{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001159299,0.001737747,0.001323682,0.00542718,0.002165233,0.009849261,0.002091056,0.002915466,0.9245241],"category_scores_gemma":[0.006839439,0.0007970319,0.001085571,0.005847362,0.000641134,0.003829143,0.002246916,0.002257409,0.8840906],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002403858,"about_ca_system_score_gemma":0.00615756,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01591363,"about_ca_topic_score_gemma":0.03715463,"domain_scores_codex":[0.998702,0.00009379484,0.00007119765,0.0001473805,0.0008325616,0.0001530278],"domain_scores_gemma":[0.9952015,0.0005448972,0.0002294907,0.0004384623,0.002468265,0.001117465],"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":[0.000009014626,0.00001399109,0.00002377301,0.00007724569,7.245332e-7,0.000007232859,0.000003802873,0.0000229487,0.00003376752,0.0005481791,0.9818897,0.01736964],"study_design_scores_gemma":[0.00001063859,0.000006777242,0.0001997801,0.000070149,0.000001142045,0.000008665761,0.00001317616,0.00003227951,0.00003056935,0.0004648337,0.9991584,0.000003510084],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.0001146831,0.0005491824,0.0004182925,0.002057596,0.006045516,0.0002926902,0.02624662,0.001380079,0.9628953],"genre_scores_gemma":[0.0004070469,0.000453581,0.0003242242,0.0006653548,0.0007178313,0.00009453295,0.01034456,0.0003482167,0.9866446],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.07547587,"threshold_uncertainty_score":0,"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."}}