{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.5957205,0.001525487,0.003995968,0.02162239,0.002699305,0.01749783,0.009306033,0.006316192,0.001847745],"category_scores_gemma":[0.8377728,0.001763623,0.00429595,0.01195575,0.02935562,0.02358997,0.01359339,0.008774247,0.0003645346],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006328898,"about_ca_system_score_gemma":0.007555582,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002997863,"about_ca_topic_score_gemma":0.004783971,"domain_scores_codex":[0.3779845,0.3804359,0.08165305,0.02575383,0.1318162,0.002356594],"domain_scores_gemma":[0.02868021,0.8816883,0.02904805,0.03492806,0.02488809,0.0007674186],"domain_codex":"methods","domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.001447051,0.0004212617,0.225925,0.01630644,0.009557727,0.0004532008,0.02225448,0.004841292,0.001233713,0.1101905,0.006223894,0.6011456],"study_design_scores_gemma":[0.0008219491,0.003488431,0.2049241,0.06091356,0.007793344,0.005004239,0.01841835,0.04356558,0.006221711,0.5959311,0.05127954,0.001638141],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1188127,0.1529199,0.6138474,0.07860024,0.003907029,0.001463285,0.001084064,0.0004040288,0.02896131],"genre_scores_gemma":[0.765453,0.01559649,0.2003186,0.011358,0.004460851,0.001530563,0.0005696705,0.0002321433,0.0004806379],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5957205,"threshold_uncertainty_score":0.4985487,"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."}}