{"id":"W4403031022","doi":"10.1007/s42330-024-00333-1","title":"A Tale of Two Exams","year":2024,"lang":"en","type":"article","venue":"Canadian Journal of Science Mathematics and Technology Education","topic":"Mathematics Education and Teaching Techniques","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Red Deer Polytechnic","funders":"","keywords":"Science education; Mathematics education; Sociology; Pedagogy; Psychology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.001989534,0.0009332271,0.0008007869,0.002216483,0.0109339,0.007541689,0.001564578,0.003521472,0.06332377],"category_scores_gemma":[0.01062306,0.0004343495,0.001165933,0.001147742,0.001980276,0.003282849,0.005423159,0.005374125,0.01236142],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003358378,"about_ca_system_score_gemma":0.004141637,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02119606,"about_ca_topic_score_gemma":0.05670013,"domain_scores_codex":[0.9979019,0.000360226,0.00006771461,0.0002105356,0.0008045386,0.0006551672],"domain_scores_gemma":[0.9928947,0.000571155,0.0001275624,0.0003219409,0.002253239,0.00383128],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0003873927,0.0005404401,0.005216766,0.0001122967,0.00006869181,0.003805297,0.006448081,0.0006143333,0.002077301,0.05920245,0.7939309,0.1275959],"study_design_scores_gemma":[0.00004838162,0.000372366,0.00956575,0.0001718143,0.00002505213,0.001550219,0.00895892,0.001239682,0.0009051994,0.02642696,0.9506139,0.0001217845],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.10349,0.003703751,0.01581488,0.1874157,0.04078105,0.0006127195,0.002036709,0.002338358,0.6438068],"genre_scores_gemma":[0.2101942,0.0007008947,0.009742442,0.008758641,0.00284265,0.00008832666,0.0008149183,0.0003796584,0.7664783],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9788039,"threshold_uncertainty_score":0.2118391,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01920645681476948,"score_gpt":0.3485168810292631,"score_spread":0.3293104242144936,"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."}}