The case
- A35 §6. Three translations and the original; manipulate wording, footing (the attributed speaker: “a fellow reveler, a concerned friend, or a disapproving outsider”), interpretive order and initial state. Compare “a classical state-space model with stochastic updates, a constrained quantum-instrument model, and a general contextual measurement model”. The caveat: “forcing a participant to make an explicit judgment is itself an intervention.”
- A38 §IX. Stimuli to “Include semantically matched but pragmatically divergent versions and pragmatically matched but lexically divergent versions.” Falsifiers include “No reproducible order dependence” and “Dynamical metrics add no predictive value.”
- A40 §XI–XII. A worked case study whose annotations are hypotheses, not facts about the texts; pilot validation, measurement invariance, classical baselines primary. The order-effect statistic Δ_(A,B) = d(P(Y | A→B), P(Y | B→A)), with the confounds “framing effects, memory, learning, task demands, or measurement disturbance”. Falsification criteria (§12.7): no added predictive value over static equivalence; order effects not reproducible; “Correspondence maps fail to generalize beyond their estimation tasks”; process models explain no extra variance; quantum models no better than constrained classical ones.
What it can show
If run, it tests CLAIM-052, CLAIM-117, CLAIM-011 and CLAIM-009. None of it has been run. The manuscript’s §10 keeps the classical-first stance and the separation of context effects from contextuality. The confound list and the held-out evaluation of correspondence maps are not in it.
Outline v3 (A48 §15.5) restated the failure conditions: no stable framing effects across paraphrases; order effects explained by prompt artifacts; no added prediction from trajectories; Φ failing on held-out tasks; TTT giving no transferable behaviour; complex models no better than simple baselines. U19’s instruction to withhold hedging removed the section (CLAIM-095). A49 said the conditions belonged in the methods “as predefined decision rules”, but outline v4 (A52) lists only outcomes.