{"id":"W4409741900","doi":"10.5220/0013491000003932","title":"Banking Strategies and Software Solutions for Generated Test Items","year":2025,"lang":"en","type":"article","venue":"","topic":"Software System Performance and Reliability","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Computer science; Test (biology); Software; Software engineering; Operating system","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":[0.002671896,0.0005444158,0.0001825498,0.002865632,0.0006969373,0.004635351,0.001358574,0.001683086,0.008143128],"category_scores_gemma":[0.0293397,0.0003508636,0.000370613,0.001096853,0.0005267647,0.002768973,0.0008417804,0.000634483,0.002198659],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001567974,"about_ca_system_score_gemma":0.001838266,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001984865,"about_ca_topic_score_gemma":0.003389856,"domain_scores_codex":[0.995842,0.002093047,0.000267362,0.0002771828,0.001077784,0.0004425658],"domain_scores_gemma":[0.9873467,0.006176722,0.001556562,0.0007946118,0.003664007,0.0004613248],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001179241,0.001215142,0.1119438,0.0004486548,0.0001842516,0.001873684,0.002690314,0.03020662,0.02077237,0.1731886,0.01561684,0.6406805],"study_design_scores_gemma":[0.0005177258,0.003583188,0.166107,0.001544347,0.0007645864,0.007112841,0.01493344,0.4273417,0.09280069,0.2265338,0.05843828,0.000322485],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7652727,0.001122748,0.1172092,0.005084778,0.0001370793,0.0004996249,0.000306675,0.0008551839,0.109512],"genre_scores_gemma":[0.9647029,0.0001742485,0.02783766,0.0003157571,0.00001484407,0.00007020503,0.0001305141,0.00008170197,0.006672176],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.008143128,"threshold_uncertainty_score":0.02724147,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01966233747416383,"score_gpt":0.2610456326138477,"score_spread":0.2413832951396838,"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."}}