{"id":"W4407378596","doi":"10.5430/jct.v14n1p174","title":"Assessing Students' Progress in Chemistry: Using Multiple-Choice Questions and Performance-Based Assessments","year":2025,"lang":"en","type":"article","venue":"Journal of Curriculum and Teaching","topic":"Various Chemistry Research Topics","field":"Chemistry","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"King Faisal University","keywords":"Mathematics education; Computer science; Management science; Chemistry; Psychology; Engineering","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007001818,0.0001645276,0.000274567,0.0001011664,0.0002467968,0.0004141155,0.0002327556,0.0001163208,0.00001358147],"category_scores_gemma":[0.000523138,0.0001484828,0.00004885792,0.0000959723,0.00008023365,0.0004673568,0.0001346544,0.001064124,5.995977e-8],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002127857,"about_ca_system_score_gemma":0.0002158529,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004970454,"about_ca_topic_score_gemma":0.00000447058,"domain_scores_codex":[0.9986506,0.00005351838,0.000446801,0.0001887096,0.0003940818,0.0002663245],"domain_scores_gemma":[0.999135,0.0002217124,0.0002769444,0.0001335663,0.0001096117,0.0001231768],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001164493,0.0002770527,0.9651165,0.0006056401,0.0000448979,0.00002980977,0.00005147574,0.00007021506,0.01199524,0.000007185825,0.0000074517,0.02178287],"study_design_scores_gemma":[0.01209372,0.0001283026,0.5954067,0.01454323,0.0004101338,0.000529447,0.003544143,0.3363517,0.03294454,0.0001529013,0.002835631,0.001059549],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9972853,0.00101812,0.0003889945,0.00009404841,0.00004723846,0.00004371868,0.000001523383,0.00001423875,0.001106782],"genre_scores_gemma":[0.9969013,0.00004644322,0.002823194,0.0000149629,0.0001176081,0.000003601517,0.000002602749,0.0000111505,0.00007918288],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3697098,"threshold_uncertainty_score":0.6054952,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02367418738733538,"score_gpt":0.3899725514857401,"score_spread":0.3662983640984047,"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."}}