{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002631714,0.0006282787,0.000423184,0.00170126,0.0006705269,0.002038268,0.0006795514,0.001252129,0.004295046],"category_scores_gemma":[0.01693346,0.0002371278,0.0004761043,0.0007514127,0.0006212312,0.001670458,0.001061815,0.001024526,0.00247639],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008137852,"about_ca_system_score_gemma":0.0006950939,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00239605,"about_ca_topic_score_gemma":0.005407903,"domain_scores_codex":[0.9976212,0.0008498975,0.0003857878,0.0003021967,0.0007095357,0.0001313679],"domain_scores_gemma":[0.9895691,0.002731572,0.003071462,0.0006093961,0.002571982,0.001446438],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002440646,0.004300444,0.8148153,0.0002089375,0.0001701161,0.000479323,0.002739568,0.0005676583,0.006002318,0.0004675437,0.01580114,0.152007],"study_design_scores_gemma":[0.0001681038,0.003963802,0.971247,0.00007154903,0.00006323206,0.0007645096,0.00222134,0.003415271,0.008968892,0.0005987286,0.008436866,0.000080618],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9845594,0.0001980215,0.00176693,0.000527175,0.00009745551,0.0003711057,0.0007064595,0.000294562,0.011479],"genre_scores_gemma":[0.987227,0.0001711021,0.005431714,0.0004090067,0.00006429254,0.0005389834,0.001168255,0.00005609171,0.00493349],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.004295046,"threshold_uncertainty_score":0.01436836,"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."}}