.
Source items. Two kinds.
- CASE-040’s topic × act cells (spending or relationships, teasing or reprimand), written by item writers.
- Naturally occurring utterances of teasing, reproach and reprimand, drawn by a stated sampling rule, with their act labelled by raters who did not select them.
Rendering arms, for each source item.
- (a) Proposition kept, footing changed: CASE-027’s kind, the sociologist’s Henley.
- (b) Proposition changed, footing kept: the necessity half of “neither necessary nor sufficient” (CLAIM-229).
- (c) Static stance and footing matched to the source, response to later framings different: the renderings CLAIM-202 needs.
- (d) A professional translation under a stated brief.
- English intralingual retellings, for comparison with the interlingual arm.
Languages (CLAIM-222). English sources. Targets: French or German, where the T/V choice is obligatory, and Japanese, where honorific marking is. In each, footing is grammatically marked on the target side only, so a rendering must choose a footing the English leaves open. Renderings by professional translators under stated briefs.
Receivers (CLAIM-226). Named populations of native readers of each language, recruited by a stated sampling frame. Source-language readers supply the source terms; target-language readers the target terms. Measurement invariance across the two communities is tested on anchor items whose act is fixed by convention (a formal apology, a greeting to a superior), so that a difference between populations is estimated and not read as distortion by transport. For historical sources (Villon, Henley, Marot) a separate arm, whose source term is scoped as fidelity to a stated modern reading, since there is no source audience to ask.
Criterion (CLAIM-232, CLAIM-244). Behavioural and task-indexed: the receiver’s choice of act (which of a stated set of acts the rendering performs) and choice among offered continuations, under stated tasks. Holistic fidelity ratings are collected as a secondary outcome, for CLAIM-240.
Predictors. All human-rated predictors are rated by raters other than those who supply the criterion.
- Human-rated: propositional sameness; sameness of proposition and illocutionary type; the static stance and footing profile.
- Machine: embedding similarity, surface similarity, topic factors, a supervised text model trained to predict act labels.
- Theory: class-level observational distortion (CLAIM-228), the order effects of sequence contexts (CLAIM-231), the directed decision term (CLAIM-259), and the change of cover (CLAIM-257).
Analysis. Judge and group random effects, so that disagreement between judges is modelled and not averaged away. Predictions scored on held-out renderings. Equivalence tests with margins and power fixed in a preregistration, which also fixes the observables and their weights.
What it can show
Whether the central theses survive rivals that hold the pragmatic point: the proposition with its force, and the static pragmatic profile. Whether the transport structure adds to the static profile. Whether any of it holds across a language pair in which footing must be chosen.
Its standing is stipulated, so as yet it shows nothing. Run, it could not show the ontological clause of the October thesis (CLAIM-150), nor which distinctions a translation ought to keep (CLAIM-252), nor anything about language pairs it does not run.
Note of 2026-10-10: the interlocutor’s reading of the design
At U61 the owner asked for “an updated proposal that takes PR#121 into account”. The assistant’s revised proposal (A232 §8) says its central experiment “follows the design filed in PR #121” as this case, and reproduces it, Stage 0 included, so nothing in the measurement stage is new. It adds three things. None is adopted here.
- Two more items to preregister. A232 Stage 6 lists “exclusion rules” and “the principal language-pair comparison” beside the tasks, observables, weights, margins and power this design already leaves to the preregistration.
- Response to later framing as a receiver outcome. A232 Stage 4 makes “How does the answer change after specified subsequent framing?” a primary outcome, beside the choice of act and of continuation. Here the criterion is the choice of act and of continuation only, and response to later framings enters as a predictor rated by other raters, through arm (c) and CLAIM-202.
- A map of outcomes (A232 §11). In the record’s terms: “Strong success” is CLAIM-220 and CLAIM-202 both surviving, though A232 also asks the transport measures to beat “strong supervised alternatives”, which CLAIM-202’s condition does not; “Partial success” is CLAIM-220 surviving and CLAIM-202 defeated; “Structural-method success only” is signatures that transfer without predicting receivers, the gap CLAIM-135 and CLAIM-129 describe between recovering structure and reaching the act; “Failure” is CLAIM-220’s defeat condition met; and “Ontological limit” is CLAIM-150.
One difference. A232 counts “the observables cannot be measured reliably” as failure. Here a Stage 0 failure is a negative answer to QUESTION-076. No defeat condition of the theses names it, so they would be left untested, not defeated.
The design above is unchanged.