{"id":"W7116213154","doi":"","title":"Creation and validation of the LEVANTE core tasks: Internationalized measures of learning and development for children ages 5-12 years","year":2025,"lang":"","type":"preprint","venue":"PsyArXiv (OSF Preprints)","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Set (abstract data type); Reliability (semiconductor); Core (optical fiber); Sample (material); Measure (data warehouse); Cognition; Selection (genetic algorithm)","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.01736643,0.0003496318,0.0008752937,0.0007141945,0.0003049338,0.0001871503,0.001098294,0.0003040934,0.0007338956],"category_scores_gemma":[0.1073159,0.0002814759,0.0002254341,0.0006525822,0.0004341561,0.0001549791,0.002262209,0.00047644,0.00002695476],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007802524,"about_ca_system_score_gemma":0.0003253397,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001423469,"about_ca_topic_score_gemma":0.00001271112,"domain_scores_codex":[0.9940706,0.001170044,0.001861708,0.001510198,0.00111837,0.0002691143],"domain_scores_gemma":[0.9782906,0.0173068,0.00248761,0.0009164047,0.0009173373,0.00008124905],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001938405,0.00007871418,0.7074832,0.0002075891,0.0003369524,1.390239e-7,0.002457276,0.002057832,0.001088355,0.0004573658,0.0001247265,0.285514],"study_design_scores_gemma":[0.001287372,0.00001780057,0.9650645,0.001034006,0.0001666439,0.000007145348,0.0007019473,0.002700442,0.01033155,0.01493537,0.003441573,0.000311695],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9006387,0.0003356477,0.09402455,0.0001666445,0.0007728367,0.001719471,0.00006156061,0.00003367716,0.002246938],"genre_scores_gemma":[0.9520697,0.001091229,0.04276565,0.00001711147,0.00004529951,0.0001103905,0.00002119799,0.00002012414,0.003859319],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2852023,"threshold_uncertainty_score":0.9999638,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2953755882102296,"score_gpt":0.4220356792820552,"score_spread":0.1266600910718256,"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."}}