{"id":"W2057277282","doi":"10.1119/1.4849159","title":"Physics Exams that Promote Collaborative Learning","year":2013,"lang":"en","type":"article","venue":"The Physics Teacher","topic":"Experimental Learning in Engineering","field":"Engineering","cited_by":65,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Formative assessment; Summative assessment; Mathematics education; Consistency (knowledge bases); Physics education; Collaborative learning; Computer science; Medical education; Student engagement; Psychology; Medicine; Artificial intelligence","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"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":[0.004786264,0.0007946198,0.0005571385,0.001310181,0.0009452082,0.00262251,0.001742785,0.001095627,0.01964941],"category_scores_gemma":[0.02241412,0.0003764314,0.0005302113,0.000998577,0.0005735032,0.001479313,0.004577407,0.001417223,0.005650821],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003945388,"about_ca_system_score_gemma":0.0007875522,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001750816,"about_ca_topic_score_gemma":0.000527949,"domain_scores_codex":[0.9953939,0.001838056,0.0003796514,0.0005693474,0.001424677,0.0003944856],"domain_scores_gemma":[0.9826519,0.009032187,0.001426264,0.00238608,0.001398198,0.003105305],"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":[0.0008949743,0.01529712,0.01355788,0.001279756,0.0001045379,0.001013801,0.006654267,0.003987492,0.05236537,0.02135736,0.03339837,0.8500891],"study_design_scores_gemma":[0.001778263,0.01880662,0.1326803,0.0007517902,0.0002040272,0.00314196,0.003087938,0.0253926,0.08484485,0.04221929,0.686697,0.0003954152],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4238568,0.0006174296,0.4295055,0.001650458,0.001053786,0.007951716,0.0008492926,0.008857098,0.1256579],"genre_scores_gemma":[0.5142758,0.0004988961,0.4206285,0.0006468472,0.0003789726,0.007454633,0.001092412,0.0005802538,0.05444362],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01964941,"threshold_uncertainty_score":0.06573385,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01094219117019928,"score_gpt":0.2093543674537619,"score_spread":0.1984121762835626,"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."}}