{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008218484,0.0003818479,0.0008061965,0.003737797,0.0008489285,0.00174274,0.001583077,0.0009789724,0.002465636],"category_scores_gemma":[0.0280131,0.0003555606,0.0007989392,0.002113954,0.001310777,0.0009098055,0.001631174,0.001312317,0.001211312],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002367086,"about_ca_system_score_gemma":0.007370042,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008659518,"about_ca_topic_score_gemma":0.01172036,"domain_scores_codex":[0.9909697,0.002302086,0.001123288,0.0005375152,0.00470032,0.0003670702],"domain_scores_gemma":[0.9769202,0.008711766,0.001147618,0.001096258,0.01122439,0.0008998569],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.001062978,0.004321205,0.2177697,0.0006286024,0.0001608859,0.0006603586,0.004697052,0.007884242,0.01286654,0.01349688,0.01020768,0.7262439],"study_design_scores_gemma":[0.0004355702,0.007729739,0.842041,0.0007607114,0.0001874908,0.001216628,0.005276571,0.03970546,0.02357933,0.01171088,0.06698102,0.0003757472],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7167738,0.0004427348,0.1763456,0.0009437649,0.0004681243,0.02224208,0.005242468,0.001618621,0.07592267],"genre_scores_gemma":[0.6805947,0.0003365878,0.2856121,0.0003288638,0.00006254492,0.01791903,0.00502816,0.0001212182,0.009996739],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.008659518,"threshold_uncertainty_score":0.043464,"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."}}