{"id":"W4400482763","doi":"10.55016/ojs/cpai.v6i1.76623","title":"Creative evaluation","year":2023,"lang":"en","type":"article","venue":"Canadian Perspectives on Academic Integrity","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Assiniboine Community College","funders":"","keywords":"Computer science; Psychology","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":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.01169315,0.0001770789,0.0002235418,0.001381161,0.0003747109,0.0001212053,0.0007114061,0.0002711655,0.0105176],"category_scores_gemma":[0.01153535,0.0001388599,0.0001121442,0.002165622,0.000161002,0.0004147492,0.00003837194,0.001505084,0.01022005],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001774643,"about_ca_system_score_gemma":0.002853109,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.01218919,"about_ca_topic_score_gemma":0.02854314,"domain_scores_codex":[0.9954473,0.0006278277,0.0004539187,0.0006741619,0.002329025,0.0004677005],"domain_scores_gemma":[0.99682,0.0007396151,0.0001541045,0.0005042491,0.00131691,0.0004651417],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.00007399108,0.000043149,0.02616736,0.000003390185,0.0000673307,0.00001399356,0.05102051,0.001480966,0.0002174291,0.287743,0.4630737,0.1700952],"study_design_scores_gemma":[0.0008054408,0.0001967017,0.4693316,0.00005305474,0.00002874787,0.00000672537,0.1874444,0.09396517,0.0003220358,0.1141539,0.133205,0.0004872174],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6143446,0.0001584323,0.00009736933,0.02741853,0.0007581932,0.0006583198,0.0001078783,0.0001259167,0.3563308],"genre_scores_gemma":[0.9885948,0.0001170388,0.00005285392,0.001599818,0.0002453947,0.0000809972,0.00002782302,0.00001255846,0.00926876],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4431643,"threshold_uncertainty_score":0.9967909,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3184331467128143,"score_gpt":0.5352887859850761,"score_spread":0.2168556392722618,"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."}}