{"id":"W2212108873","doi":"10.55016/ojs/ajer.v52i2.55134","title":"A Special Education Program Evaluation for Accountability Purposes: An In-Depth Case Study","year":2006,"lang":"en","type":"article","venue":"Alberta Journal of Educational Research","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"","keywords":"Accountability; Psychology; Mathematics education; Program evaluation; Pedagogy; Special education; Educational research; Evaluation methods; Sociology; Political science; Public administration; Engineering","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"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":[0.01157049,0.0006038236,0.0005800339,0.001951016,0.013279,0.002811978,0.002080128,0.004275039,0.00363344],"category_scores_gemma":[0.02848953,0.0005800334,0.0006265996,0.001873411,0.002699617,0.002299305,0.004020939,0.004015279,0.0003735645],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01110419,"about_ca_system_score_gemma":0.01105993,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02469294,"about_ca_topic_score_gemma":0.07692418,"domain_scores_codex":[0.9864817,0.009264503,0.0003893746,0.0003641585,0.001280133,0.002220099],"domain_scores_gemma":[0.9688559,0.01723271,0.001973797,0.00111316,0.004280844,0.006543525],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.001822832,0.03692823,0.1773185,0.001269594,0.0001776296,0.1088185,0.4042824,0.006775141,0.005370524,0.02698971,0.01972506,0.2105219],"study_design_scores_gemma":[0.0004299227,0.009396638,0.07553933,0.001136095,0.0001830574,0.03348365,0.7846856,0.01086353,0.008623526,0.005415494,0.06998059,0.0002624899],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9780068,0.0002140541,0.003456283,0.003919574,0.00007172897,0.0007011446,0.00008839599,0.00002932037,0.01351255],"genre_scores_gemma":[0.9879288,0.0002966691,0.004385525,0.000740409,0.00003222481,0.0002931894,0.00004055642,0.00002439714,0.006258176],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02469294,"threshold_uncertainty_score":0.08056682,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4967872139392358,"score_gpt":0.6637065650467828,"score_spread":0.1669193511075471,"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."}}