{"id":"W2039069397","doi":"10.1007/s10459-009-9169-z","title":"Is case-specificity content-specificity? An analysis of data from extended-matching questions","year":2009,"lang":"en","type":"article","venue":"Advances in Health Sciences Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":22,"is_retracted":false,"has_abstract":false,"ca_institutions":"Université de Montréal","funders":"","keywords":"Variance (accounting); Matching (statistics); Reliability (semiconductor); Test (biology); Psychology; Statistics; Computer science; Natural language processing; Mathematics","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":[0.05513086,0.0006329378,0.001688386,0.00424952,0.001154554,0.0016226,0.002464284,0.001886379,0.006694906],"category_scores_gemma":[0.4239516,0.0005315383,0.001677895,0.004721642,0.00191729,0.004046828,0.005458167,0.002464529,0.0009786726],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001466678,"about_ca_system_score_gemma":0.001932697,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002642922,"about_ca_topic_score_gemma":0.002068846,"domain_scores_codex":[0.9154375,0.05273145,0.01091697,0.005828465,0.01280916,0.002276302],"domain_scores_gemma":[0.1968365,0.74907,0.02193297,0.01799666,0.0130497,0.001114015],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01099377,0.001453455,0.8772823,0.002120324,0.00210927,0.0009944015,0.01817422,0.002818719,0.002970391,0.003176029,0.003266667,0.07464035],"study_design_scores_gemma":[0.0003486961,0.001198828,0.9696444,0.0004543381,0.0008042931,0.001928658,0.004725245,0.005623045,0.002651383,0.004762347,0.007756291,0.0001025454],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9833208,0.000655619,0.007958054,0.000285148,0.0000244758,0.0009033235,0.003823406,0.00004537543,0.002983734],"genre_scores_gemma":[0.9875808,0.0001270759,0.003752811,0.0002273521,0.00002365331,0.001033113,0.006787435,0.00005886538,0.0004088297],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05513086,"threshold_uncertainty_score":0.2915633,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1362837500955101,"score_gpt":0.4901600374232628,"score_spread":0.3538762873277527,"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."}}