{"id":"W3004171085","doi":"10.1039/c9rp00274j","title":"Evaluating students’ learning gains, strategies, and errors using OrgChem101's module: organic mechanisms—mastering the arrows","year":2020,"lang":"en","type":"article","venue":"Chemistry Education Research and Practice","topic":"Innovative Teaching and Learning Methods","field":"Psychology","cited_by":25,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Metacognition; Formalism (music); Mathematics education; Computer science; Test (biology); Psychology; Cognition","routes":{"ca_aff":true,"ca_fund":true,"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":[0.009953252,0.0001610136,0.0001505236,0.00004368628,0.0009492785,0.0005434987,0.0002795284,0.0001144935,0.0007194234],"category_scores_gemma":[0.01109745,0.0001400315,0.00001607462,0.0005455057,0.0001797993,0.0003585443,0.0002142882,0.001970717,0.00001877982],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007618826,"about_ca_system_score_gemma":0.0005698625,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001341933,"about_ca_topic_score_gemma":5.383548e-7,"domain_scores_codex":[0.9953963,0.002909445,0.0002520255,0.0004999845,0.000546146,0.0003960789],"domain_scores_gemma":[0.9967026,0.002175804,0.0002092474,0.0002514479,0.0004825248,0.0001783458],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"qualitative","study_design_scores_codex":[0.0003247223,0.0002095808,0.001237489,0.0002244989,0.0001367033,0.000008515552,0.03638213,0.00008845414,0.9291129,0.001403856,0.0004153128,0.03045584],"study_design_scores_gemma":[0.001448332,0.0007593666,0.006474452,0.0002360348,0.0001404576,0.0005923533,0.9192547,0.005136306,0.02002836,0.002222279,0.04294655,0.0007607898],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9729117,0.0009653268,0.007868498,0.005570482,0.0001436501,0.000241282,7.744993e-7,0.00005820844,0.01224009],"genre_scores_gemma":[0.9846652,0.00003956198,0.0120839,0.0004184377,0.000405127,0.00003637774,0.00001311002,0.00003588623,0.002302414],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9090846,"threshold_uncertainty_score":0.9972325,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3986438866750975,"score_gpt":0.5921792560203151,"score_spread":0.1935353693452176,"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."}}