{"id":"W1228266688","doi":"","title":"Comparing Performance-Based Accountability Models: A Canadian Example.","year":2008,"lang":"en","type":"article","venue":"Canadian Journal of Education / Revue canadienne de l éducation","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":21,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Humanities; Accountability; Political science; Sociology; Ethnology; Psychology; Art; Law","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"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":[0.003403952,0.0002172504,0.0003544518,0.001975502,0.0007350708,0.00025633,0.0009725226,0.000116096,0.0007617107],"category_scores_gemma":[0.001491041,0.0002197307,0.0001455591,0.001159536,0.0001498509,0.001197835,0.000008537944,0.0003409036,0.00004872222],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.009801748,"about_ca_system_score_gemma":0.08599338,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.7473954,"about_ca_topic_score_gemma":0.9846091,"domain_scores_codex":[0.9971046,0.000196206,0.001115756,0.0003798371,0.0003689342,0.0008346843],"domain_scores_gemma":[0.9924865,0.0003087368,0.000705794,0.0006383871,0.002837514,0.003023082],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001845886,0.0001718135,0.8658615,0.0000429278,0.00003779216,0.00001023053,0.045338,0.007755343,0.00006534655,0.004462425,0.06393807,0.01229814],"study_design_scores_gemma":[0.0004213532,0.0001734003,0.8220104,0.0001517995,0.00004713744,0.0004563185,0.03433008,0.01060693,0.00009891044,0.01210638,0.1191138,0.0004834252],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9825335,0.0001995239,0.0004470626,0.005715321,0.005056872,0.0002499535,0.00001776091,0.000005749052,0.00577422],"genre_scores_gemma":[0.9950476,0.00001355193,0.001281353,0.001531639,0.0007056022,0.00003607054,0.00005372826,0.00002242496,0.001308087],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2372137,"threshold_uncertainty_score":0.9939994,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3376322216252052,"score_gpt":0.3619103631765933,"score_spread":0.02427814155138808,"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."}}