{"id":"W4366723551","doi":"10.3138/cjpe.0020.009","title":"Evaluating School Improvement in Canada: A Case Example","year":2006,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Shared Health","funders":"","keywords":"Work (physics); Action (physics); Management science; Psychology; Engineering ethics; Process management; Political science; Business; Engineering","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02104518,0.0001425522,0.0002697275,0.0007503525,0.0001995304,0.0003344401,0.0003589426,0.0000512395,0.003570711],"category_scores_gemma":[0.002657804,0.0001173462,0.00008007522,0.00100441,0.00003546239,0.0005543559,0.00001420156,0.0002607215,0.00003060802],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.004247749,"about_ca_system_score_gemma":0.05314429,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9705993,"about_ca_topic_score_gemma":0.999382,"domain_scores_codex":[0.9940488,0.0005782992,0.001612624,0.0002571431,0.003081815,0.0004212584],"domain_scores_gemma":[0.9954124,0.0002972305,0.0008581113,0.0003178005,0.002619562,0.0004948989],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00001093807,0.00003249814,0.1229409,0.000003672855,0.00001002703,0.0002746037,0.0001926042,0.02397572,0.0001034844,0.0001382924,0.003516437,0.8488008],"study_design_scores_gemma":[0.003791447,0.000992781,0.5244331,0.00009124217,0.0001041167,0.001143139,0.0053563,0.429577,0.000272111,0.01284562,0.02098416,0.0004089023],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9946879,0.000345798,0.0002299345,0.001040399,0.000714624,0.00115184,0.00001031474,0.000003799121,0.001815371],"genre_scores_gemma":[0.9959395,0.00000149947,0.003386488,0.000205374,0.000208081,0.0001003519,0.00001180367,0.000009912656,0.0001369973],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.848392,"threshold_uncertainty_score":0.9995748,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4300939707557752,"score_gpt":0.5277537046643593,"score_spread":0.09765973390858412,"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."}}