{"id":"W2163148621","doi":"10.3138/jvme.37.4.377","title":"North American Veterinary Licensing Examination Pacing Study","year":2010,"lang":"en","type":"article","venue":"Journal of Veterinary Medical Education","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Medicine; Test (biology); Psychology; Word (group theory); Presentation (obstetrics); Mathematics; Surgery","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01121121,0.0001449538,0.0003445,0.0003842649,0.0001724449,0.0001605474,0.0007113962,0.0000543056,0.0004856712],"category_scores_gemma":[0.01265249,0.0001028575,0.0001221492,0.0007303132,0.0001782039,0.0005703693,0.0001080196,0.0006799319,0.00003686187],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000913033,"about_ca_system_score_gemma":0.0007430947,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003348885,"about_ca_topic_score_gemma":0.00009598347,"domain_scores_codex":[0.9937757,0.0008293942,0.001276746,0.0002792142,0.003634947,0.0002040398],"domain_scores_gemma":[0.9964067,0.0007103928,0.001051474,0.0004432587,0.000996504,0.0003916998],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.000129957,0.002337358,0.128368,0.00001000477,0.0000245197,0.00007056295,0.002871259,0.00001090839,0.005599742,0.00001002073,0.001451718,0.859116],"study_design_scores_gemma":[0.0004223891,0.004918181,0.9693245,0.00005815773,0.00002515686,0.0006573721,0.006724181,0.0002324276,0.00005160186,0.0001737802,0.01727429,0.0001380043],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.993024,0.00002652009,0.0004920223,0.001892722,0.004007377,0.0001976849,4.955371e-7,0.00001016792,0.0003490479],"genre_scores_gemma":[0.997055,0.00001430805,0.00157006,0.0003648442,0.0009172914,0.00000558504,0.000001155583,0.000008896368,0.00006287808],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.858978,"threshold_uncertainty_score":0.9956644,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2172615267746426,"score_gpt":0.4645133154714006,"score_spread":0.247251788696758,"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."}}