{"id":"W4416221412","doi":"10.1080/1091367x.2025.2579239","title":"Dealing with Expert Uncertainty in Assigning Motor Competence Scores: New Dimensions of Validity Evidence","year":2025,"lang":"en","type":"article","venue":"Measurement in Physical Education and Exercise Science","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Competence (human resources); Construct validity; Motor skill; Dreyfus model of skill acquisition; Predictive validity; External validity; Validity","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.001003064,0.00009954863,0.0002795115,0.0001918116,0.0000705881,0.00002402493,0.0001223049,0.0000238604,0.000004917224],"category_scores_gemma":[0.01204887,0.00006800332,0.00002809144,0.000910905,0.0003347775,0.0001332032,0.00004303435,0.0001388941,0.000001554939],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001663853,"about_ca_system_score_gemma":0.002199889,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001270048,"about_ca_topic_score_gemma":0.00005794252,"domain_scores_codex":[0.9985404,0.00003187357,0.0002551296,0.0003539441,0.000626325,0.0001923056],"domain_scores_gemma":[0.9981356,0.001104744,0.00006993819,0.0002301101,0.0002722449,0.0001873598],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001229204,0.006308276,0.5333597,0.0003090192,0.00001507439,0.0000077566,0.009711773,0.002715473,0.1375023,0.005012006,0.0007035863,0.3031258],"study_design_scores_gemma":[0.00174173,0.0005186938,0.880156,0.07549612,0.00007724709,0.000002996547,0.00262919,0.01264596,0.02312532,0.003260595,0.00006183475,0.0002842695],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9969832,0.0007099427,0.0002725807,0.001222995,0.000159289,0.0003457528,2.838974e-7,0.00001634595,0.0002896131],"genre_scores_gemma":[0.9978115,0.0002897092,0.001527985,0.0002768401,0.00002608677,0.00002950094,4.57597e-7,0.000003742253,0.00003420929],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3467963,"threshold_uncertainty_score":0.996273,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08983193852424332,"score_gpt":0.3876370870209674,"score_spread":0.2978051484967241,"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."}}