{"id":"W4248200253","doi":"","title":"RESEARCH AND TEACHING Collaborative Testing: Evidence of Learning in a Controlled In-Class Study of Undergraduate Students","year":2014,"lang":"en","type":"article","venue":"","topic":"Reflective Practices in Education","field":"Social Sciences","cited_by":96,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Mathematics education; Class (philosophy); Teaching method; Psychology; Science education; Pedagogy; Computer science","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01776425,0.0000592282,0.0002753108,0.0003322706,0.0002175364,0.00006751384,0.0002395426,0.00003932607,0.000003326412],"category_scores_gemma":[0.04653978,0.0000529028,0.000007455847,0.0009857997,0.0002310463,0.0004503482,0.0001038945,0.0004520445,9.718146e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001745698,"about_ca_system_score_gemma":0.000287644,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.01393794,"about_ca_topic_score_gemma":0.02673607,"domain_scores_codex":[0.9915886,0.00681951,0.0003496467,0.0002214549,0.0008244865,0.0001962924],"domain_scores_gemma":[0.9858603,0.01312153,0.0002653612,0.0001033958,0.0006117467,0.00003761249],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001983688,0.0007667076,0.9100277,0.00001280754,0.00001273578,4.093075e-7,0.07831139,0.0001937066,0.0007928376,0.007974841,0.000002808298,0.001705637],"study_design_scores_gemma":[0.003778855,0.002130816,0.5240383,0.0003310924,0.00001162742,1.177709e-7,0.4598508,0.001171947,0.00008011831,0.008441451,0.0000670775,0.00009777107],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9376299,0.00004609759,0.00002725367,0.0009481746,0.00004189749,0.0009360143,3.040458e-8,0.000007639845,0.06036304],"genre_scores_gemma":[0.9987633,0.00003008497,0.0005954129,0.000005018633,0.00001988486,0.00005480816,3.995537e-8,0.000004544788,0.0005269212],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3859895,"threshold_uncertainty_score":0.9926283,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1504244910715768,"score_gpt":0.5438010138242099,"score_spread":0.393376522752633,"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."}}