{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0105467,0.0006825445,0.0005972415,0.003860997,0.005355785,0.005139723,0.002377384,0.001506167,0.005818655],"category_scores_gemma":[0.04068341,0.0002484823,0.001302807,0.01036638,0.003639786,0.002406006,0.002706986,0.0018414,0.0003343255],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.1103435,"about_ca_system_score_gemma":0.06746984,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9826295,"about_ca_topic_score_gemma":0.9849522,"domain_scores_codex":[0.9892485,0.00458226,0.0003610726,0.00073086,0.003844868,0.001232315],"domain_scores_gemma":[0.9792506,0.009899421,0.001220118,0.0009344192,0.007961648,0.000733807],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.000289521,0.0001801458,0.04417355,0.0004593178,0.0002541803,0.0004381993,0.01022513,0.04370876,0.0003205993,0.8005106,0.02639219,0.07304771],"study_design_scores_gemma":[0.0004157103,0.0003742694,0.1313183,0.001071325,0.0007226943,0.0002628557,0.02714531,0.2596917,0.001436517,0.2877063,0.2893351,0.0005199522],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.3882006,0.007326867,0.05728764,0.03371048,0.0004013676,0.0007857677,0.007453845,0.0004102864,0.5044231],"genre_scores_gemma":[0.958296,0.001879924,0.02411207,0.0006403537,0.00002674448,0.0001475293,0.00104055,0.00006697733,0.01378989],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.1103435,"threshold_uncertainty_score":0.8006019,"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."}}