{"id":"W1971150495","doi":"10.1119/1.2999067","title":"Physics Exam Problems Reconsidered: <i>Using Logger Pro to Evaluate Student Understanding of Physics</i>","year":2008,"lang":"en","type":"article","venue":"The Physics Teacher","topic":"Innovative Teaching Methods","field":"Social Sciences","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia; Toronto Metropolitan University","funders":"","keywords":"Interactivity; Physics education; Mathematics education; Test (biology); Physics; Computer science; Multimedia; Psychology; Ecology","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":[],"consensus_categories":[],"category_scores_codex":[0.004208714,0.0002869996,0.0004451392,0.00003914597,0.001201165,0.00005964096,0.000606109,0.00008481951,0.00003766779],"category_scores_gemma":[0.0001997981,0.0002322669,0.0001569929,0.001049862,0.0008021912,0.00033841,0.0002111939,0.0005644758,0.0000326869],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006065737,"about_ca_system_score_gemma":0.0003712624,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009100915,"about_ca_topic_score_gemma":0.00004327063,"domain_scores_codex":[0.9957532,0.001778591,0.000422675,0.00041589,0.001014606,0.000615],"domain_scores_gemma":[0.9982564,0.0003863257,0.0003933531,0.0005819356,0.0002879401,0.00009403747],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00005537624,0.00141391,0.01912777,0.00007248233,0.0005270273,0.000005796667,0.5681566,0.01080589,0.05017469,0.3208702,0.003023285,0.02576698],"study_design_scores_gemma":[0.00172842,0.0003131208,0.00332026,0.0004134703,0.0003920914,0.000007401318,0.0338254,0.002780369,0.03290018,0.9197904,0.002818412,0.001710505],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6759531,0.00005214724,0.2831519,0.0009014581,0.0005043201,0.001519703,0.000007682416,0.000177745,0.03773195],"genre_scores_gemma":[0.9889819,0.00001217155,0.008360134,0.0002715883,0.001201826,0.00003987636,0.000002493887,0.00006490931,0.001065086],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5989202,"threshold_uncertainty_score":0.9471568,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4452701209894344,"score_gpt":0.4505638954160032,"score_spread":0.005293774426568754,"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."}}