{"id":"W4388813874","doi":"10.29221/jce.2023.26.4.113","title":"A study on the evaluation of foundational literacy with the case of BC, Canada and Australia","year":2023,"lang":"en","type":"article","venue":"Journal of Curriculum and Evaluation","topic":"Educational Research and Pedagogy","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Numeracy; Literacy; Curriculum; Mathematics education; Context (archaeology); Meaning (existential); Pedagogy; Psychology; Geography","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":[],"consensus_categories":[],"category_scores_codex":[0.00618343,0.00005117098,0.00008284357,0.00009271756,0.0001128216,0.00005685101,0.0001426336,0.00001109356,0.00003568261],"category_scores_gemma":[0.0004938446,0.00002475682,0.0000158632,0.0003520646,0.00004051374,0.0002342402,0.00002659941,0.00009277058,6.306072e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005373751,"about_ca_system_score_gemma":0.001274259,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.01905928,"about_ca_topic_score_gemma":0.04653058,"domain_scores_codex":[0.9977378,0.0005285253,0.0002292684,0.00007953611,0.001348555,0.00007632486],"domain_scores_gemma":[0.9971553,0.0005063138,0.0002905526,0.0001219884,0.001882979,0.00004288456],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002952439,0.001803509,0.4315018,0.000165809,0.0009519407,0.0002137823,0.06861186,0.0216124,0.0006188906,0.05008565,0.05016083,0.3739783],"study_design_scores_gemma":[0.00103065,0.0008616478,0.8860127,0.0000625455,0.0001271837,0.0002928973,0.009624942,0.09795606,0.00008509497,0.003769323,0.0001121288,0.00006487613],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9955332,0.00007158573,0.0001452458,0.003817942,0.00009672157,0.0002869037,0.000002073804,9.207602e-7,0.00004543432],"genre_scores_gemma":[0.9997846,0.00000851372,0.0000839034,0.00002556941,0.00005693455,0.00001338265,0.00000173913,0.000001525175,0.00002384425],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4545109,"threshold_uncertainty_score":0.9874729,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1700989173492516,"score_gpt":0.4656078199469245,"score_spread":0.2955089025976728,"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."}}