{"id":"W4245992123","doi":"10.1002/ev.20367","title":"Issue Information","year":2021,"lang":"en","type":"paratext","venue":"New Directions for Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Computer science; Citation; World Wide Web; Information retrieval; Library science","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"grok","categories":["insufficient_payload"],"domain":null,"study_design":"not_applicable","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"opus","categories":["insufficient_payload"],"domain":null,"study_design":"not_applicable","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.008122634,0.0003200991,0.0005006663,0.001027674,0.0005563062,0.001110469,0.0004890956,0.0004242735,0.1506526],"category_scores_gemma":[0.005698301,0.0002843623,0.0003724948,0.001598294,0.00003311803,0.001853754,0.0000749969,0.0002930232,0.0650481],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005672496,"about_ca_system_score_gemma":0.004462644,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001311316,"about_ca_topic_score_gemma":0.000267378,"domain_scores_codex":[0.9929203,0.0005301491,0.001469721,0.0005822469,0.004187302,0.0003102403],"domain_scores_gemma":[0.9914367,0.0009826337,0.00100778,0.0009394258,0.005464302,0.0001691845],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001279961,0.00002062087,0.000006590805,0.00001159947,0.00002900595,1.405158e-8,0.0003053744,0.003649432,0.000009650434,0.0001044392,0.6409153,0.3549351],"study_design_scores_gemma":[0.00122394,0.0000845906,0.000314764,0.0000546109,0.0001639939,0.000003341378,0.0005944876,0.03776553,0.000340976,0.002380333,0.9567752,0.0002981726],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.0002810697,0.002638979,0.0847837,0.007184287,0.04907193,0.003692383,0.0004559747,0.00009903762,0.8517926],"genre_scores_gemma":[0.00120942,0.000647451,0.004061347,0.001219431,0.003311564,0.001145583,0.008266319,0.00003742157,0.9801015],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.354637,"threshold_uncertainty_score":0.9999608,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2232214289072087,"score_gpt":0.5437897844404399,"score_spread":0.3205683555332312,"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."}}