{"id":"W3046538170","doi":"10.1111/emip.12382","title":"Synergy and Tension between Large‐Scale and Classroom Assessment: International Trends","year":2020,"lang":"en","type":"article","venue":"Educational Measurement Issues and Practice","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":21,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University; Brock University","funders":"","keywords":"Scale (ratio); Test (biology); Political science; Mathematics education; Pedagogy; Sociology; Psychology; Geography; Geology; Cartography","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0460766,0.0001960899,0.0005050337,0.004356804,0.0009616507,0.006938803,0.001265186,0.0009419395,0.002572443],"category_scores_gemma":[0.05747416,0.0003108683,0.0002128058,0.009207913,0.006573258,0.008068855,0.005446479,0.00269738,0.0002373255],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004967979,"about_ca_system_score_gemma":0.007635783,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01427181,"about_ca_topic_score_gemma":0.01366717,"domain_scores_codex":[0.9779327,0.009496402,0.002022549,0.003315778,0.005945744,0.00128688],"domain_scores_gemma":[0.7971092,0.1286396,0.01820693,0.008969309,0.04136628,0.005708543],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0002106703,0.0002017055,0.2032793,0.002084328,0.0001108703,0.0002203676,0.04126216,0.001305704,0.001764698,0.09733063,0.006467875,0.6457617],"study_design_scores_gemma":[0.00002368232,0.0004931898,0.7198912,0.004646169,0.00007877697,0.0009358522,0.1028281,0.002212136,0.0020134,0.02183255,0.144895,0.0001499825],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6923557,0.08575698,0.01210072,0.1065014,0.0005356059,0.00008126989,0.0004144431,0.0001727623,0.1020812],"genre_scores_gemma":[0.9889381,0.007133181,0.001986147,0.00106264,0.0001319265,0.00002246577,0.0000621931,0.00002624796,0.0006370611],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0460766,"threshold_uncertainty_score":0.2436792,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09632131387335943,"score_gpt":0.4206769110651631,"score_spread":0.3243555971918037,"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."}}