{"id":"W2524551678","doi":"","title":"THE STANDARDIZED TESTING MOVEMENT: EQUITABLE OR EXCESSIVE?","year":2003,"lang":"en","type":"article","venue":"McGill Journal of Education / Revue des sciences de l'éducation de McGill","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Context (archaeology); Humanities; Legitimacy; Political science; Philosophy; Politics; History; Law","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02745824,0.000199369,0.0003054627,0.0005081816,0.004383992,0.000643018,0.001284774,0.00007214444,0.001103361],"category_scores_gemma":[0.02541251,0.0001211898,0.0001470039,0.002668093,0.000504285,0.001539953,0.00006622697,0.0002634762,0.00006184585],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001542244,"about_ca_system_score_gemma":0.004375835,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003417951,"about_ca_topic_score_gemma":0.0001426869,"domain_scores_codex":[0.9946625,0.00107876,0.001485729,0.0003665567,0.001788022,0.0006184164],"domain_scores_gemma":[0.9908094,0.003156068,0.001862386,0.0004588363,0.003395205,0.0003180687],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0003464793,0.001642654,0.03805231,0.0001290661,0.0001451228,0.00001472107,0.009372438,0.03430199,0.008680957,0.297279,0.06773154,0.5423037],"study_design_scores_gemma":[0.0008464541,0.0007651277,0.009549249,0.0002802129,0.00007017166,0.0007970765,0.03186158,0.004853933,0.006068015,0.2509913,0.6935078,0.0004090153],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6267589,0.002144036,0.01670101,0.01763676,0.01415892,0.001132011,0.00007338986,0.000076319,0.3213187],"genre_scores_gemma":[0.9342424,0.0003617944,0.05046003,0.002886525,0.0002647456,0.00004848403,0.000001295926,0.00001936618,0.01171542],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6257763,"threshold_uncertainty_score":0.9998097,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6006801436383168,"score_gpt":0.5533757529947877,"score_spread":0.04730439064352909,"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."}}