{"id":"W48175147","doi":"","title":"Assessment of Thinking Levels in Students' Answers","year":2001,"lang":"en","type":"article","venue":"Academic exchange quarterly","topic":"Educational Assessment and Pedagogy","field":"Social Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Critical thinking; Psychology; Set (abstract data type); Higher-order thinking; Mathematics education; Construct (python library); Pedagogy; Teaching method; Computer science","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001772143,0.000100762,0.0001804846,0.0001830665,0.0001386349,0.00002642633,0.0005569245,0.0001788252,0.000692707],"category_scores_gemma":[0.00001864936,0.0001057633,0.000046124,0.0004089925,0.0001090043,0.0003432079,0.00001960553,0.0003577691,0.00001569874],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001725827,"about_ca_system_score_gemma":0.0002676956,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002629007,"about_ca_topic_score_gemma":0.003329574,"domain_scores_codex":[0.9980925,0.0002539449,0.0003338899,0.0002072473,0.0007706285,0.0003417616],"domain_scores_gemma":[0.9993395,0.0002385181,0.000155471,0.0001206091,0.00006265331,0.00008331418],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00000306037,0.0000826672,0.8672316,0.00001208786,0.00001238084,0.00000387413,0.09328637,0.000001113232,0.0001862791,0.02597864,0.0004988491,0.01270305],"study_design_scores_gemma":[0.0003957562,0.0001359265,0.8934211,0.00007370213,0.00001146349,0.000001129591,0.05777081,0.00001391382,0.000004727037,0.01559342,0.03240641,0.0001716829],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9650941,0.0003116383,0.0002386235,0.00343102,0.0004516803,0.0002605898,0.000004668234,0.00003203287,0.03017559],"genre_scores_gemma":[0.9960448,0.0003569753,0.0003241696,0.000388492,0.00044212,0.00003559527,0.000006958276,0.000008903645,0.002392035],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03551556,"threshold_uncertainty_score":0.7584656,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07763424570977516,"score_gpt":0.4663400536234029,"score_spread":0.3887058079136277,"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."}}