{"id":"W4413838709","doi":"10.24908/iqurcp19869","title":"Predictive Models for Grade 9 Mathematics Stream: Controlling for Past EQAO Performance","year":2025,"lang":"en","type":"article","venue":"Inquiry Queen s Undergraduate Research Conference Proceedings","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Mathematics education; Computer science; Mathematics","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":["metaepi_narrow","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.007419301,0.000325834,0.0005733906,0.0009480366,0.000920366,0.001442249,0.001487683,0.0001637375,0.00001284911],"category_scores_gemma":[0.002745288,0.0002569467,0.000198352,0.001264684,0.0005284106,0.001558111,0.0003211691,0.0004302644,0.00001946626],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003525768,"about_ca_system_score_gemma":0.001242749,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002639128,"about_ca_topic_score_gemma":0.000005774156,"domain_scores_codex":[0.9943725,0.00006116804,0.001095712,0.001018449,0.002391867,0.001060325],"domain_scores_gemma":[0.9866835,0.004028321,0.0003557045,0.0003617657,0.008337327,0.000233403],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.000427929,0.0005942207,0.003101767,0.0004884129,0.0001879689,2.282188e-7,0.003992164,0.000315324,0.001961781,0.9269264,0.04845487,0.01354891],"study_design_scores_gemma":[0.001042344,0.0004943704,0.0002189131,0.0002210444,0.00002559665,5.462649e-7,0.01087087,0.3150448,0.002224982,0.6654795,0.004193868,0.0001831785],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.09649929,0.00008688183,0.8198314,0.06896232,0.000772191,0.005735391,0.000102043,0.0001290827,0.00788136],"genre_scores_gemma":[0.9751332,0.0001733266,0.01164069,0.0001704748,0.0003764468,0.002747677,0.00002708638,0.00002940792,0.009701692],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8786339,"threshold_uncertainty_score":0.9999883,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3118884900090728,"score_gpt":0.484923408079207,"score_spread":0.1730349180701342,"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."}}