{"id":"W2284581455","doi":"10.1080/14926156.2015.1119332","title":"Canadian and United States Students' Performances on the OECD's PISA 2012 Problem-Solving Assessment","year":2015,"lang":"en","type":"article","venue":"Canadian Journal of Science Mathematics and Technology Education","topic":"Education and Critical Thinking Development","field":"Social Sciences","cited_by":8,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"","keywords":"Mathematics education; Set (abstract data type); Reading (process); Scale (ratio); Psychology; Computer science; Political science; Geography","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"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":[0.002762125,0.00006296323,0.00008326363,0.0008795157,0.001131217,0.0003325763,0.0004029485,0.00005244766,0.00004848934],"category_scores_gemma":[0.0004454672,0.00004432767,0.000007324529,0.0006529757,0.001427805,0.0002326974,0.00001500401,0.0001873203,0.000004895696],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005461046,"about_ca_system_score_gemma":0.01299469,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.02856268,"about_ca_topic_score_gemma":0.1305671,"domain_scores_codex":[0.9990993,0.00002607297,0.0001855566,0.00008925813,0.0003234897,0.0002763739],"domain_scores_gemma":[0.9985399,0.00007224843,0.00011196,0.00008877241,0.0005161765,0.00067095],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","study_design_scores_codex":[4.029309e-7,0.00006147045,0.0337993,0.00001352623,0.00001068258,0.000001714388,0.06748558,0.00000287688,0.000008259621,0.8831789,0.00540203,0.01003524],"study_design_scores_gemma":[0.0001259458,0.0001362408,0.006590124,0.0002491658,0.00002126224,0.00004474888,0.46215,0.0001553352,0.00004187005,0.4054782,0.1248217,0.0001854677],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8906612,0.0002088059,0.00003099795,0.1026007,0.0004364422,0.0001486283,9.896528e-7,0.000006135469,0.005906012],"genre_scores_gemma":[0.9925936,0.0001352364,0.006043396,0.0008105998,0.00003975658,0.000006680335,7.207672e-7,0.000003470293,0.0003665337],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4777007,"threshold_uncertainty_score":0.9926007,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.025483940492606,"score_gpt":0.3353345004961147,"score_spread":0.3098505600035087,"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."}}