{"id":"W4366452169","doi":"10.3138/cjpe.20.007","title":"Preparing School Evaluators: Hiroshima Pilot Test of the Japan Evaluation Society’s Accreditation Project","year":2005,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"Toronto Public Health","funders":"","keywords":"Accreditation; Certification; Certification and Accreditation; Test (biology); Context (archaeology); Medical education; Sustainability; Program evaluation; Government (linguistics); Political science; Quality (philosophy); Public relations; Psychology; Medicine; Public administration","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04267324,0.00022079,0.0003462379,0.0006034301,0.0004419537,0.0004879745,0.000979954,0.0001223899,0.001677089],"category_scores_gemma":[0.01863492,0.0001522946,0.0003503799,0.001679314,0.0001878203,0.001654145,0.00004120846,0.0003725614,0.00006432168],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001628184,"about_ca_system_score_gemma":0.021869,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004803687,"about_ca_topic_score_gemma":0.01657259,"domain_scores_codex":[0.9892831,0.001335482,0.001896298,0.0003565179,0.00673305,0.0003955708],"domain_scores_gemma":[0.9873881,0.0006271991,0.002222776,0.0006616316,0.008734359,0.0003659275],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000031355,0.0001847866,0.08682996,0.00001365899,0.00006796871,1.302293e-7,0.002966383,0.01943956,0.001007011,0.00009171812,0.02322803,0.8661395],"study_design_scores_gemma":[0.002888952,0.00139575,0.3620611,0.0002036902,0.0005332331,0.00003709049,0.003243719,0.5955685,0.001665358,0.003914558,0.02820218,0.0002858248],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9841619,0.0007611333,0.0003891916,0.003591303,0.001418871,0.004702064,0.00002095376,0.00001855923,0.004936033],"genre_scores_gemma":[0.9929133,0.00001251195,0.005668637,0.0002359686,0.0006810112,0.0002514141,0.00002162595,0.00001988765,0.0001956086],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8658536,"threshold_uncertainty_score":0.9992355,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3502415788863397,"score_gpt":0.5226132347110275,"score_spread":0.1723716558246878,"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."}}