{"id":"W7155486217","doi":"10.5281/zenodo.19733516","title":"D5.6 First report on the evaluation process","year":2025,"lang":"","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Ethics and Social Impacts of AI","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Canadian AIDS Treatment Information Exchange","funders":"European Commission","keywords":"Deliverable; Task (project management); Process (computing); Protocol (science); Task analysis; Evaluation methods","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":[0.09961081,0.001756099,0.001445168,0.005358582,0.003846836,0.01516206,0.004219092,0.004664117,0.1114246],"category_scores_gemma":[0.15569,0.001599442,0.002022685,0.002356244,0.001807875,0.005812889,0.01104629,0.005630604,0.1246938],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01608675,"about_ca_system_score_gemma":0.03340866,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01119747,"about_ca_topic_score_gemma":0.008142974,"domain_scores_codex":[0.8908845,0.03298018,0.00980833,0.006574695,0.05466675,0.005085579],"domain_scores_gemma":[0.8132942,0.0247122,0.005370214,0.01644498,0.1298569,0.01032156],"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":[0.001075663,0.0008272586,0.003170331,0.002622379,0.00007835342,0.000478087,0.007536742,0.001500281,0.01068898,0.02664164,0.6322036,0.3131767],"study_design_scores_gemma":[0.000102414,0.0006560375,0.00492851,0.001171121,0.00002839158,0.0001268778,0.001861371,0.0005599438,0.006798703,0.00306308,0.9805963,0.0001074129],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.02032967,0.003493852,0.2090459,0.02116184,0.006999929,0.09422049,0.04258399,0.01170348,0.5904609],"genre_scores_gemma":[0.08438498,0.002664689,0.226356,0.005340978,0.001529432,0.08840545,0.06822837,0.00809567,0.5149943],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.1114246,"threshold_uncertainty_score":0.5267986,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1128978279413537,"score_gpt":0.385561736901905,"score_spread":0.2726639089605513,"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."}}