{"id":"W4417181692","doi":"10.1007/s12207-025-09551-w","title":"Reading Between the Lines: Ensuring Validity in Psychometric Testing for Non-Native English Speakers","year":2025,"lang":"en","type":"article","venue":"Psychological Injury and Law","topic":"Neurobiology of Language and Bilingualism","field":"Neuroscience","cited_by":2,"is_retracted":false,"has_abstract":false,"ca_institutions":"The Scarborough Hospital; University of Toronto","funders":"","keywords":"Reading (process); Legal psychology; Test (biology); Attribution; Plaintiff; Literacy; Psychometrics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008589318,0.0001564002,0.0002455712,0.0001395793,0.0003297337,0.00004382123,0.0002938413,0.0001538911,0.000008423362],"category_scores_gemma":[0.004831714,0.00009822896,0.00005825001,0.000929673,0.0003612759,0.0000843461,0.00008962245,0.0004646397,0.00000243294],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001329842,"about_ca_system_score_gemma":0.000006005139,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001364652,"about_ca_topic_score_gemma":0.000002868943,"domain_scores_codex":[0.9984204,0.0002800977,0.0003184989,0.0005711175,0.00007381972,0.0003360277],"domain_scores_gemma":[0.9944602,0.005150622,0.00009115285,0.0002058215,0.00004910949,0.00004306993],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001363155,0.001169851,0.4363304,0.0003211073,0.00007244563,0.000308187,0.008803523,0.00001356586,0.3435172,0.03092617,0.003945753,0.1732286],"study_design_scores_gemma":[0.006968403,0.004486429,0.658135,0.0007140315,0.0001924279,0.00008403705,0.002333418,0.0002812795,0.1851283,0.09299302,0.04674809,0.001935564],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9780467,0.00002546665,0.0001592342,0.0004358875,0.0005670491,0.0004474844,0.00002724383,0.00006912329,0.02022181],"genre_scores_gemma":[0.9948487,0.00002024746,0.0003437096,0.004276758,0.0003212465,0.00003755499,0.000002221365,0.00000677224,0.0001428059],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2218045,"threshold_uncertainty_score":0.5784362,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1294776872301069,"score_gpt":0.3923401978354648,"score_spread":0.2628625106053579,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}