{"id":"W4367191237","doi":"10.5281/zenodo.7874593","title":"D5.3 Evaluation Criteria","year":2022,"lang":"en","type":"report","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Manufacturing Process and Optimization","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Canadian Automotive Partnership Council","funders":"Horizon 2020 Framework Programme; European Commission","keywords":"Computer science; Geography","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":["metaepi_narrow","sts","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.00162456,0.0002522412,0.0002397506,0.0003943276,0.001555647,0.0007963858,0.0008192814,0.0001546368,0.1548357],"category_scores_gemma":[0.0005600603,0.0002994354,0.00007724133,0.0003879245,0.00004413163,0.0001993582,0.0007028416,0.0005614308,0.001713543],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009633988,"about_ca_system_score_gemma":0.00001933316,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001936274,"about_ca_topic_score_gemma":3.002277e-7,"domain_scores_codex":[0.9973067,0.0002191174,0.0003735893,0.0004379898,0.001313417,0.0003491683],"domain_scores_gemma":[0.9981619,0.00001351583,0.0001269331,0.0005211845,0.001054356,0.0001221213],"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":[0.00001127565,0.00003858163,3.035488e-7,0.0006011013,0.0000835852,0.000007600114,0.00048266,0.03787807,0.0001098394,0.00006484946,0.768923,0.1917991],"study_design_scores_gemma":[0.0002199388,0.00006053042,0.00009950767,0.00004687546,0.00005922309,0.00006873595,0.00006700523,0.01877411,0.0001300797,0.00007339626,0.9800959,0.0003047342],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.0008347284,0.001079041,0.00663788,0.0001038439,0.001554608,0.0009724381,0.0006975462,0.003200721,0.9849192],"genre_scores_gemma":[0.7661724,0.01332979,0.00201097,0.0002296009,0.004104522,0.000003442575,0.1567371,0.03019156,0.0272206],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.9576986,"threshold_uncertainty_score":0.9999458,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07064894893957248,"score_gpt":0.2887309496388846,"score_spread":0.2180820006993121,"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."}}