{"id":"W4255656260","doi":"10.1515/iupac.79.0893","title":"Benchmark Response","year":2016,"lang":"en","type":"dataset","venue":"IUPAC Standards Online","topic":"History and advancements in chemistry","field":"Chemistry","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Glossary; Chemical nomenclature; Computer science; Hazard; Toxicology; Chemistry; Biology; Philosophy; Linguistics","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":[0.003345537,0.002083535,0.001897825,0.003602787,0.00130207,0.003933512,0.003783331,0.003398187,0.2283854],"category_scores_gemma":[0.03391828,0.0005568431,0.002177393,0.005479072,0.0005117987,0.002500652,0.002835767,0.002364326,0.2532253],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002360138,"about_ca_system_score_gemma":0.003427229,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01231827,"about_ca_topic_score_gemma":0.02574656,"domain_scores_codex":[0.9951374,0.001196429,0.0006853408,0.001402304,0.001074782,0.0005038242],"domain_scores_gemma":[0.9902697,0.003203583,0.0006089367,0.002517333,0.002845695,0.0005548442],"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":[0.0001479281,0.00002335218,0.0007428658,0.0005150815,0.00003090436,0.00001286867,0.00001032634,0.0002396795,0.00004004702,0.0005536221,0.9922119,0.005471326],"study_design_scores_gemma":[0.0004299546,0.0000530112,0.002982752,0.0005190619,0.00004708929,0.00006813879,0.0001030696,0.0005957608,0.0002323032,0.00280832,0.992126,0.00003456206],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"dataset","genre_scores_codex":[0.0003038415,0.000202348,0.0001452764,0.0004270207,0.0002072709,0.00007383665,0.9939241,0.0005866506,0.004129664],"genre_scores_gemma":[0.001010003,0.0001046078,0.000557396,0.0004603821,0.0000572996,0.0002616791,0.9937301,0.0001310526,0.003687525],"genre_candidate":"dataset","genre_consensus":"dataset","teacher_disagreement_score":0.2283854,"threshold_uncertainty_score":0.7640252,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0117349637130397,"score_gpt":0.3886429260418791,"score_spread":0.3769079623288394,"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."}}