{"id":"W1982497397","doi":"10.1016/s0191-491x(02)00014-7","title":"Matching the grade 8 TIMSS item pool to the Ontario curriculum","year":2002,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":2,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto; Lakehead University","funders":"","keywords":"Mathematics education; Curriculum; Matching (statistics); Secondary education; Psychology; Pedagogy; Mathematics; Statistics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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":[0.002452534,0.00008436642,0.00008634964,0.00005343385,0.001055732,0.00007504377,0.0003132552,0.00002526144,0.001230334],"category_scores_gemma":[0.0006635452,0.00005228855,0.00003946322,0.0004400446,0.0001416056,0.0001502098,0.00006780592,0.0001341834,0.0002568586],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009908624,"about_ca_system_score_gemma":0.0002079393,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.01374872,"about_ca_topic_score_gemma":0.2292784,"domain_scores_codex":[0.9980574,0.0003543982,0.000195315,0.0001731963,0.001015213,0.0002044794],"domain_scores_gemma":[0.9988517,0.0006296581,0.00007524994,0.0001414274,0.0002733558,0.00002860148],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.000002029562,0.0002044615,0.09309825,0.000003202017,0.00006308185,1.091562e-7,0.7985937,0.000660968,0.000003020467,0.02902945,0.07384103,0.004500731],"study_design_scores_gemma":[0.0002427172,0.0000213347,0.5927417,0.00004426318,0.0000656798,5.644851e-7,0.3024482,0.0004367595,0.000001139861,0.02763575,0.07619528,0.0001666791],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8436687,0.001187729,0.000009681055,0.114558,0.002989085,0.001091214,0.000001021848,0.00001371278,0.03648089],"genre_scores_gemma":[0.9904101,0.000130586,0.0001660917,0.0007760128,0.00101876,0.0004442054,0.00000529327,0.000004617453,0.007044313],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4996434,"threshold_uncertainty_score":0.9996827,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1656116291663467,"score_gpt":0.4685884905270746,"score_spread":0.3029768613607279,"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."}}