{"id":"W2079281756","doi":"10.1119/1.3602073","title":"The puzzling reliability of the Force Concept Inventory","year":2011,"lang":"en","type":"article","venue":"American Journal of Physics","topic":"Science Education and Pedagogy","field":"Social Sciences","cited_by":80,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; Concordia University; Vanier College; John Abbott College","funders":"","keywords":"Reliability (semiconductor); Internal consistency; Consistency (knowledge bases); Test (biology); Physics; Statistics; Psychology; Reliability engineering; Econometrics; Psychometrics; Mathematics; Computer science; Thermodynamics; Artificial intelligence; Engineering","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":[0.001228138,0.00003548845,0.00009297202,0.000008548584,0.0003868161,0.00001404837,0.0005353354,0.000009468662,0.00003279871],"category_scores_gemma":[0.0004744758,0.00001916638,0.0001025738,0.0003335838,0.002244723,0.0001351476,0.00002072139,0.0001462977,0.000001733624],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004026445,"about_ca_system_score_gemma":0.0005958698,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009043594,"about_ca_topic_score_gemma":0.00008718453,"domain_scores_codex":[0.9989629,0.0003438987,0.0001957969,0.00004803967,0.0003248124,0.0001245777],"domain_scores_gemma":[0.998703,0.0002505124,0.0005711085,0.0001518008,0.0002600635,0.00006353307],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00004137483,0.0001926703,0.1327602,0.000003052231,0.00003198637,3.92731e-7,0.5057667,0.0001025868,0.0003412266,0.1381315,0.003054895,0.2195735],"study_design_scores_gemma":[0.000318541,0.0004851849,0.09473364,0.00006370051,0.00006608215,0.000004568064,0.6871097,0.00008411415,0.006737177,0.1303347,0.07981579,0.0002467928],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9859213,0.00005462919,0.0004962714,0.002400641,0.001144556,0.00005707782,6.730554e-7,0.000003313281,0.0099215],"genre_scores_gemma":[0.9989863,0.00003305777,0.000159999,0.0002077191,0.000313794,4.945414e-7,2.346646e-8,0.00000221497,0.0002963758],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2193267,"threshold_uncertainty_score":0.8270777,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06388540973370929,"score_gpt":0.3546067386217578,"score_spread":0.2907213288880485,"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."}}