{"id":"W4281641218","doi":"10.1111/medu.14853","title":"When I say <i>…</i> response process validity evidence","year":2022,"lang":"en","type":"article","venue":"Medical Education","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Sherbrooke; McGill University","funders":"","keywords":"Psychology; Process (computing); Equity (law); Cognition; Cognitive psychology; Social psychology; Applied psychology; Computer science; Political science; Psychiatry","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":["metaresearch"],"category_scores_codex":[0.5005898,0.000942678,0.00214436,0.006181276,0.002724393,0.01224905,0.005912336,0.008595277,0.01516882],"category_scores_gemma":[0.7972859,0.0008606953,0.00602988,0.005524864,0.01498198,0.009435291,0.007352349,0.0110145,0.004083653],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005880523,"about_ca_system_score_gemma":0.01420077,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003585716,"about_ca_topic_score_gemma":0.00330117,"domain_scores_codex":[0.4905263,0.3318523,0.05485462,0.02357074,0.09243135,0.00676468],"domain_scores_gemma":[0.1253491,0.7361667,0.03790541,0.0391301,0.05931368,0.002135026],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.003476276,0.0007998012,0.04788291,0.02966684,0.005897103,0.000283467,0.01060492,0.001542403,0.001216498,0.3485809,0.07703574,0.4730131],"study_design_scores_gemma":[0.001981385,0.001917131,0.0631168,0.06578153,0.007633677,0.001067486,0.009015598,0.007538298,0.01364952,0.4963399,0.3312104,0.0007483833],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.04575058,0.06148485,0.2348611,0.427748,0.01712952,0.004512075,0.00489161,0.00067771,0.2029447],"genre_scores_gemma":[0.700106,0.01343343,0.1324985,0.1296525,0.006752565,0.006577369,0.003972421,0.0004280296,0.006579089],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.4994102,"threshold_uncertainty_score":0.6158617,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2864561348748174,"score_gpt":0.5686232695119959,"score_spread":0.2821671346371786,"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."}}