{"id":"W2188606052","doi":"10.3138/cjpe.0028.011","title":"A Good Start, but We Can Do Better","year":2014,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Globe; Engineering ethics; Computer science; Psychology; Engineering management; Process management; Knowledge management; Risk analysis (engineering); Business; Management science; Engineering","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01709848,0.0001273194,0.0002409072,0.000675054,0.0002194225,0.0006355171,0.0005163736,0.00007876074,0.00436192],"category_scores_gemma":[0.002244455,0.00009510024,0.0001355501,0.0005722105,0.00008211859,0.0004753517,0.00001149639,0.0002302048,0.000189011],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003261148,"about_ca_system_score_gemma":0.003775128,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003968465,"about_ca_topic_score_gemma":0.02268562,"domain_scores_codex":[0.9952235,0.0007588977,0.000900536,0.0002090068,0.002581328,0.0003267122],"domain_scores_gemma":[0.9955965,0.0002557448,0.0006609061,0.0003624806,0.002482795,0.0006415751],"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":[0.000006215178,0.00002006219,0.02573927,0.000002433925,0.00001774335,0.000002653651,0.0009756158,0.0006209504,0.00003454668,0.0006280677,0.007329843,0.9646226],"study_design_scores_gemma":[0.001756551,0.001104937,0.100794,0.00008468449,0.0001282149,0.00007185181,0.001887232,0.06166701,0.0001979445,0.04857545,0.7834445,0.0002875777],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9479666,0.0005312665,0.003080346,0.03060579,0.002030485,0.001178524,0.0000173077,0.00001719443,0.01457254],"genre_scores_gemma":[0.9945521,0.000007869703,0.003751094,0.0008177675,0.0004216482,0.00003706996,0.000007228682,0.00001059452,0.0003946081],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.964335,"threshold_uncertainty_score":0.9965482,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3188817265860601,"score_gpt":0.4975859348037114,"score_spread":0.1787042082176513,"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."}}