{"id":"W4367724627","doi":"10.1007/978-3-031-31234-2","title":"Performance Evaluation Methodologies and Tools","year":2023,"lang":"en","type":"book","venue":"Lecture notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering","topic":"Simulation Techniques and Applications","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"University of Ioannina; Universitat de les Illes Balears; Université de Versailles Saint-Quentin-en-Yvelines; Indian Institute of Technology Delhi; Avignon Université; University of Minnesota; Universidad de Zaragoza; Indian Institute of Technology Madras; Imperial College London; Indian Institute of Technology Roorkee; Amherst College; Indian Institute of Science; Aalto-Yliopisto; York University; Indian Institute of Technology Bombay; McMaster University","keywords":"Computer science","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01511593,0.003150815,0.001662013,0.007848972,0.0007458748,0.007013605,0.002797243,0.001444721,0.01063201],"category_scores_gemma":[0.02672586,0.001020907,0.001367499,0.007321225,0.001334983,0.004880961,0.002392829,0.002164084,0.007127294],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001846387,"about_ca_system_score_gemma":0.0026481,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001262498,"about_ca_topic_score_gemma":0.0009224394,"domain_scores_codex":[0.975093,0.009580886,0.00220436,0.001579995,0.0108141,0.0007276637],"domain_scores_gemma":[0.9823454,0.007924071,0.001265002,0.003466255,0.004644901,0.0003543095],"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":[0.00009914275,0.0002358969,0.0008377238,0.001253044,0.0001282347,0.00006105597,0.0002374748,0.02256401,0.002293558,0.2414359,0.03142427,0.6994298],"study_design_scores_gemma":[0.00008593622,0.0004945606,0.001983663,0.002027434,0.0002253605,0.0005863105,0.0004435181,0.1844971,0.01501367,0.4564252,0.3380586,0.0001587155],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.001921875,0.005898675,0.9409029,0.000750977,0.0002675206,0.0006770001,0.0009203728,0.00314034,0.04552028],"genre_scores_gemma":[0.08760212,0.008657172,0.874351,0.0004771129,0.0006481356,0.001830242,0.002865972,0.001438391,0.02212988],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01511593,"threshold_uncertainty_score":0.07994163,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2575639494538544,"score_gpt":0.4106394678136102,"score_spread":0.1530755183597558,"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."}}