{"id":"W3027538061","doi":"10.5430/jnep.v10n8p74","title":"Reliability and validity of scenario-specific versus generic simulation assessment rubrics","year":2020,"lang":"en","type":"article","venue":"Journal of Nursing Education and Practice","topic":"Simulation-Based Education in Healthcare","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Rubric; Debriefing; Reliability (semiconductor); Computer science; Psychology; Medical education; Mathematics education; Medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.04370585,0.0004826208,0.0006958068,0.00180211,0.0004611764,0.001031264,0.0008951089,0.0005011795,0.001563972],"category_scores_gemma":[0.1500085,0.0002818634,0.001301263,0.0007482665,0.0009089724,0.0009972554,0.001768491,0.0005862047,0.0006098587],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006298934,"about_ca_system_score_gemma":0.0008193741,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009508537,"about_ca_topic_score_gemma":0.001240643,"domain_scores_codex":[0.9548866,0.02523799,0.006573528,0.003029993,0.009554248,0.0007176498],"domain_scores_gemma":[0.878388,0.06450558,0.01028263,0.009797624,0.0356004,0.001425827],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002745857,0.000775233,0.6173313,0.00181634,0.001081405,0.0001970797,0.01299868,0.005034721,0.01374159,0.001457721,0.003323654,0.3394964],"study_design_scores_gemma":[0.0002724744,0.004765764,0.9177268,0.001937119,0.0006361891,0.001574425,0.005567586,0.03550711,0.01590192,0.001820629,0.01401944,0.0002705913],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9591532,0.001150386,0.03174968,0.0001447939,0.0002399455,0.0008628122,0.0003557307,0.0003206597,0.006022722],"genre_scores_gemma":[0.98387,0.0002575617,0.01444995,0.00004504991,0.00003678154,0.0004470643,0.0005014389,0.00005674794,0.0003353168],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9562941,"threshold_uncertainty_score":0.2311414,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3132676389802285,"score_gpt":0.512938179821187,"score_spread":0.1996705408409584,"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."}}