{"id":"W1972358781","doi":"10.1111/cogs.12152","title":"Peer Assessment of Aviation Performance: Inconsistent for Good Reasons","year":2014,"lang":"en","type":"article","venue":"Cognitive Science","topic":"Team Dynamics and Performance","field":"Psychology","cited_by":23,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Victoria","funders":"","keywords":"Aviation; Affect (linguistics); Psychology; Variance (accounting); Reliability (semiconductor); Cognition; Cognitive psychology; Applied psychology; Computer science; Social psychology; 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":[0.001607998,0.00008100041,0.0001181631,0.0001195184,0.0002037399,0.00002383176,0.0001983707,0.00003794444,0.00009207582],"category_scores_gemma":[0.0002409572,0.00007222059,0.00004123796,0.0003196183,0.0003970405,0.0001710299,0.00004611754,0.00008180734,0.00002625209],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003872627,"about_ca_system_score_gemma":0.000122146,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001374266,"about_ca_topic_score_gemma":0.000008002811,"domain_scores_codex":[0.9989688,0.00002954613,0.0001772606,0.0002755021,0.0002994082,0.0002494766],"domain_scores_gemma":[0.998602,0.000177187,0.0001413366,0.0001747077,0.0008419017,0.00006284549],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0000727654,0.0003260218,0.4334545,0.00006944266,0.00003175524,3.306617e-7,0.002181487,0.0000333651,0.001920109,0.2743819,0.0002892367,0.287239],"study_design_scores_gemma":[0.0005483086,0.0004797205,0.9580215,0.00004960696,0.00002308757,0.000004144022,0.0005436449,0.03795169,0.0005440559,0.000773563,0.0009320047,0.0001287249],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8480284,0.00001007731,0.05145976,0.0002358607,0.0006197955,0.0002886529,0.00004879525,0.00001774715,0.0992909],"genre_scores_gemma":[0.9963434,0.000004036267,0.001631175,0.0001249936,0.00005861391,0.00007712297,0.00001644715,0.000006505839,0.001737662],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5245669,"threshold_uncertainty_score":0.294507,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02825236003594501,"score_gpt":0.3729892835386608,"score_spread":0.3447369235027158,"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."}}