{"id":"W7155568543","doi":"10.5281/zenodo.19734032","title":"D5.9 First report on the evalaution 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; Iterative and incremental development; Troubleshooting","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":[0.04152071,0.00172504,0.000978274,0.004070431,0.002844106,0.01114946,0.003832589,0.003398356,0.09901676],"category_scores_gemma":[0.07688895,0.001459561,0.001412358,0.001609782,0.001465193,0.005095224,0.01163326,0.004307943,0.1144899],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008830132,"about_ca_system_score_gemma":0.01689145,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01071375,"about_ca_topic_score_gemma":0.007158168,"domain_scores_codex":[0.949406,0.01263877,0.00312394,0.003987479,0.02798493,0.002858869],"domain_scores_gemma":[0.9337178,0.01038168,0.002297683,0.01006375,0.03921659,0.004322333],"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":[0.0007512753,0.0007838969,0.00371411,0.00173482,0.00005565833,0.0010945,0.00885921,0.002572253,0.01596242,0.03440021,0.5241198,0.4059518],"study_design_scores_gemma":[0.00003870971,0.0004136509,0.003395397,0.0005254474,0.00001637622,0.0002137952,0.001517794,0.0005988789,0.009151752,0.00319214,0.9808567,0.00007928172],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.02655938,0.002814129,0.2768714,0.01276393,0.003960258,0.0245786,0.03429497,0.01858758,0.5995698],"genre_scores_gemma":[0.0911672,0.00209912,0.2178078,0.002345811,0.0008456714,0.02357507,0.06406754,0.0114928,0.5865991],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.09901676,"threshold_uncertainty_score":0.3312441,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06558036223926425,"score_gpt":0.3492009309193913,"score_spread":0.2836205686801271,"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."}}