{"id":"W2114622266","doi":"10.1109/icsmc.1994.400190","title":"Quantitative evaluation of expert systems","year":2002,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Concordia University","funders":"","keywords":"Computer science; Expert system; Quality (philosophy); Quality assurance; Data mining; 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0272713,0.001017655,0.0007737395,0.007088732,0.0006331113,0.003103412,0.0009777974,0.001038454,0.004869533],"category_scores_gemma":[0.1652508,0.0002365349,0.0004762541,0.003920009,0.001318631,0.00344351,0.001485915,0.0007042818,0.0007226532],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001758035,"about_ca_system_score_gemma":0.001452231,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001077017,"about_ca_topic_score_gemma":0.00102465,"domain_scores_codex":[0.9413131,0.02703991,0.003962227,0.002289332,0.02467289,0.0007224903],"domain_scores_gemma":[0.7787442,0.1477587,0.01378293,0.009796652,0.04746985,0.002447565],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001563048,0.0008676211,0.05475971,0.004238405,0.0007857242,0.0002222702,0.003541494,0.11406,0.02184328,0.08146509,0.01465757,0.7019958],"study_design_scores_gemma":[0.0004204266,0.006544258,0.1556277,0.001661205,0.0006303063,0.0009024958,0.004790247,0.5559456,0.05311935,0.1285488,0.09124953,0.0005600086],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.33888,0.004976418,0.5827703,0.001594485,0.0005995004,0.002007475,0.004205722,0.001848956,0.06311715],"genre_scores_gemma":[0.8598167,0.0009246895,0.1308803,0.0001779672,0.0001593432,0.00127391,0.002556631,0.0001975364,0.004012936],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.0272713,"threshold_uncertainty_score":0.1442261,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1592782373554071,"score_gpt":0.3645055818296493,"score_spread":0.2052273444742422,"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."}}