The claim
A78: “This opens a novel research question: Can communicative reconstruction preserve most local pragmatic judgments while substantially changing their global compatibility structure? That would be much stronger than showing that a telephone-game chain progressively changes its vocabulary or apparent speaker stance.” And: “Translation may preserve local meanings while changing the global structure of meaning. That is a nontrivial, formally expressible distinction—and potentially the most interesting connection we’ve found between pragmatic translation and the sheaf-theoretic contextuality literature.”
Its measures: D_sheaf(T) = Σ_C w_C d_C(T_C# e_i(C), e_(i+1)(τ(C))), “a graded, directed measure of preservation of local observational structure”, which is the form of the manuscript’s L_obs (CLAIM-005); and Δ_CF = CF(e_(i+1)) − CF(e_i), the change in contextual fraction, “for measurement scenarios to which the measure applies” (Abramsky, Barbosa and Mansfield; LIT-265 in the record, unread).
Manuscript Abstract: “The theory predicts when local content can survive while the global pattern of communicative relationships changes.” §11: “A transport may preserve overlaps but change the available measurement cover.”
Proposal v5 (U30) kept it: A81 §5.5, A82 §8.7 (“Distinguish increased ambiguity from an increase in formal contextuality”), and A84’s planned Appendix F example, “Two locally similar empirical models with different global compatibility”. A81 §4.4 set the contextual fraction’s role: “an observable structural descriptor rather than … the definition of translation fidelity”.
What was lost
Δ_CF and the worked example of two locally similar models with different global compatibility. The chunk-5 reader finds both dropped at C6 when Appendix F replaced the finite examples with a protocol, without critique. The manuscript’s Proposition 2 says when a global extension is preserved; nothing in it shows a case where it is not.
C7 Appendix B states the limit a drift result would meet here: “A result about state distributions is not automatically a result about a discontinuous contextuality measure, whose stability requires separate assumptions.” The contextual fraction is such a measure.
What the reading of the contextual fraction implies
The contextual fraction is now read (LIT-265, NOTE-236). Three consequences for Δ_CF:
- It is undefined at some steps. Δ_CF is undefined at any step that makes the marginals depend on context, because the contextual fraction is defined only without signalling.
- It cannot rise along free operations. By the paper’s Theorem 2, a free operation (relabelling, restriction or translation of measurements, coarse-graining, mixing with a noncontextual model) cannot raise the fraction. So Δ_CF ≤ 0 along such steps, and an increase needs a step of some other kind.
- It does not jump within a scenario. The premise quoted here, that the fraction is discontinuous, does not hold within a fixed scenario: it is Lipschitz in the probability table (THEORY-165, a derivation from the paper’s linear programme). Only a change of scenario can make it jump.
Note of 2026-10-10: A173’s request, and the lost pair recovered
A173 asked the entry to acknowledge that classical transports preserve noncontextuality. It already does (Theorem 2 of LIT-265; THEORY-174). A173’s “not merely to transport itself” holds only for increases. Coarse-graining or mixing can lower contextuality by transport alone. The lost worked pair is CASE-037 with one context changed: two models that agree on two of three contexts and on every marginal, with contextual fraction 1 and 0. The headline thesis is still absent from v7 (A178), from the second crystallized argument (A203) and from the October outline (A218).
Note of 2026-10-10: a documented case
CLAIM-201, this headline’s narrowing to a change of cover, now has a documented case, CASE-072: Hafez’s “Shirazi Turk”, whose ungendered beloved four English translators render as a maid, as God and as a boy. It shows a rendering deciding what the source left open, on the question of who is addressed. It supplies nothing for this entry’s own form, a change of global compatibility at small local distortion within a fixed scenario: it has no contextual data and no measured distortion, and its evidence is translators’ and scholars’, not readers’ agreement.
Note of 2026-10-10: a documented case
CASE-053, found by a search the owner asked for, collects the documented tests of contextuality on elicited human judgements. It supplies the evidence that this headline’s fixed-scenario form still lacks a case: no study of pragmatic or meaning judgements finds contextuality proper at an adequate sample size. LIT-264 and Wang’s thesis (2024) find none; Lo, Sadrzadeh and Mansfield’s Winograd schemas (2023), the one test with pragmatic content, find a violation of about one standard deviation, with no-signalling forced by the response format. The positive results that are reliable, Cervantes and Dzhafarov’s “Snow Queen” and its follow-up, have instructed matching as content, not pragmatics, and are disputed. It does not supply a contextual model for this entry, and it does not show that none exists: what it documents is that the adequate test has not been run. A powered rerun of Lo et al.’s design, under CbD, is what would decide the fixed-scenario form, and could be the contextuality arm of CASE-055.
Note of 2026-10-15: split under ADR-043
ADR-043’s full classification found this claim compound: that a reconstruction can keep most local judgements and change their global compatibility is shown possible by a constructed pair, CASE-037 with one context changed, which is demonstrated, while that documented reconstruction chains do this, which is what the defeat condition tests, has no observation under it, so it is interpretive. The constructed possibility is now CLAIM-320 (demonstrated), on which this claim rests. This claim keeps the thesis about translation and reconstruction chains, and its measures. Its title states that thesis and is kept; its warrant stays interpretive.
Note of 2026-10-19: the title restated as prevalence
The promotion audit’s batch 034 found the title behind its own defeat condition. The condition tests a rate over documented reconstruction chains, after CLAIM-238, and the note above moved the constructed possibility to CLAIM-320. The title still said translation “may” do this and reconstruction “can”, which CLAIM-320’s pair already settles. It now states the thesis this entry keeps, about documented chains. The manuscript’s wording quoted above is unchanged. The warrant stays interpretive: CASE-053 documents that the adequate test has not been run.
Note of 2026-10-29: what holds it back
Its defeat condition counts changes beyond sampling error, which needs a signalling estimator with bias correction and a stated sampling error, and CLAIM-235 finds none in the line. CASE-053 documents that no adequately powered CbD test on pragmatic judgements has been run.