Topics — What a work is about. Nucleation holds whatever reading the owner files here, whether or not an Anthology of the SOTA topic could also hold it; instruction for ML practice belongs to the anthology (ADR-048). The words are seeded from the clusters the reading-time triage of 2026-09-25 found, and grow by decision.
Psychometrics — measuring minds — scales, reliability and validity, latent-variable and network models of psychological constructs, and model-based estimation of cognitive parameters (ADR-025).
7 of 191 CASE documents. Back to the full index.
| # | Title | Summary | Status |
|---|---|---|---|
| CASE-040 | Topic and act crossed: a factorial design that decorrelates what a text is about from what it does | A214 §8’s design, which A218 names as the first experiment to run: topic (spending, relationships) crossed with act (teasing, reprimand), with lexical, judgement, uptake and intervention data, several factorizations, and four outcomes reported separately. A210 §5 gives a cross-corpus version. Proposed by the assistant; not run. | Active |
| CASE-055 | The discriminating fidelity study: renderings built to dissociate wording, proposition, static stance and transport structure, judged by named populations in two languages on stated tasks | A proposed study, filed on 2026-10-10 as the record’s repair of CLAIM-233, CLAIM-232, CLAIM-244, CLAIM-226, CLAIM-222 and CLAIM-227. It restores CASE-005’s three-way selection with human receivers, and adds a proposition-plus-force rival, a behavioural task-indexed criterion, predictors rated by other raters, named populations, an interlingual arm and a measurement stage. It has not been run, and its margins are not yet fixed. | Active |
| CASE-058 | FactBank’s source-relative factuality labels and Kiesling et al.’s stance-triangle annotation: one stance construct annotated reliably, the interpersonal ones not | FactBank labels one event’s factuality separately for each source that reports it, so a proposition held fixed gets different commitment labels by source, and trained annotators agree at κ = 0.82. Kiesling et al. annotate Du Bois’s stance triangle in Reddit threads, including an investment dimension defined as the split of animator from principal, and reach only α = 0.40 to 0.57, with investment and alignment below 0.4 outside one community. Together they bracket an answer to QUESTION-076: commitment relative to a source is annotatable when defined by tests, interpersonal stance only moderately. | Active |
| CASE-062 | Simplified texts for restricted and for full readers: a gain and a loss documented on different texts, documented nulls and a reversal, and a design that tests both on the same renderings | A constructed study, built on documented halves. Simplified texts, made from the source alone, raised comprehension for named restricted readers (adults and adolescents with autism; weaker second-graders), and automatic simplifications lowered it for fluent native readers, but in different studies on different texts. Edited subtitles did not help deaf viewers, and text reduction in captions helped the more proficient readers more. The stipulated design puts both populations on the same renderings, with probe items fixed in advance as answerable from both versions or from the original only. | Active |
| CASE-067 | The reception of Deborah Smith’s The Vegetarian: the Booker judges’ task, the Korean critics’ task, and one critic’s two verdicts | Deborah Smith’s English version of Han Kang’s 채식주의자 won the 2016 Man Booker International Prize, while Korean critics published lists of errors and charged it with embellishment. Each side’s task is stated in print: the judges evaluate “the finished English-language work on their own terms”, while assessment in Korea “does usually involve comparison”. Charse Yun perceives what both sides perceive and gives two verdicts by changing task. Part of the split is access, not weight: some errors are invisible to readers without Korean. | Active |
| CASE-184 | The central defeat test as stated: human judgements of fidelity or shared kind, predicted by the theory’s judged measures against a text-similarity baseline | The test that the defeat conditions of CLAIM-115, CLAIM-139, CLAIM-005 and CLAIM-078 share, filed as one design so that an argument can start from it. The criterion is human judgements of fidelity or of shared kind. The theory’s predictor is a measure of judged pragmatic observables. The rival is a text-similarity score. It has not been run, and CASE-055 is the design the record filed to repair it. | Active |
| CASE-185 | A proposed study: direct and relay translations of one set of source utterances through three languages, to find whether agreement between the routes, the task-relative loss on each route, and the preservation of response to later framing come apart, and whether correspondences fitted between the communities’ response structures on some items predict receivers’ judgements on held-out items, translators and situations | Put together from A4 §6, A5 §5 and A6’s synthesis of the exchange “Paper assessment” (2026-10-28 to 29), which proposed carrying LIT-1820’s comparison of alternative paths into CASE-055 as independent translation routes. It adds to CASE-055 a relay arm, a path-discrepancy readout set beside the task loss, two negative controls, and an arm that fits correspondences between the communities’ response matrices and scores them on held-out data. Not run; every margin is left to a preregistration. | Active |