{"id":"W309577393","doi":"","title":"Balance Trust and Accountability: Education Reformers Want More Accountability Even Though the Evidence Suggests Better Outcomes Result from Fewer Rules","year":2012,"lang":"en","type":"article","venue":"Phi Delta Kappan","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Distrust; Accountability; Dilemma; Government (linguistics); Public relations; Feeling; Sociology; Political science; Graduation (instrument); Law; Public administration; Psychology; Social psychology","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":[],"consensus_categories":[],"category_scores_codex":[0.003961251,0.0002984338,0.0003688773,0.00009787978,0.0004032732,0.000479091,0.001008508,0.0001001736,0.0005313479],"category_scores_gemma":[0.002310863,0.0001645725,0.0001239315,0.000377396,0.0003264414,0.002478419,0.0003188377,0.0002728955,0.0001667811],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002550581,"about_ca_system_score_gemma":0.0003413078,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002289445,"about_ca_topic_score_gemma":0.0003780234,"domain_scores_codex":[0.9960035,0.0002689653,0.0008511574,0.0007262272,0.001661258,0.0004888934],"domain_scores_gemma":[0.994919,0.002624733,0.0003975834,0.001356536,0.0005000741,0.0002020546],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00003689952,0.0002251196,0.9522144,0.000007561632,0.0000248051,4.794373e-8,0.003737033,9.179005e-7,0.0002355676,0.0002216837,0.007624278,0.03567173],"study_design_scores_gemma":[0.000174783,0.0000223432,0.9537024,0.00003761909,0.00003854947,0.000001471117,0.009381321,0.0001106748,0.0001872316,0.01277686,0.02334062,0.0002261605],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9552858,0.0009676733,0.00005176483,0.04100219,0.00138367,0.0005883281,0.0001407483,0.00003183344,0.0005480277],"genre_scores_gemma":[0.993129,0.00007803685,0.001300793,0.003891142,0.0007014061,0.0001745932,0.00006761714,0.0000167966,0.0006405766],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03784327,"threshold_uncertainty_score":0.6711071,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1194307520776817,"score_gpt":0.4344076124386136,"score_spread":0.3149768603609319,"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."}}