{"id":"W4408097659","doi":"10.1332/03055736y2025d000000065","title":"What is successful policy experimentation?","year":2025,"lang":"en","type":"article","venue":"Policy & Politics","topic":"Leadership, Behavior, and Decision-Making Studies","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Foundation for Canadian Studies in the United Kingdom","keywords":"Political science","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","scholarly_communication","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0008096045,0.0003687943,0.0005595831,0.001895816,0.000677439,0.001683355,0.001341549,0.0001811483,0.0002999152],"category_scores_gemma":[0.004339551,0.0002991451,0.0003116688,0.002205714,0.0005329864,0.001047453,0.0004835171,0.000229569,0.001153019],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004640879,"about_ca_system_score_gemma":0.001118169,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002979249,"about_ca_topic_score_gemma":0.00008828378,"domain_scores_codex":[0.9952196,0.0002216799,0.001092267,0.0007863931,0.001698079,0.0009819758],"domain_scores_gemma":[0.9952645,0.002147421,0.0002564679,0.001378917,0.000673471,0.0002792504],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0000239014,0.0002629615,0.0215949,0.00001556714,0.00009338436,0.00001873011,0.04452591,0.00002038178,0.0001659072,0.7663079,0.1345322,0.03243829],"study_design_scores_gemma":[0.0009962379,0.00008632557,0.02523464,0.0001463524,0.00006345909,0.00001358882,0.08044867,0.0001199457,0.008217138,0.6017109,0.2823554,0.0006073806],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6349135,0.00432073,0.001822295,0.2150893,0.004796265,0.0007779746,0.000154152,0.0003950211,0.1377307],"genre_scores_gemma":[0.9149029,0.0002529652,0.0003679458,0.02805582,0.001183929,0.00003520972,0.000004287034,0.0000267303,0.05517022],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2799894,"threshold_uncertainty_score":0.9999461,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1409832250889204,"score_gpt":0.5144076319604024,"score_spread":0.3734244068714819,"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."}}