{"id":"W7045865752","doi":"","title":"Calibration Committees and Rating Distribution Guidance Effects on Leniency Bias in Subjective Performance Evaluations","year":2023,"lang":"en","type":"dissertation","venue":"UWSpace (University of Waterloo)","topic":"Magnetic confinement fusion research","field":"Physics and Astronomy","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"University of Waterloo; Universität Wien","keywords":"Incentive; Anticipation (artificial intelligence); Distribution (mathematics); Calibration; Affect (linguistics)","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01317708,0.0004154707,0.0004501461,0.000399748,0.0005175793,0.001460807,0.00062186,0.0008449173,0.006430423],"category_scores_gemma":[0.06954088,0.0002666159,0.0003317606,0.0003560189,0.0009123555,0.0008293719,0.001044345,0.001014959,0.0007054295],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007905528,"about_ca_system_score_gemma":0.0008056831,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001568398,"about_ca_topic_score_gemma":0.002076965,"domain_scores_codex":[0.9837154,0.01159215,0.0006896806,0.001238308,0.002151456,0.0006131123],"domain_scores_gemma":[0.8368918,0.1143042,0.03034413,0.009588318,0.003857086,0.005014468],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.02475021,0.01854311,0.5049912,0.001198117,0.0007757598,0.0003748433,0.01262807,0.009752437,0.1734732,0.010048,0.003847812,0.2396172],"study_design_scores_gemma":[0.001428538,0.01458692,0.9446931,0.0001472669,0.0002340645,0.00009514024,0.001641317,0.008477546,0.02203063,0.002649063,0.003896769,0.0001197545],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"other","genre_scores_codex":[0.9913038,0.00005404128,0.001722723,0.0001379641,0.00002000342,0.0001106286,0.00004251639,0.0000466112,0.00656175],"genre_scores_gemma":[0.995972,0.00002721153,0.002570172,0.00008493396,0.0000149185,0.0001594883,0.00004005665,0.00001290425,0.00111817],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.01317708,"threshold_uncertainty_score":0.0696879,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01701864801708234,"score_gpt":0.2538297121820119,"score_spread":0.2368110641649296,"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."}}