{"id":"W2163019878","doi":"10.1177/1555343414532813","title":"Exploring the Use of Categories in the Assessment of Airline Pilots’ Performance as a Potential Source of Examiners’ Disagreement","year":2014,"lang":"en","type":"article","venue":"Journal of Cognitive Engineering and Decision Making","topic":"Human-Automation Interaction and Safety","field":"Psychology","cited_by":20,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Victoria","funders":"Griffith University","keywords":"Aviation; Aviation accident; Psychology; Applied psychology; Aviation safety; Computer science; Engineering","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009858558,0.00008801997,0.000225495,0.0002288531,0.00003722139,0.00001986927,0.0001100582,0.00002240671,0.00005561626],"category_scores_gemma":[0.0004124685,0.00005254888,0.00007129583,0.0001450634,0.00004224271,0.0002038359,0.0000301661,0.0002145932,7.037743e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001332105,"about_ca_system_score_gemma":0.00001326564,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000008820336,"about_ca_topic_score_gemma":0.000001832795,"domain_scores_codex":[0.9987404,0.0001039788,0.0006502721,0.00007180482,0.0003428194,0.00009074503],"domain_scores_gemma":[0.9974244,0.001754171,0.0004602354,0.0001009153,0.0002402169,0.00002004425],"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":[0.001262146,0.0006053951,0.02562197,0.0001816552,0.0004106339,0.00002278473,0.03539097,0.2680248,0.003397705,0.005760473,0.0001889867,0.6591325],"study_design_scores_gemma":[0.001362969,0.0007031076,0.9198808,0.001331365,0.00006820296,0.0001275889,0.007466939,0.06761725,0.0003403949,0.00008887737,0.0009090484,0.0001034731],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.796162,0.00004025181,0.2032124,0.00003432965,0.0003123569,0.00006065656,0.000001204777,0.000002723271,0.0001740174],"genre_scores_gemma":[0.9988295,0.00005025806,0.0009824882,0.00003329099,0.00007610562,0.000004807243,4.982051e-7,0.000007368108,0.00001571846],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8942588,"threshold_uncertainty_score":0.2142881,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09387926552206022,"score_gpt":0.3587214757991561,"score_spread":0.2648422102770959,"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."}}