{"id":"W4417417655","doi":"10.31234/osf.io/r4dhw_v1","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":"article","venue":"","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"],"consensus_categories":[],"category_scores_codex":[0.005361973,0.0001156569,0.0003217268,0.0004579228,0.0001856528,0.00009571845,0.0002891385,0.00007652974,0.00006115087],"category_scores_gemma":[0.051577,0.00007631574,0.00006024758,0.0007038586,0.0002046787,0.0001205978,0.0002489378,0.00008980821,2.827534e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002091778,"about_ca_system_score_gemma":0.0001203726,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006157754,"about_ca_topic_score_gemma":0.00001102775,"domain_scores_codex":[0.9978135,0.0003190914,0.0008296224,0.0003322508,0.000583914,0.0001215796],"domain_scores_gemma":[0.986738,0.01195563,0.0006857454,0.0001536272,0.0004406626,0.00002638236],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006262535,0.00002266655,0.5858176,0.00002857904,0.00008743557,2.480123e-8,0.0007011305,0.00009386611,0.001161608,0.001155058,0.0001334365,0.410736],"study_design_scores_gemma":[0.001036915,0.00006879101,0.9749577,0.0002529712,0.00004615152,0.00000185196,0.001240693,0.001296501,0.01273407,0.006135633,0.002139065,0.00008967014],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8792033,0.001307011,0.1172765,0.0001947794,0.000292213,0.0004881468,0.000009536143,0.000008642861,0.001219836],"genre_scores_gemma":[0.9581029,0.0003286253,0.03935153,0.00002206357,0.00001220443,0.000009146386,0.000004980319,0.000004234929,0.002164282],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4106463,"threshold_uncertainty_score":0.956412,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3646703116239629,"score_gpt":0.4567923133589386,"score_spread":0.09212200173497564,"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."}}