{"id":"W2296655817","doi":"10.14288/1.0077401","title":"The construction of a criterion-referenced physical education knowledge test","year":2010,"lang":"en","type":"article","venue":"cIRcle (University of British Columbia)","topic":"Physical Education and Pedagogy","field":"Health Professions","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Test (biology); Criterion-referenced test; Mathematics; Computer science; Statistics; Geology; Standardized test","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":[0.0001448991,0.00002918374,0.0001618459,0.00002294596,0.0006631984,0.0000138923,0.0002190024,0.00009343812,0.0002408512],"category_scores_gemma":[0.0002183305,0.00007230321,0.00006891264,0.0002049885,0.000454803,0.0001356759,0.00006662004,0.0003676182,0.0000799757],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003205181,"about_ca_system_score_gemma":0.0009622388,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.03272778,"about_ca_topic_score_gemma":0.1952096,"domain_scores_codex":[0.9993193,0.0001114349,0.0001345792,0.0001627612,0.0001075583,0.0001643787],"domain_scores_gemma":[0.9984686,0.000483956,0.0001920496,0.0002422854,0.0005167212,0.00009634659],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.000009887003,0.0006053111,0.02308637,0.0001177351,0.00001271168,3.141578e-7,0.003533534,4.200196e-8,0.008862507,0.0006248224,0.00967549,0.9534713],"study_design_scores_gemma":[0.0003307119,0.00004124086,0.9682686,0.00009859375,0.00001868305,0.000003269254,0.01238806,0.00006577273,0.000004176672,0.002681959,0.01602623,0.00007274777],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9797725,0.00002118952,0.00002631758,0.0004227837,0.001010914,0.0002316643,0.00006207054,0.00002813714,0.01842442],"genre_scores_gemma":[0.9970876,0.00003615267,0.0001687178,0.00003660141,0.0001892289,0.000002622435,0.00001651506,0.000006198663,0.002456338],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9533985,"threshold_uncertainty_score":0.9737134,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02475092567670564,"score_gpt":0.3330233286326581,"score_spread":0.3082724029559525,"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."}}