{"id":"W4409249547","doi":"10.64000/9yk4w3ivc","title":"The programs approach: our experiences during the first quarter of 2025","year":2025,"lang":"en","type":"preprint","venue":"","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Quarter (Canadian coin); Political science; Psychology; Geography; Archaeology","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":"medium","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":[],"domain":null,"study_design":"not_applicable","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02333253,0.0004546653,0.0003121355,0.0007261964,0.007098545,0.009585245,0.001961837,0.003076215,0.01140266],"category_scores_gemma":[0.01820181,0.0003317296,0.0004797218,0.002069317,0.004527297,0.005059415,0.01158709,0.006877382,0.003191506],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008713617,"about_ca_system_score_gemma":0.01842583,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02685198,"about_ca_topic_score_gemma":0.04876115,"domain_scores_codex":[0.9774321,0.0142736,0.0002856688,0.0008252042,0.003606508,0.003576812],"domain_scores_gemma":[0.9740316,0.002791865,0.0004163444,0.0005788727,0.003174942,0.01900641],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000685257,0.00335103,0.01587747,0.0006503311,0.00006130273,0.002438699,0.1022417,0.0017867,0.001580203,0.1229714,0.3667463,0.3816097],"study_design_scores_gemma":[0.00004853257,0.0003544914,0.005827103,0.000220784,0.000008786492,0.0003044356,0.05117409,0.0007176138,0.0006135884,0.009154159,0.9315327,0.00004365172],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"other","genre_scores_codex":[0.2586955,0.006859187,0.02854218,0.4291977,0.006892452,0.0007015238,0.001900395,0.00112638,0.2660847],"genre_scores_gemma":[0.7529055,0.004754176,0.02593752,0.0552715,0.001373134,0.0007534663,0.002471623,0.001308073,0.155225],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.02685198,"threshold_uncertainty_score":0,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2091367945894002,"score_gpt":0.475125807942161,"score_spread":0.2659890133527608,"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."}}