{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01296026,0.0004664197,0.0005655847,0.002337387,0.0006437546,0.001952928,0.0007847632,0.0006040112,0.001042699],"category_scores_gemma":[0.1434376,0.0002775892,0.000518617,0.0007824636,0.00185951,0.001266596,0.001606536,0.0006197152,0.0002328173],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006540468,"about_ca_system_score_gemma":0.0005315021,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002528422,"about_ca_topic_score_gemma":0.002299198,"domain_scores_codex":[0.9780351,0.01111737,0.001575495,0.001669105,0.007154896,0.0004480158],"domain_scores_gemma":[0.8540073,0.09919143,0.01973049,0.008297463,0.01759289,0.001180298],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001248792,0.0001514166,0.7877914,0.0005458848,0.0005674364,0.0008474612,0.03683671,0.004525654,0.0226288,0.002803818,0.0008069047,0.1412458],"study_design_scores_gemma":[0.00004220207,0.0006702296,0.932473,0.000198325,0.0001979663,0.001319294,0.0198565,0.02931766,0.007446719,0.005861399,0.002438244,0.0001783727],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9879329,0.0001801156,0.008405773,0.0001175622,0.00002059342,0.00004951324,0.0000378508,0.00003617012,0.003219563],"genre_scores_gemma":[0.9977375,0.00004970031,0.001932144,0.00001334197,0.00001199845,0.0000144258,0.00002976548,0.000009360748,0.0002017527],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01296026,"threshold_uncertainty_score":0.06854129,"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."}}